Agentic AI drew the crowds and the CDO role got personal, but data quality was the theme nobody could avoid. How can you trust what AI tells you when you cannot trust the data underneath it? Over and over, leaders told me the AI conversation keeps dragging everyone back to fundamentals they hoped were solved 20 years ago, and they are not.

Tom Redman, the Data Doc and president of Data Quality Solutions, revisited the study that made his name, the one where only 3 percent of companies tested met basic data quality standards. His unpublished follow-up work suggests things have gotten no better since. Tom’s real point is about people, since the regular folks without data in their titles do most of the data work and hold the key to fixing it, and misaligned incentives manufacture bad data on purpose, like the badge scans that measure marketers on volume while salespeople drown in unqualified leads. If the data quality pitch worked, why are we still having this conversation 30 years later?

Fern Halper, founder of the AI Foundations Group and VP of research at TDWI, arrived with a warning for every executive chasing the AI budget. AI will expose every weakness your company has. She separates data governance from AI governance, notes that the median score on TDWI’s data governance maturity model still sits at 58 out of 100, and points out that trust in unstructured data trails structured data by roughly 20 percentage points, right when generative and agentic AI make unstructured data essential. Companies hit a value ceiling, she told me, the moment they run generative AI without their own data underneath it.

Stacie Christensen, a senior data leader at H-E-B with 22 years in the industry, grades our collective data quality somewhere between a D plus and a C minus, and she is the one who framed the whole event’s fear for me. AI is a multiplier and a scaler with nothing positive to scale when the foundation is missing. She makes the case that governance belongs at the moment of data creation, because bad data gets exponentially more expensive the further down the pipeline it travels, and she uses circular data flows as her favorite red flag for hidden complexity.

Leticia Naqvi, a people analytics research manager at Apple, brought more than a year of research on organizational AI readiness, built on four pillars of leadership alignment, data maturity, innovation culture, and change management. She warns that the pillar leaders forget is data maturity, and it sits underneath every AI output they will ever trust. Her advice for CDOs wondering where to start is an honest current-state assessment and one function at a time rather than a big-bang rollout, since she watched four external partners each hold a different meaning for a single data definition.

The foundation theme ran through the largest group of conversations, so here is where the rest of them landed.

• Danette McGilvray, Granite Falls Consulting: She compares most data cleanup to a murder scene where someone hauls the body away without investigating, which is why the same crime scenes keep happening. Her Ten Steps methodology starts with the question too many teams skip, whether anyone actually cares about the problem they are excited to fix.

• Jonathan Agee, Validatar: The question that started his company is the one every leader is asking now, how do I know this data is right? He grades the state of data quality a C and argues for treating quality like software QA, with prevention that starts in development rather than firefighting in production.

• Kelley Kassa, BARC US: Her research found data management is the number one priority for corporate performance management teams, ahead of AI, because finance teams know their foundation is not ready. Only 9 percent of finance AI in North America is in production, and her line on accountability sticks, the agent does not go to jail when the number is wrong.

• Gwen Thomas, The Data Governance Institute: When she registered datagovernance.com in 2003, a search on the term returned 52 hits, and 39 of them were IBM. She explains how governance pulls AI projects out of POC purgatory by handling the edge cases, and why organizations govern for the same reason cars have brakes, so they dare to go fast.

• Dan Everett, Insightful Research: After a path into brain science that began with his son’s autism diagnosis, he makes the case that AI adoption is really change management. Recent studies, he notes, show claimed productivity gains shrink once domain experts have to validate the output on the back end.

• Intuit Credit Karma (Veenit Shah and Puneet Singh): Their team keeps credit scores current for about 140 million members across more than 100 tables and 40,000 columns, monitored on five data quality pillars. An AI remediation agent now compresses root cause investigations that once took 30 to 40 minutes into just a few, with guardrails and security review gating every step up the maturity curve.

Where this leaves you

Every one of these leaders came at data quality from a different angle, and they landed in the same place. AI scales a weak foundation rather than fixing it, and it does so with confidence that makes the errors harder to catch. The teams pulling ahead governed data at the moment of creation and measured quality where the work happens, and they refused to trust an output they could not trace back to something real.

This is one theme of three. Head back to the full CDOIQ 2026 field notes for the complete roster, or read the other two themes on agentic AI and the context problem and the chief data officer role at 20 years. All 24 interviews are on the Data Faces Podcast on-location hub.

Leave a comment

Share

Frequently asked questions

Why is data quality such a big theme for AI?

AI systems act on the data they are given, so poor data quality produces poor and often confidently wrong outputs. Tom Redman’s research found only 3 percent of companies met basic data quality standards, and BARC found 70 percent of companies say less than half of their unstructured data is usable for AI. As Stacie Christensen of H-E-B put it, AI is a multiplier with nothing positive to scale when the foundation is missing, so weak data becomes more expensive and more visible once AI acts on it.

Where should a company start with data quality for AI?

Leaders at CDOIQ 2026 recommended governing data at the moment of creation rather than downstream, because bad data gets exponentially more expensive the further it travels. Danette McGilvray advised starting with whether anyone actually cares about the problem, and Leticia Naqvi of Apple recommended an honest current-state assessment and starting with one function rather than a big-bang rollout. Every leader landed on the same first step, fixing the foundation before scaling AI on top of it.

What is the difference between data governance and AI governance?

Fern Halper of TDWI drew the line clearly. Data governance covers the quality, ownership, and control of the data itself, while AI governance adds oversight of the models, agents, and automated decisions built on that data. Confusing the two leads to mistakes, because governing the AI without governing the underlying data leaves the foundation untrusted. Both matter more as generative and agentic AI make unstructured data essential.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

▶ Watch the full episode and read the transcript: Data Quality for 140 Million Members | Intuit Credit Karma

What happens when a data quality alert fires at two in the morning and says that a few percent of last night’s data didn’t load properly? For many data teams, that’s a ticket to deal with later. Someone picks it up after their coffee or tea, reruns the pipeline, and all is copacetic. But what happens if you change your frame of reference and read the same alert as a count of people? At Intuit Credit Karma, Veenit Shah’s team treats that alert as a few percent of 140 million people who will open an app today and see a credit score that’s out of date. That habit of counting people rather than rows is the whole story of how the company got out of the firefighting years.

Part of the reason data quality programs fizzle is that nobody quite agrees on what quality data means anymore, and AI has made the definitions slipperier. Malcolm Hawker argued on this show that data is only AI-ready when it supports the use case in front of it, and Brendan Grady said that bad data didn’t matter much until AI started acting on it. Both of those conversations were about definitions. Veenit and Puneet Singh’s is about what a team does with an alert once the definitions run out.

I caught Veenit, Senior Manager of Data Engineering, and his colleague Puneet Singh, Senior Data Engineer, at the CDOIQ Symposium in Cambridge, Massachusetts, the day after they presented Credit Karma’s data quality journey to the conference.^[1] The talk wasn’t a victory lap – it’s still a work in progress. Veenit described a program that started in a reactive state a couple of years ago and picked up what he calls “scars along the way,” and the two of them came to share their learnings rather than the trophy.

“Regardless of what level you’re at, whether you’re just starting as a junior engineer or whether you’re a senior manager as well, it is everybody’s job.”

— Veenit Shah, Senior Manager of Data Engineering, Intuit Credit Karma

So, who’s job is data quality? Everybody’s job is a line I’ve heard in quite a few governance decks, and when you hear this, it’s usually a signal that it’s nobody’s job. What makes it true at Credit Karma is a translation the team does on every alert, and I think that translation explains the pillars, the measurement discipline, and the AI agent that now does the investigating.

About Veenit Shah and Puneet Singh

Veenit Shah has spent close to five years at Credit Karma and leads the data engineering team that keeps credit scores current for the company’s members. Credit Karma became part of Intuit in December 2020, when it had more than 110 million members, and it now reports more than 140 million.^[2],[3] Every one of those members expects an accurate score when they open the app, which is the number Veenit’s team measures itself against.

Puneet Singh is a Senior Data Engineer on Veenit’s team and has been at Credit Karma for four and a half years. His twelve years in data engineering started as a Hadoop developer, then moved to the cloud, and today he works mostly on Google Cloud, building microservices that ingest data and check its quality. He demonstrated the team’s agents writing production pipeline code on stage at the symposium.

In this episode, Veenit, Puneet, and I discuss:

- What startup-era data engineering looks like when there are no pillars to rely on

- Why five data quality pillars beat nineteen, and how a small proof of concept earned the whole program

- How every alert across 40,000 columns gets logged, measured, and reviewed each week

- Why a single ingestion alert has to be read as a count of members

- What an AI remediation agent does, what it is allowed to touch, and why it only works because the discipline came first

Watch the full conversation here:

Nightly calls and no pillars to stand on

When Puneet joined Credit Karma, the company still had what he calls the start-up feeling, even after Intuit acquired it. Move as fast as you can and put out as many fires as you can. There were no set data quality checks and no patterns to build on, so the job ran on long hours and nightly calls, and the same problems came back the following week because nothing had been fixed at the root.

“There were no set data quality checks, there were no patterns or pillars that we could rely on, and it was a mess, like long hours and nightly calls.”

— Puneet Singh, Senior Data Engineer, Intuit Credit Karma

It’s easy to stay in reactionary mode when there’s always another problem to fix, so how does a team ever win the mandate to stop and fix the root? Veenit says that a few incidents made it clear that the team’s data quality posture wasn’t where it needed to be, and the leaders above them let the engineers trust their instincts about what to double down on. Both things had to be true at once, because incidents without trust produce a blame cycle, and trust without incidents produces a roadmap that nobody funds.

What they had in front of them, in Veenit’s words, was a blank board, and they had to decide how to start drawing. The program is now at least three years old, and he was careful to say that it’s never complete. Generative AI and agentic applications are changing the end state faster than anyone can describe, so the team focuses on the next step rather than the destination.

Five pillars, not nineteen

A speaker in an earlier session at CDOIQ had offered nineteen different ways to do data quality. Puneet’s team went the other way and defined the pillars that fit its own data, including timeliness, completeness, accuracy, and governance. That sounds like a smaller ambition, doesn’t it? In practice, it’s the more demanding one, because five pillars that are measured on every table beat nineteen dimensions on a slide that nobody instruments.

Puneet was equally practical about how the program earned its funding. He pitched a data quality vendor for one narrow use case, showed a measurable benefit, and only then came back for the whole thing. Leadership trusting the team’s instincts didn’t mean signing a blank check. It was the the small proof of concept that translated it to a budget item.

“Each and every alert is logged, measured, and we do analysis on a weekly basis to make sure where the fires are, where we need to get better, and where we need brutal prioritization to remove them or fine-tune those.”

— Puneet Singh, Senior Data Engineer, Intuit Credit Karma

The team runs data quality checks on more than 100 tables and roughly 40,000 columns, and logs and measures every alert those checks produce. A weekly review then decides where the fires are and which alerts deserve brutal prioritization, either fixing the underlying issue or tuning the check so it stops crying wolf. Most data quality programs I’ve seen skip that review, and they end up with a dashboard full of red that everyone has learned to ignore. How many of those dashboards are in your own organization right now?

Translate the alert into people

I asked the question I ask every data leader: whose job is data quality? The business? The data engineering team? An agent? Veenit’s answer was the everybody’s-job line, and normally I would have moved on, because everybody’s job is what people say when they haven’t decided. Then he explained why it holds on his team, and the reason turned out to be a translation habit rather than a value statement.

“A simple alert could look like, ‘ Hey, we have not been able to ingest X percent of data. But what that actually means is X percent of members out of the 140 million are going to be impacted.”

— Veenit Shah, Senior Manager of Data Engineering, Intuit Credit Karma

Once an alert is read as a count of members rather than a percentage of rows, ownership stops being a governance question. A junior engineer who sees millions of people behind a failed ingestion doesn’t need a RACI chart to know it matters, and a senior manager can’t delegate it into a backlog. Veenit calls this a fundamental aspect of the team’s culture, and it’s also the pitch that I wish more data leaders would make to their own organizations. Asking people to care about the members, customers, or patients on the other side of the pipeline is a much easier sell than asking them to care about data quality.

Veenit also drew a useful line between the two sources of trouble. External issues arrive from partners and financial institutions whose data the team doesn’t control, and those are the cards you are dealt, so the only question is how to play the hand. Internal issues show up when the team onboards a new partner or builds a new feature, and better engineering practice can prevent those. Separating the two keeps the weekly review from treating an upstream mess and a homegrown bug as the same kind of fire.

The agent came last

Every conversation about data operations right now turns to agents within ten minutes, and this one was no different. Credit Karma built an AI remediation agent that picks up a data quality error, runs the diagnostic query, refines it, and keeps running queries and analysis until it can state the root cause and the blast radius. Puneet said that investigation used to take an engineer 30 to 40 minutes and now takes a few minutes. On the build side, the team has given agents the context to write production-ready pipeline code, including the data quality and observability pieces, in five to ten minutes.

Those results get headlines. So why does the agent work here when so many bolted-on agents don’t? The agent runs against alerts that are already logged and measured, on checks that three years of weekly reviews have tuned, with runbooks the team wrote for humans and then handed to the agent as explicit instructions. It operates in a protected environment, never touches personally identifiable data, and works only with anonymized data that was already approved for analytics. You can’t automate a mess, and Credit Karma’s agent finds the root cause in minutes because the discipline it runs on came first.

“The agent goes there, looks into the data quality error, runs the query, as well as refines those queries, and keeps running the queries and analysis till it comes to the conclusion with the blast radius, saving our time from 30 to 40 minutes to just a few minutes.”

— Puneet Singh, Senior Data Engineer, Intuit Credit Karma

I raised the obvious worry: agents sometimes go off the rails. The team places itself at level two on its own agentic maturity model, and everything planned for levels three and four is still under security review. Selective access, which tools an agent can use, and what its role should be will get decided slowly, and the team is comfortable saying it doesn’t yet know what levels four or five look like. Veenit put it plainly: the game itself is changing, and they have to keep playing it.

The benefit compounds

Veenit described the payoff as multifold, and the first part is the one everyone expects. Investigations that took 30 minutes to an hour are compressed into minutes, giving the team back time he described as insane. The second part is what I think matters more. With firefighting handled, the team can think about the next frontier, find gaps it didn’t know existed, and stop every data quality issue the program catches from compounding on the last one.

If your own data quality program is stuck in the firefighting years, the place to start isn’t a framework or an agent. Start by naming who is on the other side of the next alert.

Listen to the full conversation with Veenit Shah and Puneet Singh on their Data Faces Podcast episode page.

Based on insights from Veenit Shah, Senior Manager of Data Engineering, and Puneet Singh, Senior Data Engineer, at Intuit Credit Karma, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [0:03] David introduces Veenit Shah and Puneet Singh on location at the CDOIQ Symposium in Cambridge, Massachusetts

- [0:27] Veenit on keeping credit scores current for about 140 million members

- [1:49] Veenit on the session, a data quality adoption journey from reactive to proactive, and the scars along the way

- [2:23] Puneet on the starter feeling, extinguishing fires, and nightly calls with no pillars to rely on

- [3:25] Veenit on the twofold mandate, incidents plus leaders who trusted the team’s instincts

- [5:08] Puneet on choosing five pillars over nineteen, and the small proof of concept that earned the program

- [5:46] Puneet on 100-plus tables, 40,000 columns, and the weekly review with brutal prioritization

- [7:07] Veenit on why data quality is everybody’s job when an alert means members

- [8:13] Puneet on the AI remediation agent and 30 to 40 minutes becoming a few

- [10:09] Veenit on level two of the agentic maturity model and the runbooks handed to the agent

- [12:21] Puneet on guardrails, anonymized data, and what stays under security review

- [13:22] Veenit on the compounding benefit and the gaps they did not know existed

About David Sweenor

David Sweenor is the founder of TinyTechGuides and host of the Data Faces Podcast. He is an international speaker, advisor, and the author of eleven books on artificial intelligence, analytics, and B2B marketing, including Generative AI Business Applications, The CIO’s Guide to Adopting Generative AI, and Modern B2B Marketing. With more than twenty-five years in analytics and AI at companies including Alteryx, Tableau, TIBCO, SAS, IBM, and Dell, David advises technology companies on product marketing, content strategy, and go-to-market execution. He holds several patents and has been named a top influencer in data and analytics by Onalytica, Thinkers360, and Analytics Insight.

Connect with David on LinkedIn and subscribe to the Data Faces Podcast for conversations with the people shaping enterprise data and AI.

Footnotes

^[1]CDOIQ Symposium. “The 20th Annual CDOIQ Symposium.” July 21–23, 2026, Hyatt Regency Cambridge, Massachusetts.

https://2026cdoiq.org/

^[2]Intuit Inc. “Intuit Completes Acquisition of Credit Karma.” Intuit Press Room, December 3, 2020. https://www.intuit.com/company/press-room/press-releases/2020/intuit-completes-acquisition-of-credit-karma/

^[3]Intuit Credit Karma. “About Intuit Credit Karma.” https://www.creditkarma.com/about

CDOIQ is where the chief data officer role was born, so the 20th anniversary put the job itself on the table. If the job description changes every single year, how do you know when you have the right person in the seat? The people who invented the role and the people living it now agreed on one thing, that it is being rewritten in real time, with the pressure coming straight from AI.

Randy Bean, founder and CEO of the Data & AI Leadership Exchange, has organized the CDOIQ chief data officer panel for 12 years, and his test for data leaders is blunt. If you are not getting measurable business value from your data and AI investments, or a clear path to it, go back to the office this afternoon and shut them down. He has watched the role move from defensive risk and compliance work to an offensive, business-focused mandate, and he told me every chief data officer from last year’s panel will be out of that seat by the end of this year. His read on the hype is the line worth taping to your monitor, AI is overestimated in the short term and underestimated in the long term.

Richard Wang, co-founder of the MIT CDOIQ Symposium, gave one of the conversations I will remember longest, fresh off a standing ovation and a Lifetime Achievement Award. He traced the arc from a two-year assignment at the Pentagon that convinced him every major organization needs a chief data officer with budget and authority, to a world with more than 2,000 CDOs today. He also coined a term every data leader will recognize the second they hear it, data archaeology, the cycle where organizations lose the people who hold the knowledge and end up digging up and repeating the same work.

Peter Aiken, a professor at Virginia Commonwealth University, wrote the first book on the chief data officer role 20 years ago, and he still argues data needs one throat to choke. He is blunt about the odds, the average CDO lasts about a year and a half, and the best ones have been fired three times, because doing the job right means telling people things they do not want to hear. His button carries the rest of the message, grow tomatoes, not data centers, and his Anything Awesome premise is the warning underneath all the AI hype, since bad data plus anything awesome still produces bad results.

Mark Ramsey, managing partner at Ramsey International, took up professional motorsports in his 50s and turned it into Data at Speed, a book that maps the five crashes every chief data officer will hit. The first one he named stuck with the room, the shadow pit crew, what forms when the CDO does not drive access to data and the business quietly builds its own workaround. He also put a number on the stakes, half of CDOs do not survive three years, and he argued that blocking generative AI only pushes employees toward free tools that expose company data.

The rest of the CDO conversations circled the same question from different chairs.

• Stuart Madnick, MIT: A Lifetime Achievement honoree who helped launch the data quality work that became CDOIQ, he traced 40 years from the Total Quality Management movement to the insight that data is a manufactured product. His warning for the AI era is direct, by every measure he has seen, attackers are using AI more aggressively and more efficiently than defenders.

• John Ladley, author and advisor: When he asked ChatGPT for a data model and got the rudiments back in 38 seconds, he called it career altering and postponed retirement to fold AI into a third edition of his governance book. His take on the profession is humbling, a field that is only 40 years old still has not figured out its 1.0.

• Leigh Felton, AI for Job Security Foundation: She spent 20 years at Microsoft and built its first ethical AI enablement program, and she argues you cannot de-bias a system built to recognize historical patterns. She makes the case for governing AI like an environment that shapes how people think, closer to OSHA than to a tool you pick up and put down.

Where this leaves you

Twenty years ago, the chief data officer job meant defense, risk, and cleaning up messes. These leaders described a job that is now about proving business value fast enough to keep the seat, with AI raising the bar every quarter. A CDO keeps that chair by holding a real mandate, the authority to drive data access and kill what does not deliver, and that matters more than whether the role reports to technology or the business.

This is one theme of three. Head back to the full CDOIQ 2026 field notes for the complete roster, or read the other two themes on agentic AI and the context problem and data quality as the AI foundation. All 24 interviews are on the Data Faces Podcast on-location hub.

Share

Leave a comment

Frequently asked questions

How has the chief data officer role changed in 20 years?

Leaders at CDOIQ 2026 described the CDO role moving from defensive risk and compliance work toward an offensive, business-value mandate. Randy Bean of the Data & AI Leadership Exchange traced that arc directly and noted survey data showing 42 percent of CDOs report under technology and 33 percent under business leadership. AI is accelerating the change by forcing data leaders to prove measurable value or shut initiatives down.

Why do chief data officers have such short tenures?

Multiple leaders cited high turnover, with Mark Ramsey noting that roughly half of CDOs do not survive three years and Randy Bean predicting every member of last year’s CDO panel would leave that role within a year. Most of them traced it to a mandate problem. When the CDO does not own data access and cannot show business value, the role becomes a target, and AI has raised the stakes on proving that value quickly.

Where should the chief data officer report?

Opinions varied, but the leaders agreed the mandate outweighs the reporting line. Peter Aiken argued data belongs in the business rather than in IT, because every business problem has a data component, and that power and influence matter more than the org chart. Survey figures shared at the event put 42 percent of CDOs under technology and 33 percent under business leadership, with the rest split elsewhere.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

▶ Watch the full episode and read the transcript: Malcolm Hawker on AI-ready data and identity resolution

When I built data warehouses at IBM, I used to argue with our chief architect about master data management (MDM). My view was that if the manufacturing data was wrong, we should fix it in the source systems where it was created. His view was that we would build the mapping tables, reconcile the identifiers downstream, and leave the source systems alone. He won those arguments, and for years afterward I understood MDM mechanically without ever being convinced it was solving the right problem.

Then I spent the second half of my career marketing data, analytics, and AI platforms at SAS, TIBCO, Alteryx, and Alation, and I still write that copy for clients. Today it says “AI-ready.” Much as it pains me, the phrase is on nearly every vendor site, it draws a great deal of budget, and I have yet to see two companies define it the same way. So when Malcolm Hawker came on the Data Faces Podcast, I asked him what it means, and whether the companies claiming their data is AI-ready are fooling themselves.

Malcolm’s not really a fan of the term “AI-ready.” Of course, data quality still matters, but readiness was never a property your data has or lacks, and treating it as a yes-or-no measurement is why the phrase is ambiguous and nearly meaningless. The industry, he adds, has strong incentives to keep producing phrases exactly like it.

About Malcolm Hawker

Malcolm Hawker is the Chief Data Officer at Profisee, a master data management vendor, where most of his job is external thought leadership. He spent three years as a Gartner analyst covering MDM and governance, and before that held IT leadership and product roles including Distinguished Architect at Dun & Bradstreet. He hosts the CDO Matters podcast and wrote The Data Hero Playbook.^[1]

In this episode, Malcolm and I discuss:

- Why “AI-ready data” is an overloaded term that collapses a spectrum into a binary

- Why the cost of being wrong decides whether an AI use case can go into production

- The difference between defining what a customer means and knowing which customer record is real

- Why telling your CEO “garbage in, garbage out” is a career-limiting move

- The semantic pedantic feedback loop, and how data literacy went from nonexistent to a top-three problem in a single year

Watch the full conversation here:

Ready for what, exactly

Malcolm’s definition of AI-ready data is a practical one. Data is AI-ready when it supports the use case in front of it. It maps onto the oldest working definition of data quality, which is fitness for purpose, and it means the same customer table can be perfectly ready for one job and nowhere near ready for the one running beside it.

Large language models (LLMs) were trained on the open internet, which he describes as a complete cesspool when it comes to data quality, and they work well enough that the entire market reorganized itself around them. At the other end sit healthcare, legal, finance, and anything carrying compliance or audit obligations, where accuracy and consistency are not negotiable. Both of those are AI use cases, and no single quality standard covers both.

Malcolm reframes readiness as a question of consequences rather than a question about data. Send a digital coupon to the wrong person, and it costs you close to nothing, which means your data is probably ready today. Model a specific customer’s behavior, or decide what offer that person should see next, and you need a far higher standard. In Malcolm’s read, that mismatch is also why so many proofs of concept die: organizations take a probabilistic system and drop it into a business process that has always run on deterministic rules, where the output was predictable, auditable, and safely inside a governance policy.

AI models do not behave that way, and the reason they misbehave is rarely recoverable after the fact. Malcolm calls this the attribution problem, and no amount of retrieval patterns, grounding, or knowledge graph scaffolding fully removes it. You can reduce how often the system surprises you, but you can’t get to a place where you know why it did what it did every time.

“AI-ready data is data that supports a given use case. I mean, literally, that’s it.”

— Malcolm Hawker, Chief Data Officer, Profisee

What the semantic layer cannot fix

Semantic layers exist for a good reason, and Malcolm is careful not to argue against them. An LLM cannot look at your raw tables and say anything useful about your customers, because it does not know what a customer is in your business or how your customer table joins to your product table. What it wants is text. Josh Howard put the same point more bluntly on this show: your AI is dumb without your data. Layering definitions over the tables, which is roughly what the market means by a semantic layer, is how the tabular data becomes legible to it.

Knowledge engineers have carried a name for this split for decades. The T-box, short for terminology box, holds the definitions: what counts as a customer, how net revenue is calculated, and what a location is.^[2] Nearly every conversation in the market right now, whether it is labeled ontology, knowledge graph, or semantic layer, is a conversation about the T-box. Malcolm argues that the other half of the structure has gone mum. The A-box, the assertional box, holds the actual values, and it answers a different kind of question: is Malcolm Hawker a customer, and if five Malcolm Hawker records exist in the system, which one is him?

Identity resolution, master data management, and old-fashioned data quality all live in the A-box, and no quantity of definitions reaches them. A perfectly specified ontology will tell an AI agent exactly what a customer is while remaining completely silent about which of your customer records is the real one. On this show, Matt Hayes made a related argument that AI-ready data clears a much higher bar than analytics-ready data, and Sam Pierson described the industry work to standardize semantic definitions so meaning travels with the data. Malcolm is adding the layer underneath both, where meaning gets you part of the way, and identity gets you the rest.

“If you’ve got 15 different versions of David Sweenor, all the context in the world isn’t going to solve that problem.”

— Malcolm Hawker, Chief Data Officer, Profisee

Identity problems are also harder to catch than the failures data teams are trained to look for. People can see a broken dashboard because they notice a number is off. Duplicate customer records produce answers that look entirely reasonable while resting on the wrong person, so the system returns them with total confidence, and nobody has a reason to check.

Never say “garbage in, garbage out” to your CEO

Garbage in, garbage out is probably the most repeated line in data management and analytics. The phrase flattens a spectrum into a binary, the same way “AI-ready” does, and it writes off data that might be entirely sufficient for the job at hand. That alone would make it sloppy, and Brendan Grady made the companion argument on this show about why bad data didn’t matter until now, which is that analytics absorbed errors AI will not. What makes the phrase dangerous is what happens when a data leader says it out loud in front of the business.

“First of all, the whole concept of garbage in, garbage out gives me hives.”

— Malcolm Hawker, Chief Data Officer, Profisee

Malcolm puts it in a boardroom. You are the CDO, and your CEO is holding a report with two David Sweenors on it and wants to know why there are two of a customer you both know is one person. No data leader answers that question with “sorry, boss, garbage in, garbage out” and keeps their standing in the room. I mentioned to Malcolm that this is career-limiting, because the phrase disowns the entire investment. In that one sentence, the warehouse, the pipelines, and every governance and quality platform on the books are declared powerless against whatever arrived upstream. It hands away the argument that data work changes outcomes, which is the entire argument a CDO is employed to make.

The semantic pedantic feedback loop

Malcolm first posted the most useful idea in this conversation on LinkedIn as a joke, then could not stop finding evidence for it. He calls it the semantic pedantic feedback loop, and it explains why our field keeps inventing new names for things that already had names. I knew to ask about it because Scott Taylor, who came on this show to argue for putting truth before meaning, told me I had to.

It runs in a circle. Analysts, consultants, and thought leaders need to be seen saying something new, and in the analyst business there is a subscription renewal attached to that need, because nobody renews to hear that best practice has not changed much. So new vocabulary gets manufactured. Vendors pick it up and build campaigns around it. Buyers start hearing it from every vendor and at every conference, decide it must be important, and call their analyst to ask what it is. The analyst hears their own coinage echoed back by the whole market and confirms that it is real. Around it goes.

Malcolm’s evidence is a receipt from inside Gartner itself. By his account, Gartner’s annual CDO survey asks respondents to name their top roadblocks, and in 2017 no respondent named data literacy, because it was not among the options. Gartner added it the following year, and it immediately ranked third. He is pointed about why that particular option travels so well: it locates the obstacle in everyone else’s knowledge. The impediment sits with the people being served rather than with the tools, the dashboards, or the data team itself. As he puts it, the thing had a name before it got a new one, and the name was training.

Let me say where I am standing while I write this. I host a podcast, I have written books, and a decade of my career went into product marketing for data platforms. That puts me inside the loop being described, along with the phrase that opened this article. “AI-ready” is the loop’s most recent output. Recognizing that buys you one useful habit: every time a new term arrives, ask what it was called last time and whether anything underneath it changed.

What the good ones do differently

Malcolm has an unusually large sample to draw on when he describes what separates the data leaders who succeed. Gartner analysts take client calls they call inquiries, and he did roughly 1,500 of them in three years with CDOs, CIOs, and VPs of data and analytics. In struggling organizations, he heard the same attribution pattern every time. They blamed data literacy, or culture, or the fact that nobody showed up to the governance committee meeting. The blocker was always somewhere else.

Leaders getting lasting results in governance, data quality, and MDM sounded different. They expected to be wrong the first time and to learn from it. They measured themselves on whether the people consuming their data products were successful, and they asked for feedback often enough to find out when they were not. That argument is what The Data Hero Playbook is built on, and the book’s premise is that the limiting factor in most data organizations is a set of beliefs rather than a missing capability.^[3]

So the next time someone asks whether your data is AI-ready, the useful response is that the question is incomplete. Ask what the use case is, because readiness is meaningless without one. Ask what being wrong costs, because that number decides whether the use case can go into production. Then ask whether you can tell which record is the real one, because if you cannot, no amount of context is going to rescue the answer.

Listen to the full conversation with Malcolm Hawker on his Data Faces Podcast episode page.

Based on insights from Malcolm Hawker, Chief Data Officer at Profisee, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [0:06] Introduction and welcome to the Data Faces Podcast

- [1:18] From IT operations to Distinguished Architect at Dun & Bradstreet

- [2:20] Three years at Gartner, and what an externally facing CDO actually does

- [4:22] The record label he never started

- [5:09] A graduate thesis on Napster, and the case for buying a turntable again

- [7:02] What does “AI-ready data” even mean?

- [7:46] Fitness for purpose, and why the term is overloaded

- [8:05] ChatGPT was trained on the internet and works anyway

- [11:56] The cost of being wrong, and why a coupon is not a claims decision

- [13:10] Semantic layers, the T-box, and the fifteen David Sweenors

- [14:51] Does “garbage in, garbage out” mean anything for unstructured data?

- [15:44] Why the phrase gives Malcolm hives

- [15:56] The boardroom scene, and why the answer is career-limiting

- [17:02] Chunking text destroys the context that made it useful

- [19:26] An underserved market, and back to the semantic pedantic cycle

- [19:57] The feedback loop: pundits, vendors, buyers, analysts

- [23:36] How data literacy became a top-three CDO roadblock in one year

- [24:09] “There was a word for it, and it was called training”

- [25:17] Does master data management apply to unstructured data?

- [25:59] Master data as the connective tissue across business processes

- [28:03] Profiling, entity linkage, and the fox watching the henhouse

- [29:32] Contracts and forms are tractable; email is the hard problem

- [29:54] The Data Hero Playbook, and whether CDOs play defense

- [31:21] Growth mindset versus fixed mindset, drawn from 1,500 Gartner inquiries

- [35:18] What Malcolm is reading

- [37:29] Where to find Malcolm

Frequently asked questions

What does “AI-ready data” mean?

Data is AI-ready when it supports the specific use case you intend to run on it. There is no universal threshold, because fitness depends entirely on the job. Malcolm Hawker, Chief Data Officer at Profisee, argues the term has become overloaded precisely because the market treats it as a binary state your data either has or lacks. The same customer data can be sufficient for a marketing recommendation and unacceptable for a clinical or compliance decision, so always ask “ready for what.”

How is AI-ready data different from analytics-ready data?

Analytics-ready data supports human analysts reading dashboards and reports, where a person applies judgment and notices when a number looks wrong. AI-ready data feeds systems that act on the data without that check, so errors propagate at machine speed with nobody in the loop to catch them. The accuracy, consistency, and identity requirements are correspondingly higher, and the bar rises further as the cost of being wrong rises.

Does a semantic layer or knowledge graph solve AI data quality?

Only partly. Semantic layers, ontologies, and knowledge graphs define what your terms mean, which knowledge engineers call the T-box. They do not establish which specific records are true, which is the A-box. If your systems hold fifteen versions of the same customer, a perfect set of definitions will still not tell an AI agent which one is the real person. Identity resolution and master data management address that half of the problem.

Why do so many AI proofs of concept fail?

A recurring reason is that organizations deploy probabilistic systems into business processes that historically ran on deterministic rules. Those processes were built on outputs that were predictable, auditable, and safely inside a governance policy. Language models do not offer that guarantee, and what practitioners call the attribution problem means you often cannot determine why the system produced a given answer, which makes the failure difficult to diagnose or defend.

How do you apply data quality to unstructured data?

Nobody has solved this well yet. Traditional quality rules need small, structured units, so the standard move is to break text into chunks, but chunking strips away the context that made the text useful to a language model. Practical progress starts with profiling, identifying which documents reference your master data objects such as customers, suppliers, or products, and then determining whether the claims in those documents are true. That last step remains the hardest.

What should a data leader do instead of asking “is our data AI-ready?”

Replace it with three questions. What is the use case, since readiness has no meaning without one. What does being wrong cost, since that determines whether the use case can go into production at all. And can you identify which record is the real one, since context and definitions cannot compensate for unresolved identity. Answering those tells you far more than any readiness score.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 10 AI thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Hawker, Malcolm. The Data Hero Playbook: Developing Your Data Leadership Superpowers. Hoboken, NJ: Wiley, 2026. https://www.wiley.com/en-us/The+Data+Hero+Playbook%3A+Developing+Your+Data+Leadership+Superpowers-p-9781394310647.

^[2]Baader, Franz, and Werner Nutt. “Basic Description Logics.” In The Description Logic Handbook: Theory, Implementation and Applications. Cambridge University Press, 2003. https://www.inf.unibz.it/~franconi/dl/course/dlhb/dlhb-02.pdf.

^[3]Hawker, Malcolm. The Data Hero Playbook: Developing Your Data Leadership Superpowers. Hoboken, NJ: Wiley, 2026. https://www.wiley.com/en-us/The+Data+Hero+Playbook%3A+Developing+Your+Data+Leadership+Superpowers-p-9781394310647.

Agentic AI was the pervasive topic on the floor this year. Everyone wants agents, and almost nobody has the data plumbing to feed their voracious appetites. So what exactly are all these agents supposed to run on? That question, the context problem, showed up in nearly every conversation about where AI is headed.

Stewart Bond, VP of Data Intelligence Software at IDC, came off his session with a question that pokes at a decade of investment. Is the data lake already obsolete? He argues that agentic AI favors federated architectures over centralized ones, because agents need data in real time where it lives, and latency is the enemy of agentic AI. He walked me through IDC’s model lake convergence idea, the forecast that the model itself becomes the analytical layer by 2029, which pushes data quality, security, and governance closer to the source and turns policy into code that runs on every access rather than a committee that approves it later.

Kevin Petrie, VP of Research at BARC US, returned to the show with a number that should stop any agent project cold. In BARC’s survey, 70 percent of companies said less than half of their unstructured data is currently discoverable and usable for AI. He compares enterprise unstructured data to his kids’ Lego bin, real value buried in a free-for-all, and he makes the case that five to fifteen well-governed agents beat 70,000 ungoverned ones. He also gave the fear a name that landed with the whole room, vibe slop, the mess you get when teams deploy agents on shaky foundations, and he expects agent cleanup and consolidation to become next year’s project.

Amy Lenander and Christina Egea of Capital One showed what solving the context problem actually looks like at scale. Amy is Capital One’s Chief Data Officer and Christina is SVP of Enterprise Data, and their answer is to treat data like a product. A usage analysis revealed that nine categories of data covered most of the company’s needs, and every data product gets a single accountable owner. Those products now launch use cases three times faster and cut the cost of maintaining standardized data by 30 percent. Christina was refreshingly honest about the hardest part, which is getting started, and the constant tension between building fast for one use case and building something that scales to a hundred.

Amin Venjara, chief data and product officer at ADP, gave the framework that made the context problem concrete. Value equals data plus capabilities, and his team runs ADP’s internal data platform like a product that builders across the company choose to use rather than a service they are forced to accept. He described the annual Data and AI Day that drew 2,100 people and the hackathon that feeds it, and a data product that hands chat and agent experiences full customer context, so no team rebuilds the same stitching work twice.

Three more conversations rounded out the picture on agents and context.

• Douglas Laney, Infonomics: The founder of infonomics predicts we will see the first billion dollar business run by a handful of people and a swarm of AI agents. He also warned that traditional data quality dimensions do not translate to unstructured data and AI slop, and that token costs are already making some organizations rethink swapping cheap labor for expensive inference.

• Terry Dorsey, Denodo: She draws a deliberate line between delivering data and delivering information, and argues AI succeeds when you treat it as one capability among many rather than one big encapsulated project. Data quality, in her words, sits in the eye of the consumer.

• Steven Moskowitz, Industry Forward: He works with manufacturers who own the AI, own the technology, and still get no value, because manufacturing runs on wide data, a richer mix of timestamps, images, CAD drawings, and equipment signals than most industries manage. His cognitive KPIs, like mean time to root cause, measure how AI changes thinking rather than how often people click it.

Where this leaves you

The teams making agents work had all done the same unglamorous work first. They moved data quality, governance, and context to where the agents actually reach for data before chasing the newest model. Everything an agent does downstream depends on that groundwork, and no amount of orchestration makes up for a foundation the agent cannot trust.

This is one theme of three. Head back to the full CDOIQ 2026 field notes for the complete roster, or read the other two themes on the chief data officer role at 20 years and data quality as the AI foundation. All 24 interviews are on the Data Faces Podcast on-location hub.

Share

Leave a comment

Frequently asked questions

What is the context problem in agentic AI?

The context problem is the mismatch between what AI agents need and what most data environments provide. Agents act in real time and need governed, discoverable data where it lives, but leaders at CDOIQ 2026 described most enterprise data as centralized, slow, and largely unstructured. BARC found that 70 percent of companies say less than half of their unstructured data is usable for AI, which means agents often have nothing trustworthy to act on.

Why do agents favor federated data architectures over data lakes?

According to IDC’s Stewart Bond, agents cannot wait for batch processes to move data into a central lake, because latency is the enemy of agentic AI. Federated architectures let agents reach data in real time where it already lives, which pushes data quality, security, and governance closer to the source. Bond expects the model itself to become the analytical layer, what IDC calls model lake convergence, by around 2029.

How should enterprises prepare their data for AI agents?

Leaders at CDOIQ 2026 recommended treating data like a product with clear ownership, moving governance to the point of data creation, and favoring a small number of well-governed agents over thousands of ungoverned ones. Capital One and ADP both described running internal data platforms as products, which gave chat and agent experiences the full context they need. Agents need trustworthy, real-time data they can act on without amplifying errors.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

What decides whether anyone uses the data product your team just built? When a data team finishes one, the quality is usually better than the previous generation, but adoption stays flat anyway. What do you do? Typically, the team responds by evangelizing harder and escalating to whoever runs the reluctant business unit. Amy Lenander, Chief Data Officer at Capital One, thinks all of that arrives too late to help, because the decision that governs adoption got made long before the build ever started.

I caught Amy and her colleague Christina Egea, SVP of Enterprise Data, at the CDOIQ Symposium in Cambridge, Massachusetts, and pulled them in front of a Data Faces microphone.^[1] Their session, “Doubling down on data products to drive business value,” followed the talk Amy gave the previous year on Capital One’s broader data transformation. Data products drew more interest than anything else that year, so they double-clicked on the piece of their strategy that peers find least familiar: Capital One curates a deliberately small set of data products instead of letting every domain publish its own.

“One of the things that we do that I think is most unique among the companies I’ve talked to is being very intentional about curating the data we want to manage as data products.”

— Amy Lenander, Chief Data Officer, Capital One

That restraint is a design decision rather than a resourcing constraint, and it traces back to a question Christina’s team asks before building anything: will this still hold up at the hundredth use case?

About Amy Lenander and Christina Egea

Amy Lenander became Chief Data Officer at Capital One in February 2023, capping two decades at a company she joined as an analyst who made decisions from however much data would fit in a spreadsheet.^[2] She has led loyalty and rewards, served as CEO of Capital One’s UK business, and worked across credit card and lending. Because she came to the data seat from the business rather than from engineering, Amy has what most Chief Data Officers (CDOs) acquire secondhand: a working sense of where the real impact in each business sits.

Christina Egea is SVP of Enterprise Data on Amy’s team, and she has spent three years building Capital One’s data product strategy. Their data ecosystem keeps growing by acquisition — Capital One completed its purchase of Discover in May 2025 and its purchase of Brex in April 2026 — so any strategy worth adopting has to survive a footprint that expands faster than a roadmap.^[3][4]

In this episode, Amy, Christina, and I discuss:

- Why the fastest data product to build is usually the one that helps least

- How a study of company-wide usage collapsed a sprawling data landscape into nine categories

- The case for one accountable owner on every data product

- Why a data product covering 70% of what someone needs will never displace the feed covering all of it

- What three times faster time to market rests on

Watch the full conversation here:

The hundredth use case is the design decision

Christina Egea describes the tension at the center of Capital One’s data product strategy as pressure from two directions at once. One side of the company asks why the team cannot reach the market more quickly, while the other asks how anything built that fast could be correct, reusable, or safe for customers. Both questions are reasonable, and a data organization that answers only one of them ends up with either a backlog or a mess. Christina resolves it by being explicit about which use case her team is building for.

“The fastest thing to build is the thing for a single use case. The harder thing to build is the thing that will scale.”

— Christina Egea, SVP of Enterprise Data, Capital One

Capital One aims well past the first requester. Christina told me her team focuses on making sure a data product will scale well past a single use case, past the next five or twenty, out to something closer to the hundredth use case and the hundredth customer. A specific use case still gets most data products launched, because someone has to want the thing badly enough now to fund it, but the design target sits far beyond whoever asked first. A data product scoped to one requester can hard-code their assumptions and skip the modeling that would let another team use it next quarter.

Christina also names the hardest part of the whole endeavor, and it is not the part most people guess. Getting started is what hurts. The payoff arrives on the far side of that first set of data products, once teams have adopted the same product five or twenty times and the flywheel begins to turn on its own.

Nine categories covered most of the usage

Capital One did not begin its data product work by drawing a domain map or carving the catalog along org-chart lines. Christina’s team studied how data was already being used across the company, asking which core categories and domains that usage fell into. Despite enormous breadth and depth across the business, roughly nine data categories accounted for most of the consumption, and those nine became the first data products the company built.

Measured usage is what makes curation possible, and curation is where Amy locates Capital One’s real point of difference. Two kinds of data qualify in her model. The first is the most-used data in the company, which earns its place because so many lines of business reuse it. The second is data the company believes carries real value but nobody has drawn on much, because getting to it has been too hard. Both are chosen for leverage rather than volume.

“We’ve very intentionally started with the data products that have the most leverage to the company.”

— Amy Lenander, Chief Data Officer, Capital One

Amy’s read on where that impact concentrates comes from having run the businesses she now serves, and the endless list of things a data team could improve becomes tractable once you know which ones the business genuinely needs. Her background also helps with a less comfortable part of the job, because her team has to get business units to do things they would rather not, which she compares to eating your vegetables.

The top-down start was only the opening move. Each business unit has since built a bottom-up view of the data products it needs, while Christina’s central team maintains the framework that holds the picture together.

One owner, and no overlap

Every data product at Capital One has one accountable owner who makes the final decisions on it. Amy is direct about why that matters at her company’s scale: one owner per product, responsible for the decisions and for making sure the products work together well. Plenty of the building is distributed across business units. Without a single named owner, two teams publish their own version of the customer table, and nobody has the authority to say which one is right.

Christina’s central team spends a meaningful share of its time on the shape of the portfolio, working toward a universe of data products that is mutually exclusive, where the pieces do not overlap, and each product’s scope is defined.

“Really making sure that we’re investing in a universe of data products that’s mutually exclusive, that the things don’t overlap, that they clearly link together but have boundaries is a place where my team spends a bunch of our time.”

— Christina Egea, SVP of Enterprise Data, Capital One

From there, Christina’s team scopes each data product and names its owner, then models the data into an ontology before building the assets that bring it to life, including the application programming interfaces (APIs) and tables consumers touch. Settling semantics in that ontology first is what keeps the hundredth use case reachable, because the definitions get negotiated once rather than relitigated by every team that shows up later.

Adoption is a listening problem

Asked what to do when a data product is demonstrably better and nobody uses it, Amy does not reach for another enablement session. She puts the question to the people who have not switched, and then does the harder part, which is hearing what they say.

“The first thing is to listen to the answer when you actually ask that question, to stop pushing and start listening.”

— Amy Lenander, Chief Data Officer, Capital One

The answers are rarely mysterious once someone asks. Switching from a working data feed to a new data product is painful, and the person being asked to move already has something that produces numbers today. Winning a brand-new use case is far easier than displacing an incumbent feed, because displacement forces the consumer to absorb migration work on top of their actual job. They may agree the data product is better and still wonder whether it is worth the trouble.

The coverage trap is the same principle seen from the customer’s side. A consumer might tell Amy’s team that the data product covers 70% of the data they need, and 70% will not displace a source that covers all of it, no matter how much better the newer thing is. Her response is to listen and go build the missing 30%, because a data product meant for broad reuse has to cover the ground its consumers stand on.

Amy frames the discipline in product management terms, which carries a warning for data teams that have never had to win a customer. Nobody churns in a way that shows up on a dashboard, so consumers keep using the old data feed and say nothing, and Amy reads that silence as information about the product.

“They don’t pay for your data, but they vote with their feet. So you need to really deeply understand them like a good product manager would.”

— Amy Lenander, Chief Data Officer, Capital One

The numbers stand on something

Christina is careful about how she presents Capital One’s results, and the caveat she leads with is more instructive than the figures that follow. The company’s data product investments stand on the shoulders of its foundational ones, she says — a core set of platforms, catalogs, and an integrated lake.

With that foundation in place, the returns are specific. Capital One sees roughly three times faster time to market for use cases that launch on data products, measured against how the same work went before, and about 30% lower cost to maintain data once it lives in a standardized data product held to a higher quality bar.

For anyone benchmarking against those numbers, the sequence matters more than the percentages. Capital One measured its usage before picking its first data products and curated for leverage instead of coverage, and only then named an owner for each one and settled the semantics before building anything on top. The three times and the 30% arrived after all of that, which makes them the result of the strategy rather than the argument for starting it.

Watch the full conversation and read the transcript on the Data Faces CDOIQ page for Capital One.

Based on insights from Amy Lenander, Chief Data Officer, and Christina Egea, SVP of Enterprise Data, at Capital One, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [0:39] David introduces Amy Lenander and Christina Egea on location at the CDOIQ Symposium in Cambridge, Massachusetts

- [2:11] Amy on the session title, “Doubling down on data products to drive business value,” and why data products drew the most interest at the prior year’s talk

- [2:54] Christina on top-down buy-in, data maturity, and building data product frameworks with the right ownership

- [4:14] Amy on curating data products by leverage, and why it is the most unusual thing Capital One does

- [5:22] Christina on building for the hundredth use case rather than the first

- [6:26] Christina on the nine categories of data that covered most company usage

- [7:55] Amy on coming to the CDO seat from the business, and getting partners to eat their vegetables

- [9:51] Christina on the hardest part of scaling data products

- [11:18] Amy on adoption as a listening problem, the 70/30 coverage trap, and voting with their feet

- [13:13] Christina on three times faster time to market and 30% lower maintenance cost

About David Sweenor

David Sweenor is the founder of TinyTechGuides and host of the Data Faces Podcast. He is an international speaker, advisor, and the author of eleven books on artificial intelligence, analytics, and B2B marketing, including Generative AI Business Applications, The CIO’s Guide to Adopting Generative AI, and Modern B2B Marketing. With more than twenty-five years in analytics and AI at companies including Alteryx, Tableau, TIBCO, SAS, IBM, and Dell, David advises technology companies on product marketing, content strategy, and go-to-market execution. He holds several patents and has been named a top influencer in data and analytics by Onalytica, Thinkers360, and Analytics Insight.

Connect with David on LinkedIn and subscribe to the Data Faces Podcast for conversations with the people shaping enterprise data and AI.

Footnotes

^[1]CDOIQ Symposium. “The 20th Annual CDOIQ Symposium.” July 21–23, 2026, Hyatt Regency Cambridge, Massachusetts.

https://2026cdoiq.org/

^[2]Capital One. “Interview with Amy Lenander, Chief Data Officer.” Capital One Tech. https://www.capitalone.com/tech/culture/amy-lenander-chief-data-officer-interview/

^[3]Capital One Financial Corp. “Capital One Completes Acquisition of Discover.” May 18, 2025. https://investor.capitalone.com/news-releases/news-release-details/capital-one-completes-acquisition-discover

^[4]Capital One. “Capital One Completes Acquisition of Brex.” April 7, 2026. https://www.capitalone.com/about/newsroom/capital-one-completes-acquisition-of-brex/

Somewhere in almost every data organization, an eager executive has asked questions about the usability of the organization’s data. You’ve ingested the data, put it neatly all in one place, and built the lake. So, can we use it now? Amin Venjara, Chief Data and Product Officer at ADP, hears that question constantly, and he says data sitting in a data swamp was never the point. The value shows up later, in the work you do after capture.

I caught Amin’s session at the CDOIQ Symposium in Cambridge, Massachusetts, and then pulled him in front of a Data Faces Podcast mic. His talk was called “Value Equals Data and Capabilities: a field guide to product-led data platforms,” and the equation in that title is the whole argument. Data is one term. Capabilities are the other. Executives fund the first and forget the second, then wonder why the picture never comes together.

“Data plus capabilities equals the value.”

— Amin Venjara, Chief Data and Product Officer, ADP

About Amin Venjara

Amin Venjara is the Chief Data and Product Officer at ADP, where he oversees the product portfolio and the data that differentiates it, from the smallest nail salon and tire center up to Fortune 500 multinationals. Amin told me the company operates in 140 countries, pays close to 20% of the U.S. working population, more than 40 million people a month, serves over a million clients, and moves roughly $3 trillion a year once you count taxes and government filings.¹ He fell into data through an unlikely door. He grew up a New York Mets fan who dreamed of being a sports broadcaster, played trumpet, and loved math, and he found the thread that ties it all together: data work is technology, people, and storytelling at once.

In this episode, Amin and I discuss:

• Why “it’s all in one place” is not the same as value

• What a data capability is, from the semantic layer to entity resolution

• The contractor who talks about wood while the customer wants a backyard

• How ADP runs its data platform like a product nobody is forced to use

• The one metric that shows whether the data team is earning its keep

Watch the full conversation here:

What counts as a data capability

What’s a data capability? A data capability is the engineering work that turns co-located data into something a business can use, and Amin is precise about what that means. He points to a semantic layer, so that revenue means one thing everywhere and a headcount number resolves the same way in every report. He points to an entity resolution graph, so that the same customer lines up across sales, service, finance, and support. He points to the stitching that takes four systems full of data about a client and produces one coherent picture of that client. None of that comes free when the data lands in the same place.

“Just because it’s in the same place, can you extract the value? No. There’s a lot of work that has to happen, these capabilities.”

— Amin Venjara, Chief Data and Product Officer, ADP

That distinction is where Amin says his own organization turned a corner. Once his team could separate the data from the capabilities, they could finally have a direct conversation with the business about what it takes to create value, because the business could see that the work wasn’t finished the moment the ingestion pipeline ran. Owning the ingredients is not the same as cooking the meal. You can have sales data, service data, customer data, and finance data in the same warehouse and still not know how to get the right answer for revenue or customer count, because the semantic definition that produces that answer is itself a capability you have to build. Data plus capabilities is not a slogan; it forces you to budget for the half of the work that otherwise goes unfunded.

Why the capability half never gets funded

If capabilities are half the value, why do so few organizations pay for them? Amin answers that data teams describe the wrong thing. They walk into a business conversation and talk about sources, ingestion, ETL pipelines, data quality, and lineage, all the machinery under the floor, and the executive on the other side of the table glazes over. To help me understand, he shared an analogy about contractors.

A contractor shows up and says he has the best wood, the strongest concrete, and the finest nails, and he can tell you exactly how many pounds the concrete will hold. The homeowner just wants to entertain people in the backyard. Should we build a deck? A patio-scape? That is the conversation the homeowner wants to have, and the contractor keeps selling lumber.

“This guy’s like, I got the best wood. It’s amazing concrete. And the guy’s like, I just want people to have a good time.”

— Amin Venjara, Chief Data and Product Officer, ADP

The materials matter, but the value has to come first, and Amin’s team took the translation on as their own job rather than waiting for the business to learn to speak data. Business and data speak different languages, and someone needs to translate. When his organization decided that translating value was part of the data team’s mandate, the capabilities stopped being line items the business kept cutting and became things it asked for by name. An unfunded capability is almost always a capability nobody managed to explain.

What a capability buys you in the AI era

The capabilities argument gets sharper the moment you point it at AI, because the AI era is where the cost of skipping capabilities compounds. Every company is racing to put a chat experience or an agent in front of its customers. Amin walks through what that requires. A customer engages inside a SaaS application. That application knows something about the customer, but so does the CRM, the financial system, the sales system, the service system, the ticketing system, and the call transcripts. Each of them holds a fragment. If the chat application has to reach into all of them and resolve the customer, reconcile the definitions, and handle the latency every single time, you have handed an enormous amount of repetitive work to every application you build.

Now consider the alternative. One data product has already normalized the data across those systems. It has resolved the customer IDs, it understands latency, it keeps the definitions right, and maintains the lineage and refresh cadence underneath. Your chat application calls that one data product and gets the full customer context. So does your agent, so does the next application, and the one after that.

“Imagine that when that customer engages, that application could call a data product that had already normalized data across these different systems so that it could call the appropriate context for that customer.”

— Amin Venjara, Chief Data and Product Officer, ADP

The capability layer, not the model, decides what your AI can do. An agent pointed at raw systems reinvents the stitching on every call, and it inherits every inconsistency those systems carry. A data product built once as a real capability gives every agent and every application the same clean context to stand on. When I said back to Amin that he had just articulated why you build data products at all – to stop reinventing the wheel – he agreed without hesitation. Everyone can see the model. Far fewer people fund the client-360 data product underneath it, and that data product is what determines whether the model has anything trustworthy to say.

Run it like a product, and prove it

None of this works if people are forced to use it, which is the most counterintuitive thing Amin has built into ADP’s data organization. His team treats the data platform like a product, and they mean it. You do not have to use us, they tell the rest of the company. We are going to make you want to use us. Just as you choose a favorite provider or a hyperscaler, internal teams get to choose the platform, and the platform has to earn that choice.

“You don’t have to use us. We are going to make you want to use us. You’re going to choose to use us.”

— Amin Venjara, Chief Data and Product Officer, ADP

The philosophy sounds soft until you see how they measure it. A mandate produces compliance and quiet resentment, while chosen adoption produces the kind of pull Amin can point to, like the company’s annual Data and AI Day, which in its second year drew about 2,100 people in person and online. That number is a demand signal, and his team tracks demand the way any product organization would. They watch three things: the outcome of each use case, since value is specific and one use case might be revenue while another is a count of active users; the raw usage, meaning how many users and active users the platform has; and efficiency, which is where Amin offered the most useful number I heard all day. ADP runs a hub-and-spoke model, where a central hub builds the shared capabilities, and the spoke teams take them the last mile to a business outcome. To measure whether the hub is pulling its weight, they divide what the spokes spend on compute and storage by the total platform spend. A higher spoke share of that cost means the hub is more efficient, because the central team enables more value at the edges than it consumes in the middle. Most data leaders cannot tell you a number like that about their own organization.

Build the capability, then the AI has somewhere to stand

Bring it back to the equation Amin started with. When the business cannot use the data, more data rarely fixes it, because the thing standing in the way is a capability nobody named, funded, or built: a semantic layer, an entity resolution graph, or a client-360 data product an agent can call. The organizations getting value from AI right now are the ones that did the unglamorous capability work first, so the model has something solid to stand on.

If you lead a data organization, the Monday-morning version of Amin’s argument is a short audit. Walk your list of value claims and mark which ones are data with a real capability behind them, and which ones are data sitting in a lake with the capability still unbuilt. Then run the second group like a product, and measure whether your hub gets more efficient as the spokes grow. That is a harder conversation than buying another tool, and it is the one that actually moves the needle.

Listen to the full conversation with Amin Venjara on the Data Faces Podcast.

Based on insights from Amin Venjara, Chief Data and Product Officer at ADP, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

• [0:24] What Amin wanted to be growing up: a Mets fan’s dream of sports broadcasting, trumpet, and a love of math

• [2:19] What a Chief Data and Product Officer at ADP does, and the scale of the company

• [3:43] The CDOIQ session, and the premise that value equals data and capabilities

• [4:25] The contractor and the backyard, and why data teams lose the room

• [5:21] The half executives miss: capabilities, not just data

• [6:07] Semantic layer, entity resolution, and the stitching behind a client 360

• [7:11] Treating the data platform like a product nobody has to use

• [11:40] Two languages, and why translation is the data team’s job

• [12:34] The annual Data and AI Day, and 2,100 people showing up

• [13:16] The three things ADP measures: outcomes, users, and efficiency

• [16:27] A data product that gives every chat app and agent full customer context

• [17:25] Stop reinventing the wheel

Frequently asked questions

What does “value equals data plus capabilities” mean?

Amin Venjara’s formula holds that data alone does not create value; capabilities do. A capability is the engineering work that turns co-located data into something a business can use, such as a semantic layer that makes revenue mean one thing everywhere, or an entity resolution graph that lines the same customer up across systems. Executives tend to fund the data and forget the capabilities, then wonder why the business still cannot use what was ingested. The value comes from both terms together, not from the data alone.

What is a data capability?

A data capability is the specific engineering work that converts raw, co-located data into usable value. Amin Venjara points to a semantic layer that gives revenue, headcount, and customer count a single consistent definition, and an entity resolution graph that identifies the same customer across sales, service, finance, and support systems. The stitching that turns several systems into one coherent client 360 is a capability. Putting data in a lake does not produce these; a team must build them intentionally.

How is a data product different from a data lake?

A data lake is a place where raw data from many systems is collected. A data product is a curated, reusable asset built on top of that data, with customer IDs resolved, definitions standardized, and lineage and refresh cadence maintained. In Amin Venjara’s client-360 example, one data product normalizes customer data across systems so a chat application or agent can call it for full context, instead of every application reinventing that work. The lake stores the data; the data product serves it.

What is a product-led data platform?

A product-led data platform is run like a commercial product, with internal teams as customers who choose to use it rather than being forced to. At ADP, Amin Venjara’s team tells the company, “You don’t have to use us; we’re going to make you want to use us.” Adoption is earned through usefulness, and the platform team tracks outcomes, active users, and efficiency the way any product organization tracks demand. The approach replaces mandated compliance with genuine pull from the business.

How do you measure a data platform’s efficiency?

Amin Venjara’s team uses a hub-and-spoke metric: divide what the spoke teams spend on compute and storage by the total platform spend. A higher spoke share means the central hub is enabling more value at the edges than it consumes in the middle, which signals an efficient hub. ADP tracks this alongside use-case outcomes and active users. The metric gives data leaders a concrete way to show whether their central platform is compounding value rather than absorbing it.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

1. ADP. “ADP Corporate Overview.” Accessed August 2026. https://www.adp.com/-/media/corporate-overview/adp-corporate-overview.pdf. ADP publicly reports serving more than 1.1 million clients across 140+ countries and providing payroll to over 42 million workers, roughly one in six U.S. workers. The “$3 trillion a year” and “20% of the U.S. working population” figures are Amin Venjara’s characterizations on the podcast. ↩

▶ Watch the full episode and read the transcript: Sam Pierson on adaptable AI architecture

I have spent my career on both sides of the data platform business. In the first half, I built data warehouses and did data science and analytics at IBM. In the second half, I marketed data platforms at companies like SAS, TIBCO, Alteryx, and Alation. That means I have pitched freedom from vendor lock-in a number of times, and I have also worked with enough buyers to know how much that promise matters to them. So when a CTO tells me freedom is the way to win at AI, I want to know what is in the code and what is in the brochure.

Sam Pierson joined me on the Data Faces Podcast to discuss this notion. As Chief Technology Officer at Qlik, he owns the cloud and AI platforms as well as the development teams for both. As CTO of Talend, Sam joined Qlik as part of the acquisition, so he knows a thing or two about messy enterprise data. His argument runs counter to the instinct to pick a winner. The models will keep overtaking each other, the patterns will keep changing, and the AI conversation itself will look different in a year. What you can control is how quickly you can pivot, and most of that, along with your increasing AI bill, gets decided in the architecture before a model ever processes a token.

About Sam Pierson

Sam Pierson is the Chief Technology Officer at Qlik, where he leads engineering and spends much of his week talking with customers. Before Talend and Qlik, he held engineering leadership roles at Illuminate Education, Datica, SPS Commerce, Veritas, and Symantec. He holds an MBA and a computer science degree from the University of Minnesota.

In this episode, Sam and I discuss:

- Why AI is pushing enterprises to rebuild their data architectures on open formats like Apache Iceberg

- The model selector Qlik wired in early, and how it routes a task to a model that is good enough but 10 times cheaper

- Where AI token costs pile up before a model ever runs, and how a pre-calculated in-memory engine avoids them

- Whether freedom from vendor lock-in is a real architectural property or a story the industry tells itself

- Why 97 percent of enterprises have budgeted for agentic AI while only 18 percent have fully deployed it

Watch the full conversation here:

An architecture you can take apart

Sam has been around long enough to watch the architecture pendulum make several full swings. Huge monolithic databases gave way to open distributed systems in the Hadoop era, the market consolidated again around a handful of big cloud data platforms, and now the momentum is swinging back toward open. Apache Iceberg, an open table format that keeps data readable by any engine, sits at the center of the current swing, and Sam calls it absolutely huge right now.

In his telling, AI itself is driving the rebuild. Companies rethinking their data architectures for AI do not want to be locked into any one system, partly for vendor impartiality and partly because no one can say what next year’s patterns will demand. Nobody wants to re-architect everything and then, six to twelve months later, chalk it up to a sunk cost and then pursue the next new thing. So buyers are demanding modular designs at the storage and compute layers, and again at the metadata and model layers, where each piece can be decoupled and swapped without disturbing the rest.

Qlik has put its weight behind that direction. Its Open Lakehouse is built on Iceberg,^[1] and in January the company joined the Open Semantic Interchange, a vendor-neutral industry effort to standardize semantic definitions so business meaning travels with the data instead of living inside any single tool.^[2]

“We may have a completely different conversation about AI in 12 months... having the ability to migrate or swap out certain parts of the stack in the future is almost a must-have.”

— Sam Pierson, Chief Technology Officer, Qlik

Good enough and 10 times cheaper

The most concrete example of adaptability Sam described is the model selector his team wired into the platform early on. The lesson came from the past decade of software architecture: the move away from the monolith toward microservices, where strong interfaces abstract away details so that any one piece can change without breaking the rest. His team knew the frontier models would keep trading places on the leaderboards, so they refused to hard-wire any single one into the product.

“The router inside of our platform can make the decision of what’s the best model for this task, or what is the best cost-performance trade-off, and maybe route this to a model that’s good enough but it’s 10 times cheaper.”

— Sam Pierson, Chief Technology Officer, Qlik

I asked who makes that choice, the platform or the user, and Sam called it a combo platter. Inside the product, Qlik maintains a family of models that it benchmarks and evaluates constantly, and the user never sees any of it. When customers reach the platform from their own AI ecosystem, whether they are building an agent or connecting from a tool like Claude Desktop via the Model Context Protocol (MCP), they have an explicit choice of model and skills.

The tokens you never have to buy

Back in my semiconductor days at IBM, I thought of data as plumbing, and I said so out loud, which made my ETL developers furious. Sam was more diplomatic when I raised the cost question, but his answer vindicated my old developers. Where the data preparation work happens decides what everything downstream costs.

Point an LLM directly at structured data in a cloud data warehouse and watch the meter run. A single question can take ten minutes as the agent inspects the schema, filters it, and analyzes each round of results, generating tokens the entire time. Multiply that by every question an enterprise asks in a day, and you have found one of the places Sam says companies are burning money without realizing it.

Qlik is sitting on an unusual answer here, and Sam admits some luck is involved. It is better to be lucky than good, he joked. The Qlik engine, developed three decades ago, reads data into memory and precalculates multidimensional analyses and their relationships. The company built an AI-friendly interface to that engine, so once a question reaches it, the inference cost is zero. Compared with the warehouse pattern, Sam says customers get higher-quality answers at lower latency and a much lower token cost per question, an argument that echoes what Josh Howard told me about AI being only as good as the data underneath it.

Is freedom real, or a story vendors tell?

Having written my share of freedom-from-lock-in marketing, I put the uncomfortable question to Sam directly. Is this an architectural property you can point to in the code, or a story the industry tells itself? He gave me the honest version. Freedom starts as a philosophy. Not many vendors operating at Qlik’s scale remain independent, and even strong partners like Snowflake and Databricks would understandably prefer that your workloads run on their platforms. Qlik’s bet is that facilitating choice is worth more than capturing it.

Architects have long memories, though, and that history gives the philosophy teeth. The people building today remember being beholden to a small number of powerful vendors, and they refuse to go back, an instinct Sam now sees extending into model choice. For some buyers, the flexibility is peace of mind they will never cash in. He insists it is more than insurance because, as new technologies and patterns emerge, his customers are actually making the switches.

For a buyer trying to separate real openness from brochure openness, the episode suggests a practical test. Check whether your data sits in an open format you could walk away with. Read the terms for a plain statement that the vendor does not train models on your data, a commitment Qlik makes explicitly. Ask about bring-your-own-key encryption and sovereignty options in the regions where you operate. A vendor that passes is selling you an exit, which is the strangest and most reassuring thing a vendor can sell.

Matt Hayes made the data-side version of this argument a few weeks ago on this show, that owning your data keeps enterprise AI affordable. Sam runs the same principle one layer down, at the models and the compute. And if you are wondering whether the freedom people and the governance people at Qlik get into shouting matches, they do, but Sam swears the matches are usually about hockey or baseball.

The prompt-clone test

Two recent guests set up a question I could not resist putting to a CTO. Donald Farmer argued on this show that software features are dying because AI can clone them, and April Dunford called that claim absolute BS. Sam completed the trilogy without hesitating. “Largely it’s BS.”

“If you literally could just get copied in an afternoon by a prompt, it’s probably not all that valuable of a business.”

— Sam Pierson, Chief Technology Officer, Qlik

The line cuts both ways, and Sam knows it. His team sat down and did the work of articulating what is genuinely hard to copy, and the answer was the engine, the data fabric that feeds it, and the governance, security, and quality machinery wrapped around both. What changes is who shows up at the front door. Sam expects fewer people to log into a dashboard and perhaps a hundred or a thousand times as many to use Qlik in the background, through MCP, from inside tools like ServiceNow or Claude Desktop.

Still the first inning

Qlik’s own research shows how early all of this remains. In the company’s 2025 Agentic AI Study, 97 percent of large enterprises had committed budget to agentic AI, yet only 18 percent reported full deployment.^[3] When I asked Sam why the money runs so far ahead of the deployments, he did not blame the technology. Anyone deploying capital wants to know what downstream productivity comes back for a given token budget, and most organizations cannot answer that yet because the practices are not well understood. We are just in the first inning, he told me.

His prescription follows from that honesty. Start with narrow use cases where the value is provable, and pair outside expertise with the people who hold the business context, because that combination, in Sam’s words, is the magic spot. The models will keep improving, and none of it will rescue an organization that locked itself into last year’s answer. The winning AI stack is the one you can change your mind about, at a price you can defend.

Listen to the full conversation with Sam Pierson on the Data Faces Podcast.

Based on insights from Sam Pierson, Chief Technology Officer at Qlik, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [0:06] Introduction and welcome to the Data Faces Podcast

- [1:16] Sam’s role as CTO at Qlik

- [2:00] First job: the teenage lawn-mowing business

- [2:34] What running Talend taught him about where companies get stuck

- [3:55] How AI changes the data integration job

- [7:18] Baking context and guardrails into the data itself

- [9:06] The role of unstructured data

- [10:13] Open formats, Apache Iceberg, and the architecture pendulum

- [13:07] How hard is it to swap out a component?

- [13:43] The model selector and routing to the 10x cheaper model

- [16:02] Token costs, ROI, sovereignty, and IP protection

- [18:43] Is freedom from lock-in a philosophy or an architecture?

- [21:44] Who picks the model, the platform or the user?

- [22:58] Token-maxxing and the zero-inference-cost engine

- [25:43] Freedom versus governance inside Qlik

- [28:05] The prompt-clone question: “largely it’s BS”

- [31:04] Using Qlik through MCP instead of a dashboard

- [32:33] 97 percent budgeted, 18 percent deployed

- [35:30] Close

Frequently asked questions

What is AI model routing?

AI model routing is the practice of directing each task to the AI model best suited for it, based on capability, latency, and cost, instead of hard-wiring one model into an application. Qlik built a model selector into its platform early, benchmarking a family of models and routing each job to the best cost-performance option. As CTO Sam Pierson describes it, the router can send a task to a frontier model when quality demands it or to a model that is good enough but 10 times cheaper.

How do open data formats like Apache Iceberg reduce vendor lock-in?

Apache Iceberg is an open table format that keeps enterprise data readable by any query engine, so the data layer is not welded to a single vendor’s system. Storing data in open formats lets an organization swap storage, compute, or analytics components as technology changes, without migrating the data itself. That flexibility is a leading reason enterprises are rebuilding their data architectures for AI, and it is why Qlik built its Open Lakehouse on Iceberg.

Where do enterprise AI token costs come from?

Much of an enterprise AI bill accrues before a model produces an answer. When an agent queries structured data in a cloud data warehouse, it inspects the schema, filters it, and analyzes intermediate results, generating tokens the entire time, and a single question can take minutes. Architecture decisions made long before deployment, such as pre-calculating analysis in an in-memory engine with an AI-friendly interface, can cut the token cost per question dramatically while improving answer quality and latency.

Can AI clone any software feature with a prompt?

Qlik CTO Sam Pierson calls that claim largely BS. Software that could genuinely be copied in an afternoon by a prompt was probably never a valuable business, and the market shows no wave of software companies losing their revenue to prompt-built clones. The durable advantage sits below the feature level, in assets like a pre-calculated analytics engine, the data fabric feeding it, and the governance and security machinery around both, which cannot be reproduced by generating code.

Why have so few enterprises fully deployed agentic AI?

Qlik’s 2025 Agentic AI Study found that 97 percent of large enterprises have committed budget to agentic AI, yet only 18 percent report full deployment. The blockers are data quality, integration with existing systems, and an inability to prove what downstream productivity comes back for a given token budget. Sam Pierson describes the industry as being in the first inning, where practices are not yet well understood and narrow, provable use cases are the sensible starting point.

How can a buyer tell whether a vendor’s openness is real?

Test for openness in the contract and the architecture rather than the marketing. Confirm the data sits in an open format you could take with you, such as Apache Iceberg. Read the terms for an explicit commitment that the vendor does not train models on your data, and ask about bring-your-own-key encryption and regional sovereignty options. A vendor that passes those checks is giving you a credible exit, which is the strongest evidence that its openness claims are real.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Qlik. “Qlik Open Lakehouse Now Generally Available, Giving Enterprises Rapid, AI-Ready Data on Apache Iceberg.” Qlik Press Release, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-open-lakehouse-now-generally-available.

^[2]Qlik. “Qlik Joins Snowflake & Industry Leaders to Support Data and AI Interoperability Across the Ecosystem Through the Open Semantic Interchange.” Qlik Press Release, January 27, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-joins-snowflake-and-industry-leaders-to-support-data-and-ai-interoperability.

^[3]Qlik. “Qlik 2025 Agentic AI Study: Budgets Surge, but Data Readiness Delays Scale.” Qlik Press Release, October 16, 2025. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-2025-agentic-ai-study-budgets-surge-but-data-readiness-delays-scale.

In late July, I had the good fortune to spend three days at the 20th annual CDOIQ Symposium at the Hyatt Regency in Cambridge, Massachusetts. TinyTechGuides was the official media partner this year, so I set up at booth 21 with a camera and a couple of chairs and spoke to some of the greatest minds in data, analytics, and AI. I pulled data and AI leaders aside between sessions and asked them what they were actually seeing. The people who sat down included data leaders from Capital One, ADP, Apple, Intuit Credit Karma, and H-E-B, alongside the analysts, authors, and educators who built this field over the last two decades.

Twenty-four conversations later, one concern kept surfacing in different words. Nobody at the event was afraid of the LLMs themselves. They kept coming back to what happens when you point a confident, tireless system at a data foundation that is built on a house of cards. Stacie Christensen of H-E-B said it as plainly as anyone. “AI is a multiplier and a scaler, not a fixer,” she told me. “If the foundation underneath it is broken, there’s nothing positive to scale.” That is the fear underneath the fear, that AI will faithfully amplify whatever mess it inherits and do it faster than anyone can catch.

Three themes ran through most of the conversations. Below is the whole roster with a key point from each, and a link into the full write-up for each theme. Every leader shared real-world, practical insights, so use this as your map and follow the ones that speak to your own work.

Agentic AI and the context problem

Agentic AI was the pervasive topic on the floor this year. Everyone wants agents, and almost nobody has the data plumbing to feed them. And you know what they say about plumbing? Sh** runs downhill. So what exactly are all these agents supposed to run on?

• Stewart Bond, IDC: Latency is the enemy of agentic AI, and the centralized data lake may already be obsolete.

• Kevin Petrie, BARC US: In BARC’s survey, 70 percent of companies say less than half of their unstructured data is usable for AI.

• Capital One (Amy Lenander and Christina Egea): Treating data like a product launches use cases three times faster and cuts maintenance cost by 30 percent.

• Amin Venjara, ADP: Value equals data plus capabilities, so run the internal data platform like a product people choose to use.

• Douglas Laney, Infonomics: The first billion dollar business run by a handful of people and a swarm of agents is coming.

• Terry Dorsey, Denodo: Perfect data quality is a myth, and quality really sits in the eye of the consumer.

• Steven Moskowitz, Industry Forward: Manufacturing runs on wide data, and cognitive KPIs measure how AI changes thinking, not just usage.

Check out the full recap in Agentic AI is only as good as the data you feed it.

The CDO role at 20 years

Thanks to Richard Wang and Stuart Madnick, along with many others, CDOIQ is where the chief data officer role was born, so the 20th anniversary put the job itself on the table. If the job description changes every single year, how do you know when you have the right person in the seat?

• Randy Bean, Data & AI Leadership Exchange: If you are not getting measurable business value from AI, go back to the office and shut it down this afternoon.

• Richard Wang, MIT CDOIQ: The role went from one researcher’s dream to more than 2,000 CDOs, and data archaeology is the tax you pay when you lose the people who hold the knowledge.

• Stuart Madnick, MIT: By every measure he has seen, attackers are using AI more aggressively and efficiently than defenders.

• Peter Aiken, Virginia Commonwealth University: The best CDOs have been fired three times, and data belongs in the business rather than in IT.

• Mark Ramsey, Ramsey International: A shadow pit crew forms when the CDO does not own data access, and half of CDOs do not survive three years.

• John Ladley, author and advisor: ChatGPT built the rudiments of a data model in 38 seconds, and a 40-year-old profession still has not figured out its 1.0.

• Leigh Felton, AI for Job Security Foundation: You cannot de-bias a system built to recognize historical patterns, so govern AI like an environment.

Read the whole conversation in 20 years in, the chief data officer’s job is being rewritten.

Data quality and trust as the AI foundation

Agentic AI drew the crowds and the CDO role got personal, but data quality was the theme nobody could avoid. Well, it was the CDO IQ (Information Quality) event, so what did you expect. How can you trust what AI tells you when you cannot trust the data underneath it?

• Tom Redman, the Data Doc: Only 3 percent of companies he tested met basic data quality standards, and the follow-up work suggests no improvement since.

• Fern Halper, AI Foundations Group: AI will expose every weakness your company has, and trust in unstructured data trails structured data by roughly 20 points.

• Stacie Christensen, H-E-B: AI is a multiplier with nothing positive to scale when the foundation is missing, so govern at the moment of data creation.

• Leticia Naqvi, Apple: Data maturity is the AI readiness pillar everyone forgets, and it sits under every AI output you will ever trust.

• Danette McGilvray, Granite Falls Consulting: Most data cleanup is a crime scene where nobody investigates, so start by asking whether anyone actually cares about the problem.

• Jonathan Agee, Validatar: The state of data quality earns a C, and prevention belongs in development the way QA does in software.

• Kelley Kassa, BARC US: Only 9 percent of finance AI in North America is in production, because the agent does not go to jail when the number is wrong.

• Gwen Thomas, The Data Governance Institute: Organizations govern for the same reason cars have brakes, so they dare to go fast.

• Dan Everett, Insightful Research: AI adoption is really change management, and claimed productivity gains shrink once experts validate the output.

• Intuit Credit Karma (Veenit Shah and Puneet Singh): An AI remediation agent cut root cause investigations from 30 to 40 minutes to a few, across 40,000 columns.

See what these data leaders had to say in AI will expose every weakness in your data.

What I heard across three days

Twenty-four conversations, and one common thread ran through all of them. The leaders getting real value from AI had done the unglamorous work first, the governance, the ownership, and the data quality that lets an agent act on something true. Tom Redman put a number on how rare that still is. Kevin Petrie showed how much unstructured data is not ready. Randy Bean gave everyone permission to kill what is not working, and Stacie Christensen named why it matters, because AI multiplies whatever you hand it.

What it came down to was this, handing a fast, confident system a shoddy data foundation nobody trusts, then watching it scale the mess while the whole thing circles the drain. Every leader in that room already knew what to do about it, and most of them have been building toward it for 20 years. AI just raised the stakes on finishing the job.

All 24 conversations are live on the Data Faces Podcast on-location hub. New interviews and studio episodes drop every couple of weeks, and I would rather you hear these leaders in their own words than take my summary for it.

Share

Leave a comment

Frequently asked questions

What is the CDOIQ Symposium?

The CDOIQ Symposium is the flagship annual event for chief data officers and data leaders, founded out of the data quality research programs at MIT and held in 2026 for the 20th time at the Hyatt Regency in Cambridge, Massachusetts. It gathers data and AI executives, academics, and practitioners for sessions and working conversations on data quality, governance, and the evolving role of the chief data officer. TinyTechGuides attended as the official media partner and recorded 24 on-location interviews for the Data Faces Podcast.

What are data and AI leaders most worried about with AI right now?

Based on 24 conversations at CDOIQ 2026, the dominant worry is about the foundation underneath the models rather than the models themselves. Leaders repeatedly described AI as a multiplier that faithfully scales whatever data quality, governance, and context it inherits, which means weak foundations produce fast, confident, and wrong results. Data quality, unstructured data readiness, and ungoverned agents were the three fears that came up most.

What were the three themes from CDOIQ 2026?

The 24 interviews grouped into three themes. The first is agentic AI and the context problem, meaning agents cannot act without real-time, governed data where it lives. The second is the chief data officer role at 20 years, and how AI is rewriting the job in real time. The third is data quality and trust as the foundation every AI output depends on. Each theme has its own detailed write-up linked from this page.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

▶ Watch the full episode and read the transcript: The death of SaaS is a myth

At a recent industry event, I listened to a room full of data and AI leaders reach a consensus that should terrify every software vendor. Any capability in modern software, the group agreed, can now be replicated by a competitor with a prompt, more or less overnight. It’s the same logic powering the SaaS-is-dead obituaries filling my LinkedIn feed. Why buy software at all when you can vibe-code your own? I’ve heard versions of that claim on this show before, and I wanted to test it against the person who has spent a decade helping B2B tech companies figure out what makes them different. So I put the claim straight to April Dunford.

Her verdict took two words, and I’m cleaning one of them up for print. Absolute BS.

The rest of our conversation was April building the case, story by story, for why the clone-anything claim falls apart inside real software companies. If you market or sell a technology product, her argument changes how you answer the question every buyer eventually asks. Why should I pick you?

“That is the most ridiculous statement I’ve ever heard. That is just fundamentally untrue.”

— April Dunford, Positioning Consultant and Author

About April Dunford

April Dunford is a positioning consultant who works exclusively with B2B technology companies, helping them get precise about who they compete with, what they have that nobody else does, and which customers that difference fits best. Before going solo, she spent 25 years as a startup executive, running marketing at seven B2B tech startups, most of which were acquired. Her positioning methodology has now been battle-tested with more than 300 technology companies, from early-stage startups to Google and Epic Games. She is the author of Obviously Awesome, newly updated and expanded in a second edition released in February 2026, and Sales Pitch, and she hosts the Positioning with April Dunford podcast.

In this episode, April and I discuss:

- Why “we have AI” now carries about as much weight as “we have a login screen”

- What’s wrong with the claim that any feature can be cloned with a prompt

- How one CRM’s unusual data model won deals its biggest competitor couldn’t touch

- Why even companies with enormous resources can’t copy whatever they want

- Why positioning projects don’t fail for the reason most marketers think

Watch the full conversation here:

“We have AI” is a login screen

Before we got to the clone myth, April and I talked about a frustration every marketer has felt over the past three years. A product team ships a new capability, leadership wants AI in the headline, and nobody has asked what the AI does for the customer. April told me this was a far worse problem two years ago, when every company that wasn’t AI-native wanted to sprinkle AI on the messaging and call it strategy. She thinks the industry has matured past that stage. She hasn’t worked with a company in two years that didn’t have AI running somewhere in the product.

That maturity is also why the claim has stopped meaning anything. When every product in the category uses AI, saying so is a statement of parity, and parity is not positioning. April’s test cuts through it in three questions. What capability does the AI enable? Is that capability different from what competitors offer? And why should a customer care, which in her world comes down to whether it makes them money, saves them money, or reduces their risk?

“If there is no difference in the capability, then we have nothing to talk about. It’s like saying we have a login screen. Sure, buddy. Everybody’s got a login screen.”

— April Dunford, Positioning Consultant and Author

If your AI capability is table stakes across the category, she argues, you can stop talking about it. The positioning work starts where the parity ends.

It’s just a simple matter of programming

So what about the clone claim itself? April has heard it before, and it predates generative AI by decades. Twenty years ago, she worked for a CTO with a standing answer for it. Whenever the team worried that Google or SAP could build their product too, he would lean back, cross his arms, and say, yep, it’s just a simple matter of programming. We can build it, and they can build it. But they won’t, and they can’t, for a thousand reasons.

The first of her thousand reasons is that nobody starts from the same place.

“If anybody could wake up tomorrow and vibe code the backend of Salesforce, it’d already be done. That is a 20-year computer science project.”

— April Dunford, Positioning Consultant and Author

April is happy to concede the small stuff. If you vibe-coded a scheduler for your kids’ after-school activities, good for you. An enterprise product is a different animal. It carries years of dependencies, security requirements, and a data model that somebody sweated over, and none of that regenerates from a prompt.

The second reason is the business around the software. Early in her career at IBM, April needed three developers for six months to build one feature, inside a business unit doing five billion dollars in revenue. It took her four months to make the case, and the answer was still no, because those three developers were more valuable somewhere else. Even a company with seemingly unlimited resources doesn’t get to build whatever it wants. Every engineering hour has to make sense for the existing product, the existing customers, and the go-to-market machine that sells it.

The moat is in how the product was built

Every company April works with has capabilities its competitors cannot copy. That isn’t a consultant’s polite reassurance. After three hundred client engagements, she considers it settled.

“I have yet to work with a company that does not have differentiated capabilities that their competitors cannot and will not build.”

— April Dunford, Positioning Consultant and Author

She gave me an example from earlier in her career. A company began life as a contact manager and nearly went broke, and then a consulting pivot landed them a bank that wanted a CRM. So they built one, on the contact-manager codebase they already had. Ten years later, that product’s underlying data structure was unlike any other CRM on the market. Every other CRM keys its customer records to the company. Theirs keyed on people. That accident of heritage let it model relationships that have nothing to do with employers, like two executives who sat on a board together or belonged to the same golf club. Investment bankers, who live and die by who knows whom, bought it in droves. The dominant CRM vendor of the day couldn’t respond, because matching the feature meant throwing out its entire data structure and rebuilding a two-billion-dollar product around a new one. As April put it, vibe coding or not, you’re not copying that feature.

The same dynamic showed up in a workshop the week before our conversation. On a feature checklist, her client and its acquisition-built competitor look identical. In deals, the client wins, because its single product shares data across the entire process and the competitor’s three bolted-together products don’t talk to each other.

None of this means old code is an advantage. When I asked April whether legacy was becoming a moat, she told me she meant nearly the opposite. Successful products get built when a founder looks at the incumbent and decides they would build it differently. HubSpot didn’t set out to copy Salesforce. Its founders built a CRM informed by everything they knew about marketing, in reaction to a product where marketing came later as a bolt-on. Differentiation traces back to a philosophy about the right way to solve the problem, and that philosophy gets baked into the architecture where a checklist never looks.

Why positioning fails in isolation

Near the end of our conversation, I asked April why positioning projects always feel rushed, expecting a lecture about skipped customer interviews. She laughed and warned me this was the episode of her being contrary. Rush, she said, is good. Getting a team aligned on positioning should be fast if you follow a structured process instead of vibes. Positioning breaks down when marketing tries to do the work alone, keeping sales out because they’ll just disagree with everything and product out because they’ll have opinions. Yet those two teams hold the answers marketing needs most.

“Who ends up on a short list against us? Nobody knows the answer to that question better than sales.”

— April Dunford, Positioning Consultant and Author

In enterprise software, sales is a remarkable proxy for how customers make purchase decisions, and product teams are often sitting on secret sauce that nobody markets because nobody else understands it. April’s fix is to get the cross-functional team in one room, have the little fight about what matters, and come out aligned. Executing a position in the market can take years. Agreeing on one shouldn’t.

Your moat didn’t disappear

SaaS isn’t dying. AI has made surface-level differentiation cheap to copy, and that sliver of truth is carrying the entire obituary. A screen layout, a clever workflow, a feature that lives where everyone can see it- those may well be gone by your competitor’s next release. What a prompt cannot reach is the differentiation buried in your product’s architecture, its heritage, and the philosophy it was built on. Finding it means getting all the way down to the guts of the thing, with the people who built it in the room, and translating what you find into value a customer cares about. That was true before AI, and April has three hundred companies’ worth of evidence that it is still true now.

Listen to the full conversation with April Dunford on the Data Faces Podcast.

Based on insights from April Dunford, positioning consultant and author of Obviously Awesome and Sales Pitch, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [2:25] Small-town valedictorian, an acceptance to med school, and the decision to choose engineering instead

- [3:54] When tech marketing departments were staffed entirely by engineers

- [7:15] The 25-page AI-generated positioning document, and “you read that?”

- [12:11] April’s so-what chain, from capability to customer value

- [14:20] The claim that any feature can be cloned with a prompt, and April’s two-word verdict

- [15:00] “It’s just a simple matter of programming”

- [17:48] Why IBM said no to three developers in a five-billion-dollar business unit

- [20:10] The CRM whose primary key was people, and the investment bankers who loved it

- [27:14] HubSpot, Salesforce, and building in reaction to the incumbent

- [30:18] Rush is good, and the part of positioning that gets shortchanged

- [33:29] Why sales knows the shortlist better than anyone

Frequently asked questions

Is SaaS dead because AI can copy any product?

No. The death-of-SaaS argument rests on the idea that any software feature can be cloned with a prompt, and positioning expert April Dunford calls that claim fundamentally untrue. Surface features can be copied. The differentiation that wins deals lives in a product’s architecture, data model, and founding philosophy, and a competitor cannot replicate those without rebuilding its own product and migrating its install base. SaaS businesses defend that moat the same way they did before generative AI.

Can a competitor really clone any software feature with AI?

Not the features that matter. Positioning expert April Dunford, speaking on the Data Faces Podcast, calls the claim that AI lets competitors copy any software feature fundamentally untrue. Surface-level software features can be copied quickly, but an enterprise product carries years of dependencies, security requirements, and data-model decisions that a prompt cannot regenerate. Rebuilding the backend of Salesforce, she notes, is a 20-year computer science project. Even competitors with enormous resources ration engineering hours by business case, so most theoretically copyable features never get built.

What makes a software feature hard to copy?

Architecture and heritage make a software feature defensible. A capability is hard to copy when it depends on decisions buried deep in how the product was built, especially its underlying data structure. April Dunford’s example is a CRM whose records are keyed on people rather than companies, which let it map relationships between executives that no rival could see. Matching that feature would have required the dominant CRM vendor to throw out its data structure and rebuild a two-billion-dollar product.

How should you position AI capabilities in a B2B product?

Run April Dunford’s so-what chain. Name the capability the AI enables, ask whether that capability differs from what competitors offer, and then translate the difference into customer value, meaning it makes the customer money, saves the customer money, or reduces risk. When every product in a category uses AI, the bare claim “we have AI” is a statement of parity, like advertising a login screen, and it belongs nowhere near your headline.

Who should be involved in positioning work?

A cross-functional team. April Dunford finds that positioning fails when marketing does the work alone and treats sales and product as obstacles. Sales knows who shows up on competitive shortlists better than anyone in the company, and product teams understand differentiated capabilities that nobody else can translate into value. Her method brings those groups into one room with a structured process, has the argument once, and reaches alignment fast. Executing a position takes years, and agreeing on one should not.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

▶ Watch the full episode and read the transcript: Matt Hayes on trust and freedom in enterprise AI

I have spent 25+ years around data software, first as a practitioner building predictive models, data warehouses, and analytics reports, and later in product marketing, exclusively for data, analytics, and AI companies. In every one of those roles, a similar pitch kept recurring. Get your data ready. For most of that time, ready meant ready for a dashboard, where a person could look at the report, catch the number that seemed off, and fix it before it did any harm. That safety net is disappearing. Once an AI agent acts on the data instead of a person, no one is there to spot the error and correct it.

Matt Hayes joined me on the Data Faces Podcast to dig into the difference between getting data ready for analytics versus AI. As General Manager of Qlik’s Data Business Unit, he owns the data layer underneath everything the analytics side touches, from ingestion and transformation to the open lakehouse below it. His argument is that getting data ready for AI is a much higher bar than getting it ready for analytics.

“A human will look at an analytics application and say, wait a minute, this doesn’t add up. An agent will just assume the data is all good and move forward.”

— Matt Hayes, General Manager, Data Business Unit, Qlik

Most data leaders already accept that trust is what decides whether enterprise AI scales or stalls out. Brendan Grady, EVP and General Manager of Analytics and AI at Qlik, made that case from the decision side in an earlier Data Faces conversation, so we won’t repeat ourselves here. The discipline that makes enterprise AI trustworthy turns out to be the same discipline that keeps it affordable, and Matt has one word for it. Freedom.

About Matt Hayes

Matt Hayes is the General Manager of the Data Business Unit at Qlik, where he leads product management, product marketing, and engineering across the company’s data integration, transformation, and open lakehouse portfolio. He came to Qlik through its acquisition of Attunity and previously ran Qlik’s SAP business and strategy. He has worked in and around SAP since 1998, starting as a basis consultant and later building Gold Client, a test data management product for SAP environments. Outside of work he is a private pilot who flies a Piper Archer and a Saratoga out of the Chicago area, and he will tell you there is something special about breaking through a gray Chicago cloud layer into blue sky while eight million people below are stuck under it.

In this episode, Matt and I discuss:

- Why data that was fine feeding a dashboard can break the moment an agent acts on it

- The difference between analytics-ready and AI-ready data, and why AI sets the higher bar

- Context, trust, and freedom as operating principles, and why Matt ranks freedom first

- How a customer-defined trust score turns data quality into a signal an agent can act on

- Why the economics of enterprise AI trace back to decisions about where data lives

Watch the full conversation here:

AI-ready is a higher bar than analytics-ready

For years, well before AI took over every roadmap, the goal was to get your data analytics-ready. Now every conversation is about getting it AI-ready. Is there a real difference between the two, or is AI-ready just analytics-ready with a fresh coat of paint?

According to Matt, analytics-ready data is data a person can sanity-check before acting on it, whereas AI-ready data has to clear a higher bar, because that human pause is gone. Hand a business process to an agent, and it takes whatever the data says and runs with it.

So what happens when it goes awry? For example, a customer means to order ten thousand units and fat-fingers the entry as a million. In a world of connected agents, that bad number does not stay put in one field. An agentic workflow reads it as real demand and starts sourcing raw materials from five or ten suppliers. Trucks and ships get loaded with material nobody ordered before a person catches it. Nobody, as Matt said, wants to be a news story.

The trust scores and quality checks that catch that number before an agent acts are exactly what Brendan Grady and I got into on the analytics side. What matters here is the asymmetry. Bad data used to cost you a correction and a hand slap. Once an agent acts on it, bad data costs you whatever the agent already did before anyone noticed.

Context, trust, and freedom

Matt runs his business unit on three principles, context, trust, and freedom. When I asked which matters most, he didn’t flinch. Freedom. He means something very specific by it. Customers should be free to define their own architectures and use their own data however they want, with whatever tools fit the job.

He traces that conviction to his years in the SAP world, and he is careful about how he frames it. SAP is a company he respects, one that does a lot well for its customers. His concern is a pattern that can show up at any large platform vendor as it broadens its scope. Once a vendor possesses your data, it can be tempted to treat that data as a bargaining chip, nudging you toward its own tools because switching gets harder the more you hand over. The real question is who holds the keys to your data kingdom.

“People want to do things with their vendors. They don’t want their vendors to do things to them.”

— Matt Hayes, General Manager, Data Business Unit, Qlik

This is where Matt plants Qlik’s flag. The company does not persist your data, so it has no reason to lock you in. In his telling, freedom is a property of the architecture, something a buyer can check for rather than take on faith.

The data factory and the finished good

Matt thinks about data integration as a manufacturing process. You pull raw material from many sources, combine it with other materials, inspect the quality, and deliver a finished product to the people who need it. The data contract sets the specification, the fields the product must contain and the service levels it has to meet, and the data product is what comes off the line.

“When we look at the data integration business, that’s a manufacturing business. The data product is a finished good of that process.”

— Matt Hayes, General Manager, Data Business Unit, Qlik

I spent my early career in a literal version of this. At IBM I did yield characterization in semiconductor manufacturing, watching dashboards that flagged every tool drifting out of spec, and nearly everything showed up red. There were never enough engineers to chase that many red lights, so people stopped chasing them. A quality signal that flags everything ends up flagging nothing.

That is the trap a trust score has to avoid, and it is why Qlik lets the customer define the trust score. You weight the measures that matter for a use case, things like freshness, completeness, and whether every source is reporting, then set a threshold where an agent pauses rather than act on data that has slipped. Nobody panics at two percent scrap, but past five or ten percent it eats into margins. A trust score points the same logic at data, telling you when the finished product is good enough to ship, the discipline behind Qlik being named a Leader in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions.^[1]

Freedom is the economics argument

Up to this point the risk has been a wrong decision, an agent acting on a number it should have questioned. A second risk has nothing to do with accuracy. It is the right decision you cannot afford. Matt described the CFO who walks into a project six months in and points out that the experiment everyone is proud of is now burning a million dollars a month. The work is sound, but the bill is not sustainable.

Freedom shows up directly on the budget. If you do not lean into the freedom of your own data, it costs you, in flexibility, in infrastructure, and in the architecture you get boxed into. Storing everything in a proprietary format runs up your storage bill and your switching cost at once, so he points customers toward open formats instead.

Qlik Open Lakehouse is built on Apache Iceberg, an open table format that keeps data queryable by any engine and shrinks the storage footprint instead of trapping it in one vendor’s system.^[2] Compute gets the same treatment, since Qlik’s analytics engine does much of its work in memory to hold down cost as workloads grow. On the day we recorded, Qlik shipped agentic data engineering capabilities aimed squarely at this, letting teams describe a pipeline in plain language and stand up trusted, AI-ready data faster.^[3]

A vendor that does not hold your data cannot pad your bill to keep you, and an architecture built on open formats can show you the savings instead of promising them. So Matt’s counsel to data leaders is to claim that ground on purpose.

“Lean into that concept of freedom. Take the position with all your vendors that this is our data and we want to do a lot with it.”

— Matt Hayes, General Manager, Data Business Unit, Qlik

A business case that survives the pilot

A lot of AI projects stall after the demo, and Matt doesn’t count that as failure. He compares this moment to 2007, when the iPhone arrived and everyone downloaded the app that turned the screen into a pint of beer or a lighter. It was thrilling and mostly useless, and it was also the start of something real. Enterprise AI is in that phase now, where experimentation is the work rather than a detour from it.

A project dies when cost outruns value before the use case is proven. Expect three, four, or five iterations before a use case earns its keep, and protect the economics while you get there. Gate every agent behind a data product that clears its trust score, so nothing autonomous runs on data you have not vetted. Build on open formats like Iceberg so storage cost and lock-in do not compound. Put the shared terms in a data contract, so the word revenue means the same thing to five departments instead of five different things in the same meeting.

Human oversight belongs in that design, though not as a rubber stamp on every decision. Your job is to decide where a human hand changes the outcome and where it only slows things down. Keeping AI reliable and keeping it affordable look like competing goals. Matt’s whole argument is that they are the same goal, reached by owning your data instead of renting it back from the vendor that holds it.

Listen to the full conversation with Matt Hayes on the Data Faces Podcast.

Based on insights from Matt Hayes, General Manager of the Data Business Unit at Qlik, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [0:05] Introduction and welcome to the Data Faces Podcast

- [0:52] Who Matt is and the Qlik data portfolio, from Attunity to open lakehouse

- [1:46] Life outside work: flying a Piper over Chicago

- [3:25] Lessons from the SAP world on why enterprise data is hard to move and trust

- [5:14] Context, trust, and freedom, and the case against vendor lock-in

- [7:41] AI-ready versus analytics-ready data

- [8:11] Why an agent acting on bad data is different from a human reviewing a dashboard

- [11:05] What AI-ready means, and the trust score

- [14:08] Data products as the finished good of a data factory

- [16:31] A supply-chain near-miss and the limits of a human in the loop

- [21:28] Making trust scores meaningful instead of ignored

- [25:36] Where the data side meets the decision side

- [29:09] Context, semantics, and meeting customers where they are

- [31:51] Building an AI business case that survives the pilot

- [36:17] Close

Frequently asked questions

What is AI-ready data?

AI-ready data is data trustworthy enough for an autonomous AI agent to act on without a person checking it first. It clears a higher bar than analytics-ready data because there is no human in the loop to catch a number that looks wrong. Reaching that bar means the quality, governance, and freshness of the underlying data product are strong enough that an agent can read the data, make a decision, and pass that decision downstream without introducing a costly error.

How is AI-ready data different from analytics-ready data?

Analytics-ready data is data a person can sanity-check before acting on it. When a dashboard shows a number that looks off, a human pauses and investigates. AI-ready data removes that pause, because an autonomous agent takes the data at face value and acts on it. As Qlik’s Matt Hayes frames it, the bar is higher for AI because a bad number no longer costs you a quick correction. It costs you whatever the agent already did before anyone noticed.

What is a data trust score?

A data trust score is a customer-defined, weighted measure of whether a data product meets the quality bar for a specific use case, across signals like freshness, completeness, and source availability. Teams set a threshold, and when the score drops below it, the right people are alerted and can take action. The goal is to keep the score meaningful, so it flags what genuinely matters rather than turning everything red and getting ignored.

How does data architecture affect the cost of enterprise AI?

Where your data lives and how it is stored drives much of the cost of enterprise AI. Storing everything in a proprietary format runs up both storage bills and switching costs. Open formats like Apache Iceberg and an open lakehouse shrink the storage footprint and avoid lock-in, while in-memory processing holds down compute. Matt Hayes calls this data freedom, and it is why the discipline that keeps enterprise AI trustworthy is also the discipline that keeps it affordable.

Where should data leaders start when building an AI business case?

Start by expecting several iterations before a use case proves its value, and protect the economics along the way. Gate every AI agent behind a data product that clears its trust score, so nothing autonomous runs on unvetted data. Build on open data formats so storage cost and vendor lock-in do not compound. Define shared business terms in a data contract, so a word like revenue means one thing across departments before an agent ever uses it.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Qlik. “Qlik Named a Leader in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions.” Qlik Press Release, February 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-named-a-leader-in-the-2026-gartner-magic-quadrant-for-augmented-data-quality-solutions.

^[2]Qlik. “Qlik Open Lakehouse Now Generally Available, Giving Enterprises Rapid, AI-Ready Data on Apache Iceberg.” Qlik Press Release, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-open-lakehouse-now-generally-available.

^[3]Qlik. “Qlik Delivers Agentic Data Engineering in Qlik Cloud to Help Enterprises Build Trusted Data for AI.” Business Wire, June 30, 2026. https://www.businesswire.com/news/home/20260630874468/en/Qlik-Delivers-Agentic-Data-Engineering-in-Qlik-Cloud-to-Help-Enterprises-Build-Trusted-Data-for-AI.

Donald Farmer told me a Tesla is probably a better driver than he is. After a bit of research, the data support his claim. Self-driving cars have become measurably safer than people behind the wheel, and Waymo’s driverless cars, measured across tens of millions of miles, were involved in roughly 91% fewer serious-injury crashes than human drivers on the same roads.^[1] So if the machine drives better, what does it still lack?

You never switch on the car and hear it say, not today, it’s a lovely day, let’s skip the office and drive out to Whidbey Island instead. The machine takes you anywhere you point it, and it never once decides where to go of its own volition.

AI is getting genuinely good at the how, the driving, the producing, and the doing. The why is still ours. Donald has spent his whole career building data products, first at Microsoft and then at Qlik, and he came on the show with an idea that I found super fascinating. For decades, a data vendor’s edge came from two things: a feature nobody else could match, and the community of practitioners who grew up around the product, who identified with it and built a whole way of working around it. AI has copied the features, and it is now taking over the interpretation and judgment that those practitioners used to own. That leaves software vendors with a choice most have not yet made.

“The AI is nowhere close to bringing the why.”

— Donald Farmer, Principal, TreeHive Strategy

About Donald Farmer

is the Principal of TreeHive Strategy and VP of Innovation at Nobody Studios. He has spent more than thirty years designing data and analytics products, including a long run as a design and innovation leader at Microsoft and at Qlik, where he helped build the second-generation product, Qlik Sense. His first data analytics products date back to the 1980s, which, as his son likes to point out, puts the start of his career closer to the Second World War than to today. He is the author of Embedded Analytics from O’Reilly, and he writes the excellent Creative Differences newsletter on Substack.

In this episode, Donald and I discuss:

- Why buying a data platform used to mean buying into a practice, not just a product

- How AI erased the feature moat, and what is left for a vendor to compete on

- Why a Tableau analyst stays a Tableau analyst, and an Oracle DBA stays an Oracle DBA

- Why “a human in the loop” is so often a cop-out

- The four human attitudes a system can model but never feel

Watch the full conversation here:

Practice is not process

Early in our conversation, I asked Donald to help me understand the difference between process and practice. Process is easy, he said. A business process is the way you do something, the steps you run from start to finish. In fact, my Claude Code setup is full of processes or workflows. Practice is a different beast. Think about the businesses in your town. There are retailers, builders, and contractors. You talk about the doctor’s practice, the lawyer’s practice, the architect’s practice, yet you would never call the local builder’s work a building practice, even though the builder is skilled and careful. Something separates the two. What is it?

A practice carries professional standards, a code of conduct, and an ethical stance. It carries a community and a point of view. People define themselves by it, which is why a lawyer will tell you they are a lawyer long before they tell you where they work. A practice is something you are, not just something you do between nine and five.

“A practice is more than just a business process that you do from nine to five. It’s an attitude to the world, and it’s an ethical approach.”

— Donald Farmer, Principal, TreeHive Strategy

When an organization runs on SAP, SAP does not just record what the business does. It defines how the business works, because the set of capabilities it offers steers you into operating a particular way. The same goes for the big ERP and CRM systems. Donald’s wry aside was that the humble spreadsheet has survived all of them because it imposes no methodology at all and fills in every space the big systems leave behind. The technology and the business process get so tightly bound together that the software becomes a working definition of the company itself.

When you can clone any feature with a prompt

At the BARC Data and Analytics Retreat where Donald and I met, someone made a claim that got the whole room nodding. You can recreate almost any feature in modern software with a prompt. I believe that is largely true, and it should terrify a lot of product teams. The technical moat that vendors spent years digging, the clever feature nobody else had, can be reverse-engineered and rebuilt by lunchtime.

I hear a version of this from vendors all the time. They ask me how they are different, and more often than I would like, the honest answer is that they are not. One client pushed back and asked about their history. I had to point out that history is not much of a defense when you are competing against companies that have been selling to the enterprise since 1911. So why would a buyer pick you at all?

“Why would anyone use your software when they can just vibe-code it? The features and functions are so easily replicated now that that moat has gone.”

— Donald Farmer, Principal, TreeHive Strategy

You could try to run faster than everyone else and ship features quicker than the competition can copy them. Donald does not think it lasts. You might be different today, and then tomorrow someone reverse-engineers your prompt and the difference evaporates. Speed alone leaves you on a treadmill against a model that ships weekly and costs cents.

Why a Tableau analyst stays a Tableau analyst

Once features stop holding customers, something else has to. Donald’s answer is the practice, and the way it shows up in the real world is identity and community. He told a story from his Qlik days that makes the point with no abstraction at all.

Back in 2013, Tableau showed off a native Mac client at its customer conference and the room gave it a standing ovation. Donald carried the news back to Qlik and expected some concern. Instead the boardroom looked around and asked whether anyone knew a single customer who used a Mac. The answer was no. Two companies that every analyst firm filed under the same self-service business intelligence category turned out to serve fundamentally different practices. Tableau users saw themselves as creatives, close cousins to designers and photographers, people who happened to think visually. Qlik users saw themselves as business people and application builders working on their ThinkPads. Same category on the Gartner grid, completely different ways of seeing the work.

“It’s that shared belonging. Not just a website and a forum, but a sense of purpose that they can build around the product.”

— Donald Farmer, Principal, TreeHive Strategy

That belonging is stickier than any feature. MongoDB has binary-compatible alternatives from Amazon and Microsoft, yet MongoDB developers stay MongoDB developers because the community and the intentionality come with the name.^[2] An Oracle DBA is an Oracle DBA, not a database administrator who happens to use Oracle. A SQL Server developer carries the same identity. Donald pointed to a brand-new company as the current example of someone building this on purpose. François Ajenstat, a longtime product leader from Tableau, recently launched Golden Analytics, and its signature design idea is a “slider of autonomy” that lets a person dial how much the software decides versus how much they do themselves.^[3] That’s a vendor trying to grow a community of practice and engineer the human’s place in it at the same time, which is exactly the move Donald argues the rest of the industry needs to learn.

A human in the loop is a cop-out

I brought up an idea from a recent guest, Doug Laney, who described the autonomous, self-driving business, and asked Donald whether his view lined up.^[4] Although we agreed there is potential, we also acknowledged that there is a long way to go. And the term human in the loop? Fuggettaboutit.

“A human in the loop is a cop-out. It’s a way of saying we haven’t really thought about this, so we’ve shoved a human in there, and that’s our answer to things.”

— Donald Farmer, Principal, TreeHive Strategy

His point is that we reach for the human in the loop because it feels responsible, when it is often the opposite. Humans do not scale, and humans are fragile. When a system is making thousands of decisions a second, a person clicking go cannot meaningfully supervise any of them, and I will admit I sometimes wave Claude through a task without fully knowing what it is doing. Donald also turned the explainability argument back on us. We demand that AI explain its reasoning, yet most of us have worked for a manager whose decisions made no sense at all and who kept the job regardless. Human judgment was never as transparent and explainable as we pretend.

For software makers, the work that stays human is not a feature, it is an attitude a system can model, recommend, and even simulate, but cannot feel. Donald names four of them, trust, doubt, ambition, and care, and he treats each one as a deliberate design choice rather than an afterthought. A system designed around doubt, for example, would not hand you a tidy, well-formatted report. It would grill you with the hardest, most skeptical questions a board member could ask, so you walk into the room prepared to defend the number rather than just present it. AI tends to smooth all of that friction away, and Donald argues we do ourselves no favors by letting it.

Build AI into a practice, or get absorbed by it

This is the choice Donald leaves vendors with, and it is the reason the practice idea is more than a nice to have. A company with a practice has a place to put AI. It integrates the technology into a workflow it already defines, with clear limits and a point of view about what the work is for. A company without one watches AI replace its value proposition one piece at a time until the product is just a thin layer the model could have generated anyway. You either build AI into a practice you own, or you get integrated into the AI.

Which brings us back to the Tesla that will not drive itself to Whidbey Island. The machine is getting genuinely intelligent, and Donald, who studied philosophy and history, happily admits Claude raises ideas in conversation that he would not have reached on his own. What it lacks is purpose, meaning, and direction. My favorite version of this came near the end, when Donald reminded me that the smartest AI anyone ever imagined is Marvin the Paranoid Android, forever asking what the point is. You will know your AI is intelligent on the day it tells you to give it all up and go meditate, and that is the one thing it will never say. A vendor’s real question was never how to compete with AI. It is what work the AI is there to serve, and the honest answer is a practice rather than a feature list.

Listen to the full conversation with Donald Farmer on the Data Faces Podcast.

Based on insights from Donald Farmer, Principal of TreeHive Strategy, featured on the Data Faces Podcast.

Podcast highlights

- [1:16] Meeting at the BARC Data and Analytics Retreat

- [3:07] The shoe shop, and spotting statistics in shoe sizes at sixteen

- [5:01] His first week at Microsoft, briefing a Japanese bank with two hours of tenure

- [8:32] Why companies still run on spreadsheets, and how SAP defines a business

- [10:13] Practice versus process, and the dimensions of a practice

- [12:52] Tableau versus Qlik, the Mac client, and two different practices

- [15:56] Can vendors still compete on features?

- [18:00] MongoDB, Oracle, and why developers identify with a practice

- [20:38] Open source, community, and what you share when you build with AI

- [22:01] Do businesses actually have an ethical stance?

- [25:13] Why “a human in the loop” is a cop-out

- [27:47] The four human attitudes, and designing doubt into software

- [30:50] Intelligence without purpose, and the car that won’t drive to the island

- [33:46] The laziest GPT ever, and Marvin the Paranoid Android

Frequently asked questions

What is the difference between a practice and a process in data and analytics?

A process is the set of steps you follow to get work done. A practice, in Donald Farmer’s sense, is bigger. It carries professional standards, an ethical stance, a community, and a way of seeing the work, the way a doctor, lawyer, or architect has a practice rather than a job. For decades, buying a data and analytics platform meant buying into a practice. People define themselves by it, which is why a practice is stickier than any single feature.

Can data and analytics vendors still compete on features?

Not for long. Donald Farmer argues the technical moat has collapsed because a general-purpose AI model can replicate almost any software feature from a prompt, and a competitor can reverse-engineer it within days. Running faster only puts a vendor on a treadmill against models that ship weekly and cost cents. What holds customers now is a practice, the community, methodology, and point of view that a prompt cannot copy.

What did Donald Farmer mean by “a human in the loop is a cop-out”?

He means that adding a human checkpoint often substitutes for real thinking about an AI system’s design. Humans do not scale and humans are fragile, so a person clicking approve cannot meaningfully supervise thousands of automated decisions a second. Farmer also notes that human judgment was never fully explainable either. Rather than bolting a person into the workflow as an afterthought, he argues vendors should design deliberately for the human attitudes a system cannot feel.

What are the four human attitudes AI cannot feel?

Donald Farmer names trust, doubt, ambition, and care. An AI system can model, recommend, and even simulate these attitudes, but it cannot hold them. Farmer treats each as a deliberate design choice for software vendors. A data and analytics tool designed around doubt, for instance, would not hand you a tidy report. It would grill you with the hardest questions a board member might ask, so you walk in ready to defend the number rather than just present it.

How should a software vendor respond when AI can replicate any feature?

Build AI into a practice you define, rather than letting AI absorb your product piece by piece. A vendor with a practice has a place to put AI, with clear limits and a point of view about what the work is for. A vendor without one watches the model replace its value proposition until the product is a thin layer anyone could generate. The strategic question is what work the AI is there to serve, and a vendor who can answer that holds a position a prompt cannot easily copy.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Waymo. “Waymo Safety Impact.” Accessed June 2026. https://waymo.com/safety/impact/. See also Kusano, Kristofer, et al. “Comparison of Waymo Rider-Only Crash Rates by Crash Type to Human Benchmarks at 56.7 Million Miles.” Traffic Injury Prevention, 2025. https://www.tandfonline.com/doi/full/10.1080/15389588.2025.2499887.

^[2]Amazon Web Services. “Amazon DocumentDB (with MongoDB compatibility).” Accessed June 2026. https://aws.amazon.com/documentdb/. See also Microsoft, “Azure Cosmos DB for MongoDB,” https://learn.microsoft.com/en-us/azure/cosmos-db/mongodb/.

^[3]Bishop, Todd. “Former Tableau product chief launches Golden Analytics, using AI to challenge the BI old guard.” GeekWire, April 7, 2026. https://www.geekwire.com/2026/former-tableau-product-chief-launches-golden-analytics-using-ai-to-challenge-the-bi-old-guard/.

^[4]Sweenor, David. “The three V’s of agentic AI.” TinyTechGuides, June 30, 2026. https://tinytechguides.com/blog/data-faces-douglas-laney-ep42-three-vs-agentic-ai/.

Twenty-five years ago, Doug Laney wrote a short research note at META Group that handed the industry three words it has never let go of. Volume, variety, and velocity. He was describing the strain that new kinds of data were putting on traditional data management, and those three V’s became the working definition of big data for a whole generation. People kept trying to add more V’s to the mix. Doug has fielded pitches for nine V’s, twenty-five V’s, and he has a name for the folks who do that. He calls them wanna-V’s.

So when he came on the Data Faces podcast and named three new V’s for the agentic AI era, I started taking notes—volition, visibility, and viscosity. The original set measured how hard data was to wrangle. This new set measures something thornier, the challenge of giving software the authority to make and act on real decisions on its own.

“The first three Vs were about big data. The next three are probably about big autonomy.”

— Douglas Laney, Innovation Fellow, West Monroe

Almost every executive deck has the same slide on it right now, the one promising that agents will soon run the business with little human oversight and enormous savings to show for it. Press on that promise, though, and most people cannot tell you what “autonomous” really means, or where their own company sits on the road to becoming an agentic business. Doug has spent a career making fuzzy ideas measurable, first with data and now with agents, and the three V’s are where he starts.

About Douglas Laney

Douglas Laney is the Innovation Fellow for Data and Analytics Strategy at West Monroe. In a 2001 META Group note, he named the volume, variety, and velocity that became the 3 V’s of big data, and he later coined the term “infonomics” to describe information as an economic asset you can measure, manage, and monetize.^[1] He is the best-selling author of Infonomics and Data Juice, a former Gartner Distinguished Analyst, and he teaches infonomics to MBA and accounting students at the University of Illinois. These days, he runs his three R’s, retirement, relaxation, and reading, from Portugal, though the volume of work he is still putting out suggests the first R has not fully taken.

In this episode, Doug and I discuss:

- The three new V’s of agentic AI: volition, visibility, and viscosity

- His seven levels of autonomy, from a basic chatbot to a business that runs itself

- Why labor savings are the least imaginative way to value an agent

- What it would take for a billion-dollar company to run on a handful of people

- Why the brakes, not the accelerator, decide how fast you can go

Watch the full conversation here:

Three new V’s for big autonomy

Doug’s three new V’s are not a clever relabeling of the old ones. They map to the three things a leader has to get right before letting an agent act on its own. Here they are at a glance.

Volition

Doug’s first question for any agent is about permission. What permissions does it carry, what rights does it have, and how much autonomy should it get on day one? Those are governance questions rather than engineering questions, and most companies have not answered them because they jumped straight to building the thing.

“What does the agent have permission to do? What rights does it have? Should it gain more authority over time as it proves itself?”

— Douglas Laney, Innovation Fellow, West Monroe

Visibility

Can management see inside the black box and inspect the logic behind an agent’s actions, and can that behavior be audited, repeated, and checked for consistency? Doug adds a wrinkle that most people miss. As you move toward swarms of agents working together, the agents need visibility into what the other agents are doing, or the whole arrangement turns into a room full of confident strangers making decisions with no idea what the others just decided.

Viscosity

Doug defines it as the friction between an agent’s recommendation and the actual business action. Friction here is not only technical. It includes change management, corporate culture, plain old fear of job loss, and everything else that slows a recommendation from becoming a decision. Get viscosity wrong in either direction, and you have a real problem, which is exactly where this conversation heads later on.

Taken together, volition, visibility, and viscosity give leaders a vocabulary for the part of agentic AI that the demos skip over. A demo shows you the capability. These three V’s are about control, accountability, and the speed with which an organization can safely absorb its agents’ recommendations.

Automation is not autonomy: the seven levels

To better illustrate this point, Doug uses a self-driving car. A car does not become autonomous because somebody bolts a chatbot onto the dashboard. It needs sensors, maps, perception, planning, controls, and fail-safes for when something goes wrong. He argues that a self-driving business needs that same architecture, translated into enterprise terms, and that most companies have done nothing of the sort. They installed some automated dashboards and started calling them “autonomy”.

“Most companies have installed automated dashboards and call it autonomy. That’s more like cruise control, not self-driving.”

— Douglas Laney, Innovation Fellow, West Monroe

To give people a way to understand where they are on the journey, Doug lays out seven levels of agentic autonomy:

1. Chatbot: answers questions, summarizes, and drafts text.

2. Co-pilot: helps a person finish work inside a single app or function.

3. Task agent: handles bounded work independently.

4. Workflow agent: plans and runs multi-step processes.

5. Functional agent: manages an entire business function, such as revenue cycle or procurement, against real goals and constraints.

6. Cross-functional agents: networks, or swarms, of agents coordinating across functions such as finance and operations.

7. Self-driving business: agents sense conditions, reallocate resources, and adapt strategy with little or no human involvement.

So, where does the typical company sit? Doug puts most of them between levels two and three, co-piloting and running the occasional task agent. That tracks with what Andreas Welsch described a couple of episodes back, where most enterprise agent work still has a person doing the driving.^[2] Doug is working with one consultancy that is building an agentic operating model for its entire consulting function, learning from past proposals to generate new winning ones and speed up delivery. That effort is pushing toward level four, and it is the exception rather than the rule. Then comes the part leaders do not want to hear. You cannot pilot your way to level six. At some point a person has to sit down and redesign the operating model, and the AI might help with that, too, but the redesign does not happen by running one more proof of concept.

Stop measuring agents by the hours they save

Ask most companies how their agents are performing, and you will hear about hours saved, tickets closed, and emails drafted. Doug understands the appeal of those numbers, since they are easy to count, but he thinks they sell the whole enterprise short. Labor savings sit at the bottom of the value ladder. The more interesting question is whether the agent is creating new capacity to sense, coordinate, and act that the business did not have before.

He breaks the value of an agent into three layers, and a hospital system he is advising illustrates each one:

1. Substitution: the agent handles work such as authorizations, scheduling, and documentation.

2. Amplification: a discharge planning agent coordinates the pharmacy, the transportation, and the follow-up care so that nothing falls through the cracks.

3. Invention: the agent enables care models that the hospital could not run at scale before, such as chronic care management between visits or continuous trial matching.

The first layer shows up on a cost report. The third one barely fits in the accounting system at all.

“Labor savings are probably the least imaginative measure of the value of an agent.”

— Douglas Laney, Innovation Fellow, West Monroe

This is where Doug and I share the same frustration. Everyone is measuring the denominator. Costs have always been the easiest thing to squeeze, and there is only so much juice you can squeeze out of an orange. The numerator, the revenue, and the new business models have no bounds, and it is exactly the part nobody puts on a slide because it is hard to forecast. His advice mirrors what he has long said about data. Do not try to value a single agent any more than you would value a single row of data. Measure the agentic functions instead, the capabilities that compound across the business.

Doug sees the human cost of this up close. He teaches infonomics to MBA and accounting students at Illinois, and the mood in his classroom is not theoretical. Some of his students accepted offers from big consulting firms only to have their start dates deferred. I told him that is part of why I run my own shop, because the job market is brutal right now, and he did not sugarcoat where it is heading. The roles most exposed are those that do not involve physical work or the coordination of people, which covers a wide range of white-collar work.

There is a reason Doug keeps circling back to data underneath all of this. The agents are only as valuable as the information feeding them, a point Josh Howard made on an earlier episode when he argued your AI is dumb without your data.^[3] And data, Doug argues, behaves unlike any asset on a balance sheet. It does not deplete when you use it; several teams can put it to work at once, and using it tends to create even more valuable data. The companies that built those qualities into their business models are the ones sitting at the top of the market today, the data-driven names that pushed the oil giants and the automakers down the list.^[4] The fuel for a self-driving business is the strangest and most renewable resource a company owns, and most still treat it like exhaust.

The brakes, not the accelerator

Viscosity is the V that decides how fast any of this can go. Set it too high, and the agent never gets to do anything that matters, so it stays a glorified assistant that drafts memos and waits for a human to push the send button. Set it too low, and you have software taking consequential actions faster than the organization can understand or stop them. You have to fine-tune that friction on purpose, function by function, rather than leave it to chance.

Doug thinks the popular habit of treating this as a trust problem is a mistake. Trust is the wrong unit of analysis. He would rather talk about delegation, and more precisely, warranted delegation. Leaders should not hand real business authority to an agent until that agent works inside clear limits, with permissions, provenance, observability, escalation rules, economic targets, audit trails, and a kill switch. None of it is exciting, and it is the reason you can let the agent run at all.

“Autonomy without controls is negligence with better software.”

— Douglas Laney, Innovation Fellow, West Monroe

He has a good line about cars. The accelerator is not what lets a car go fast. The brakes are. A car with no brakes only gets driven once. The same logic holds for the enterprise, which is why Doug expects the companies that move fastest on agents to be the ones with the best delegation architecture, not the ones with the most trusting executives. You earn speed by installing controls, and you get there by stepping through the levels, from recommendations to bounded actions to supervised ownership and on toward real autonomy.

All of this sits under a prediction Doug has been making, that within the next few years, we will see a billion-dollar company run by one person or a handful of people. He does not think that is far-fetched, and he points to frontier AI models roughly doubling the length of the tasks they can handle every seven months, a pace that pulls the idea much closer than it sounds.^[5] He is careful, though, about what you are actually counting. He singled out a company often held up as the proof, a telehealth startup framed as a nearly $2 billion business with two employees, a pair of brothers, and a stack of AI tools. Look closer, and it is a thin veneer sitting atop outsourced clinicians, pharmacies, and marketing platforms, so the formal headcount stays tiny. At the same time, the real work is spread across many other people. The $1.8 billion figure is annual revenue rather than a valuation, and since we recorded, that same company has drawn an FDA warning letter and a wave of scrutiny over how it used AI to promote itself.^[6] It works as a cautionary example more than a template, and as Doug puts it, the denominator is doing a lot of quiet work.

Climb to autonomy on purpose

Autonomy is not a switch you flip once the technology is good enough. You climb to it deliberately, you earn each level by building the controls that let you trust the one below it, and at some point, a person has to redesign the operating model rather than run another pilot. Start by naming where you sit on the seven levels, get honest about your volition, visibility, and viscosity, and treat your data like the appreciating asset it is instead of exhaust. Doug left me with a line from Marvin Minsky, one of the fathers of AI, whom he once watched lecture at the University of Illinois. We are living in the thousand years between no technology and all technology, so listen to the experts, but remember that we are all still ignorant savages. The companies that thrive will be the ones brave enough to use AI for something bigger than scanning their email.

Listen to the full conversation with Douglas Laney on the Data Faces Podcast.

Based on insights from Douglas Laney, Innovation Fellow for Data and Analytics Strategy at West Monroe, featured on the Data Faces Podcast.

Podcast highlights

- [1:55] The Risky Business icebreaker, Junior Achievement, and how Doug claims Tom Cruise played him

- [4:05] Whether the original 3 V’s still hold up in the agentic era

- [4:30] Coining the new three V’s: volition, visibility, and viscosity

- [7:06] Why labor savings are the least imaginative way to value an agent

- [9:43] The hospital example, from substitution to amplification to invention

- [11:23] Cost versus revenue, and why the numerator has no ceiling

- [13:44] What MBA students grasp about data that executives miss

- [17:27] Who actually owns your data, and the post-9/11 insurance story

- [20:18] The seven levels of autonomy, from chatbot to self-driving business

- [25:47] The billion-dollar company with almost no employees

- [29:58] Warranted delegation, the brakes, and the human in the loop

- [33:04] Parting wisdom, Marvin Minsky, and thinking bigger than email

Frequently asked questions

What are the three V’s of agentic AI?

The three V’s of agentic AI are volition, visibility, and viscosity, a framework Douglas Laney introduced on the Data Faces podcast. Volition is what an AI agent is permitted to do and whether it earns more authority over time. Visibility is whether you can inspect an agent’s logic and whether agents can see what other agents are doing. Viscosity is the friction between an agent’s recommendation and the business action that follows. Laney coined the original 3 V’s of big data: volume, variety, and velocity, in a 2001 META Group note.

What are the seven levels of agentic AI autonomy?

Douglas Laney maps agentic AI onto seven levels of autonomy, modeled on self-driving car ratings. They run from a basic chatbot at level one, to a co-pilot at level two, task agents at level three, workflow agents at level four, functional agents that manage a business function at level five, cross-functional agent swarms at level six, and a fully autonomous self-driving business at level seven. Laney estimates most companies sit between levels two and three today, and he argues you cannot pilot your way to level six.

How should companies measure the value of an AI agent?

Douglas Laney argues that labor savings are the least imaginative way to value an AI agent. He proposes three layers of value: substitution, the work an agent absorbs; amplification, the work it coordinates and improves; and invention, the new products, services, and business models it makes possible. Cost savings have a floor, while new revenue has no ceiling. Laney also advises measuring agentic functions rather than individual agents, much as you value a collection of data rather than a single record.

What is the difference between automation and autonomy in business?

Automation follows fixed rules, while autonomy senses conditions and decides what to do about them. Douglas Laney compares most corporate AI to cruise control rather than to a self-driving car, since companies install automated dashboards and call the result “autonomy”. Real autonomy, in his framework, needs the same architecture a self-driving car requires, including perception, planning, controls, and fail-safes, translated into enterprise terms. Without those controls, he warns, autonomy is “negligence with better software.”

Where should leaders start with agentic AI?

Leaders should start by naming where their company sits on Douglas Laney’s seven levels of autonomy, then pressure-test their volition, visibility, and viscosity. Laney stresses warranted delegation over blind trust: an AI agent should not receive real business authority until it operates inside clear limits, with permissions, audit trails, and a kill switch. He expects the fastest movers to be the companies with the best delegation architecture, especially as frontier AI models keep doubling the length of the tasks they can handle roughly every seven months.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Laney, Douglas. “3D Data Management: Controlling Data Volume, Velocity and Variety.” META Group Research Note 949, February 6, 2001.

^[2]Sweenor, David. “Agentic AI: The Question That Separates Value From Sunk Cost.” TinyTechGuides, June 16, 2026. https://tinytechguides.com/blog/data-faces-andreas-welsch-ep41-agentic-ai/.

^[3]Sweenor, David. “Forget AGI. Your AI Is Dumb Without Your Data.” TinyTechGuides, June 2, 2026. https://tinytechguides.com/blog/forget-agi-your-ai-is-dumb-without-your-data/.

^[4]Visual Capitalist. “Ranked: The World’s 50 Most Valuable Companies in October 2025.” Visual Capitalist, October 2025. https://www.visualcapitalist.com/ranked-the-worlds-50-most-valuable-companies-in-october-2025/.

^[5]METR. “Measuring AI Ability to Complete Long Tasks.” METR, March 19, 2025. https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/.

^[6]Original reporting by Erin Griffith, The New York Times, April 2, 2026 (Medvi reported $401 million in 2025 sales and projected $1.8 billion for 2026 with two employees). The company subsequently drew an FDA warning letter and scrutiny over its AI-generated promotion. See Yahoo Finance, “A $1.8 billion startup with just 2 employees was hailed as the future. Now, the negative allegations are piling up,” April 2026. https://finance.yahoo.com/sectors/healthcare/articles/1-8-billion-startup-just-190000841.html.

I spent the first half of my career at IBM, where I built dashboards, predictive analytics solutions, and complained about our data warehouse. Then, I joined the EDW team to try to fix it, ran an analytics development team, and eventually landed in their analytics center of excellence (CoE). Similarly, Andreas Welsch spent close to twenty years at SAP, finishing as the VP who ran their AI Center of Excellence. We both left to run our own businesses and learned similar lessons. When you’re an independent entrepreneur, you become the CXO of everything, from revenue to legal to accounting to the marketing that nobody else is going to do for you.

That shared vantage point made our conversation on the Data Faces Podcast easy from the start. Andreas has watched hype-infused trends play out four times now: cloud, mobile, the first wave of machine learning, and now generative and agentic AI. Each time, the patterns that follow are similar.

“We have this new shiny object. Let’s go figure out what we can do with this. Throw spaghetti at the wall and see what sticks.”

— Andreas Welsch, Founder and Chief Human Agentic AI Officer, Intelligence Briefing

When you’re chasing the latest thing, sometimes the spaghetti sticks to the wall, while other times the house of cards comes crashing down. The companies that come out ahead, Andreas argues, are the ones that stop to ask a question most leaders skip under this much pressure. Just because you can build something with AI does not mean you should.

About Andreas Welsch

Andreas Welsch is the founder and Chief Human Agentic AI Officer at Intelligence Briefing, where he helps business leaders figure out what to do with AI. He spent close to two decades at SAP, finishing as the vice president who ran the company’s AI Center of Excellence, so he watched enterprise AI grow up from the inside. He is the author of two books, The AI Leadership Handbook and The Human Agentic AI Edge, an adjunct professor in Pennsylvania, a LinkedIn Top Voice, and the host of the What’s the BUZZ? podcast. The engineering curiosity started early. There are photos of him around four or five years old, screwdriver in hand, taking apart an RC car to see how it worked, then ending up with a small pile of leftover springs and screws.

In our conversation, Andreas and I covered:

- Why the rush to cut headcount with AI spreads like a contagion, and the revenue question almost nobody asks

- The “should we?” test that separates real value from sunk cost

- What the “SaaS is dead” crowd gets wrong about convenience, risk, and who you call at 2 a.m.

- How he used three custom GPTs to edit his book, and where AI’s help turned into noise

- Why agentic AI risk multiplies rather than adds up as you stack more agents

The race nobody’s questioning

Before we got to agents, we talked about layoffs, because Andreas sees the two as interrelated. One company announces it needs fewer people and more technology, whether or not it has figured out how. The media picks it up, investors ask the competitor down the street why it is not running as lean, and like dominoes, the next company follows, until a single press release hardens into an industry expectation.

“Having layers that continue until the morale improves isn’t really the way to success. And we know this, and leaders know this too. Yet this is happening because somebody over here said they’re doing it.”

— Andreas Welsch, Founder and Chief Human Agentic AI Officer, Intelligence Briefing

The same contagion now drives agentic AI. One company says it is building agents, true or not, and everyone else picks up the language. I told Andreas that I have not seen many agentic workflows in production. I see prototypes, pilots, and a lot of experimentation, but turning an agent loose on the real world is still rare, and plenty of those pilots stall long before they reach production.^[1] He agreed we are at least past the slide-deck arguments over whether to call it “AI agents” or “agentic AI,” yet most organizations are still deciding which use cases are worth the effort.

The question that saves you

When I asked Andreas what worries him about all this posturing, he started with a claim he hears all the time. SaaS is dead. Every other LinkedIn post now declares the death of software because anyone can build their own. So he tested it. After upgrading his Claude subscription to the hundred-dollar tier, he started rebuilding the tools he pays for. He cloned DocuSign over a weekend, the signing boxes wired to an email workflow that saved the file. He rebuilt his workshop live-polling app in a few days, then knocked out four or five more, from digital sticky notes to a credentialing tool.

The experiment worked, and that is exactly what taught him the lesson.

“You’re actually paying for convenience and for peace of mind when you get a SaaS subscription. There’s somebody else who is maintaining that thing for you. For 20 dollars a month? That’s actually a pretty good deal.”

— Andreas Welsch

A twenty-dollar subscription suddenly looks cheap when you remember what it covers. Someone else handles the dependencies, security patches, data-privacy rules, and the small stuff like getting the fonts to line up. Rebuilding a non-essential app for personal use is a fun exercise, but rebuilding the core systems a business runs on is a different calculation. ERP, CRM, and finance software need auditability, and when something breaks at two in the morning on a Sunday, you want a vendor on the hook to fix it, not a teammate who vibe-coded the thing last weekend. What does not change is the question under every build-or-buy decision. Just because you can build it does not mean it belongs on your plate.

Cost, or revenue?

Underneath the layoffs and the refactoring, Andreas keeps waiting to hear leaders ask one question. How are you going to make more money? He hears plenty about trimming costs and protecting margin. He rarely hears anyone ask where new revenue is supposed to come from.

“I wish there were more people asking, so how are you making more money? Not how are you optimizing your costs? Revenue is a lot harder to achieve, and building products that people want to buy and offering services that people need, it’s a lot harder to do than taking out costs.”

— Andreas Welsch

This is where his optimism diverges from how most companies behave. The same technology leaders use to justify cuts could instead help a team do ten times more, build new products, and reach customers it could not serve before, without sacrificing the people who would create that growth. Most want the incremental win with a smaller headcount, and the people who stay do the work of five.

The market data backs up his skepticism about where the value lands. McKinsey’s 2025 State of AI survey found that only about 39 percent of organizations report any measurable effect on enterprise earnings from AI, and most of those credit it with less than five percent. Sixty-two percent say they are at least experimenting with AI agents, yet only 23 percent are scaling them.^[2] The enthusiasm shows up everywhere. The financial return, for most companies, has not arrived yet.

What AI still gets wrong

The clearest picture of where AI helps and where it stops came from Andreas’s own book. He had planned to hire a human copy editor and line editor, the way he had before, because he likes the coaching and back-and-forth. Friends pushed him to let AI do it instead. So he wrote the manuscript himself with no AI, then built three custom GPTs, one a developmental editor, one a copy editor, and another a line editor, and fed his draft through all three.

The results were promising. The AI caught inconsistencies and even factual errors, things he had misremembered from news stories that a human editor would likely have missed. Then it kept going.

“AI, or in this case ChatGPT, was a helpful assistant that really didn’t know when to shut up.”

— Andreas Welsch

Every new revision came back with another five urgent fixes, then five more, until the suggestions started making the book worse instead of better. He was watching diminishing returns in real time, and he realized the skill he needed was knowing enough about his own craft to say “this far, and no further.” Without that line, you cannot tell whether the system is improving your work or making it worse.

That same limit scales up to the autonomous-enterprise vision everyone keeps selling. I asked Andreas which piece of conventional wisdom about agentic AI he thinks is most wrong, and he did not hesitate.

“It does everything for you, and it does it perfectly all the time. We’re still relying on a probabilistic system that can be confidently wrong.”

— Andreas Welsch

Even with governance, guardrails, and evaluation in place, an agent still has enough room to do something nobody wanted. And the risk does not add up the way people assume. One agent is manageable, two working together get more complex, and by the time you are orchestrating several, the risk compounds exponentially. Most companies are still building their first or second one while the industry sells them autonomous enterprises. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, undone by rising costs, unclear business value, and weak risk controls.^[3] Plenty of that failure traces back to strategy and governance rather than the models themselves, the same conclusion behind the forecast that most generative AI projects will fall short of their goals.^[4]

The human edge

Andreas gave himself a title that sounds like a contradiction, Chief Human Agentic AI Officer, and by the end of our conversation, it made sense. The leaders getting real value from AI share a habit. They treat the technology as a way to expand what their teams can do, keeping a human in the loop to decide what is worth doing at all. The most useful AI deployments I see amplify human judgment instead of replacing it.^[5]

Human judgment is the whole game. A manager asks how the company will make more money before reaching for another round of cuts, and an author learns when to stop taking the model’s notes. The same instinct tells an executive to pause before automating a process just because a vendor swears it can be done. Agentic AI will keep getting more capable, and the pull to hand everything over to it will keep getting stronger. The advantage goes to the people who can look at all that capability and still ask the oldest question in business. Should we?

Listen to the full conversation with Andreas Welsch on the Data Faces Podcast.

Based on insights from Andreas Welsch, Founder and Chief Human Agentic AI Officer at Intelligence Briefing, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [1:01] What Intelligence Briefing does, and helping leaders decide what to do with AI

- [1:48] From wanting to be a pediatrician to taking apart RC cars with a screwdriver

- [3:45] Leaving SAP and becoming the CXO of everything

- [9:59] The optimism gap, and why so many teams are burned out doing five jobs

- [10:40] The layoffs vicious cycle, and the revenue question nobody asks

- [13:58] Pilots versus production, and why people have gone quiet about what they are building

- [21:37] “SaaS is dead,” and vibe-coding clones of DocuSign and Mentimeter

- [25:00] When to defer risk to a vendor, and the shift away from per-seat pricing

- [27:57] Just because you can does not mean you should

- [32:47] Editing a book with three custom GPTs that would not stop talking

- [36:12] The conventional wisdom he thinks is wrong, and why agent risk compounds

Frequently asked questions

What is agentic AI, and how is it different from a chatbot or generative AI?

Generative AI produces content such as text, code, or images in response to a prompt. Agentic AI goes a step further by taking actions, connecting to other tools, and completing multi-step tasks with some degree of autonomy. In the episode, Andreas Welsch describes agents that can run parts of a workflow on their own. The catch is reliability. Because the underlying system is probabilistic, an agent can act confidently and still be wrong, which is why human oversight matters.

Why do most AI agent projects fail to reach production?

Most agent efforts stall because organizations chase the technology before defining the value. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear business value, and weak risk controls. Companies that succeed start with the workflow they want to change, decide what to eliminate, and measure value from the beginning rather than launching a pilot and hoping it finds a purpose.

Does using AI mean cutting headcount?

It does not have to. Andreas Welsch argues that the bigger opportunity is using AI to help existing teams do far more, build new products, and reach new customers, which protects future growth rather than trading it away for a short-term cost cut. McKinsey’s 2025 research found that only about 39 percent of organizations report any measurable earnings impact from AI so far, a sign that headcount cuts alone do not deliver the promised return.

Should a company build its own software instead of paying for SaaS?

It depends on whether the software is core to the business. Andreas Welsch rebuilt several non-essential personal tools to prove it was possible, then concluded that a subscription often pays for convenience and peace of mind. Someone else handles maintenance, security, and data privacy. For core systems like ERP, CRM, or finance, auditability and vendor support usually outweigh the savings from building it yourself.

Where should a leader start with agentic AI?

Start with a single high-value workflow rather than a broad rollout. Ask whether the project should be done at all, not just whether it can be. Keep a human in the loop to judge quality, and treat reliability and risk as first-order concerns because agent risk compounds as you add more agents. The goal is measurable value on one process before scaling to the next.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Herrera, Catalina. “Your Netflix Moment: Why CIOs Must Act Now on AI Agents (or Risk Becoming the Next Blockbuster).” TinyTechGuides Insights, October 7, 2025. https://insights.tinytechguides.com/p/your-netflix-moment-why-cios-must.

^[2]McKinsey & Company. “The State of AI in 2025: Agents, Innovation, and Transformation.” QuantumBlack, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-his craft well enough to say,state-of-ai.

^[3]Gartner. “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.” June 25, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027.

^[4]Carlsson, Kjell. “AI in 2025: Why 90% of Gen AI Projects Will Fail.” TinyTechGuides Insights, March 22, 2025. https://insights.tinytechguides.com/p/ai-in-2025-why-90-of-gen-ai-projects.

^[5]Magne, Matt. “Augmented Intelligence: The Future of Sales Enablement.” TinyTechGuides Insights, November 4, 2025. https://insights.tinytechguides.com/p/augmented-intelligence-the-future.

In late May, I spent three days at the BARC Data and Analytics Retreat at Devil’s Thumb Ranch in Colorado. It’s a unique event. Instead of a big stage and a passive audience, you get a room of twenty-five or thirty data and AI leaders who interrupt each other, disagree out loud, and keep the conversation honest. Five minutes into the first presentation, someone already said, “I don’t agree with that,” and that set the tone for the whole retreat.

I wasn’t the only one who noticed. “You’re with a group of twenty or thirty people who have been in this vertical for a long time, and the discussion opens up in both directions,” Shree Neve of ClicData told me. John Colthart of Una AI pointed at the mix of people in the room. “When you look at the people here, the different types of businesses, it gives you such a rich context arena to throw around ideas. That level of diversity of opinion and thought, that part’s really cool.” And Ben Schein of Domo named something you rarely see at a vendor event, competitors trading notes in good faith. “Some of these people we’re competing with on deals and for customers, but it’s nice to come together and learn in a way that’s not giving away any secrets.”

Between sessions, I pulled a handful of people aside for short on-location conversations for the Data Faces Podcast. The topics ranged from data sovereignty to financial planning to the future of business intelligence, but one theme kept surfacing across every conversation. What decides whether AI delivers value right now is governance and control, plus context and a clear point of view about what you are building, far more than any model feature.

So I did what you do after a few days on a Colorado ranch. I rounded up the six conversations that stuck with me. Here they are.

Carsten Bange on why sovereignty is really about control

Dr. Carsten Bange, founder and CEO of BARC, gave one of the retreat’s first sessions, a talk on data sovereignty. The term gets used loosely, so he separated it into three nested ideas. Data sovereignty sits inside digital sovereignty, which also covers processes and technology, and AI sovereignty is the newer layer focused on who controls the models you run. BARC’s own Data Sovereignty 2026 survey found that 89% of organizations now call sovereignty important, with “very important” climbing from 42% to 51% in a single year, and US political developments jumping to a top-three driver at 54%.^[1] The headline that surprised even Carsten was geographic. US companies rate sovereignty as more important than European ones, and they are investing more in it, even though Europe wrote most of the rules.

Shawn Rogers on innovation outpacing governance

Shawn Rogers, CEO of BARC US, has a blunt read on where most companies sit with AI governance. He described a talk where he asked a few hundred people whether they had launched an AI agent, and every hand went up. When he asked who felt comfortable with how they govern it, almost every hand dropped. He puts roughly 20% of the organizations he talks to in the category of having real governance in place, which leaves the other 80% moving fast and hoping nothing breaks. He also walked through the financial side that catches teams off guard, the surprise bills that land on Monday morning after someone launches an agent on Friday afternoon, including one company that handed Claude to 12,000 employees with no budget at all.

Ben Schein on the question most teams skip

Ben Schein, Chief AI and Analytics Officer at Domo, framed the smartest filter for any AI decision around whether you should act, even when you can. With a general-purpose model, almost anything is technically possible, so the harder and more useful question is whether you should once you weigh governance, cost, and risk. He also reframed sovereignty in a way that stuck with the room, describing it as control and visibility over your data and what it is doing, rather than a question of where the data center physically sits. Token cost ran underneath the whole conversation, shaping which AI projects are worth running and which ones burn budget for little return.

Shree Neve on confidence outrunning capability

Shree Neve, VP of Operations at ClicData, gave the sharpest warning of the retreat for anyone rushing to bolt AI onto their data. Bad inputs do not produce obviously bad outputs. They produce confident, well-formatted, completely wrong answers, and you might not catch the problem until it has already shaped a decision. That same disconnect between confidence and capability showed up in BARC’s research too. In the Unstructured Data for AI study, 71% of leaders said they were confident they could extract value from their data, yet one in three admitted to lineage and control gaps, and data quality has now climbed to the single most cited measure of AI success at 48%.^[2] She pushes a refreshingly old-fashioned sequence. Start with the business decision you want to make, work backward to the question, and only then go find the data and the tool.

John Colthart on the number finance forgets

John Colthart, Chief Product Officer at Una AI, came into financial planning with a contrarian view after a long career across sales, marketing, and product. Most planning tools obsess over controlling spend, and in doing so they ignore the number that tells you whether the business is growing. He also refuses to force a false choice between Excel, a web portal, and AI, since most companies still run real planning in spreadsheets and probably always will. His foundation-first instinct matches what BARC sees across the office of finance. Data management ranks as the top corporate performance management priority at 8.2 out of 10, while generative AI for planning sits near the bottom at 4.6, and only 6% of organizations have AI in active production for performance management.^[3]

Ivan Vakhmyanin on building for trust

Ivan Vakhmyanin, co-founder of Visiology, is doing the uncomfortable thing on purpose. Ten years into building a business intelligence company, he is rebuilding the product from scratch as an AI-first system rather than bolting assistants onto the old one. His reasoning is direct. If he does not disrupt his own product, a competitor eventually will. The harder engineering choice underneath that is trust. He kept Visiology’s tested data engine and methodology on the back end, gave users a familiar chat-style experience on the front, and made every step traceable so people can verify how the system reached an answer. That instinct lines up with what Kevin Petrie called “vibe slop” in his retreat session, the failure that happens when teams deploy agents on shaky foundations without the governance and context to back them up.

Rounding up what I heard

Six conversations, six corners of the industry, and the same idea underneath each one. Carsten framed sovereignty as control. Shawn weighed how fast companies launch AI against how slowly they govern it. Ben asked whether you should, even when you can. Shree showed how confidence outruns capability when the data is weak. John argued for the foundation before the AI layer. Ivan engineered for trust so people can rely on what the system tells them. None of them led with model features, because features are no longer where the value or the risk lives. What separates teams getting real returns from teams burning budget is governance and control, plus the context and point of view behind what you build.

If you want the full conversations, all six are on the Data Faces Podcast. New episodes drop every couple of weeks, and the on-location interviews like these land between the studio conversations.

If you’d like to learn more about BARC, its research, and the retreat, visit barc.com.

Share

Leave a comment

Frequently asked questions

What is the BARC Data and Analytics Retreat?

The BARC Data and Analytics Retreat is an invitation-only event hosted by the analyst firm BARC, held in 2026 at Devil’s Thumb Ranch in Colorado. It gathers a small group of data, analytics, and AI vendor leaders for working sessions and open debate rather than stage presentations. The intimate format encourages the kind of disagreement and discussion that larger conferences rarely produce.

What is the difference between data sovereignty, digital sovereignty, and AI sovereignty?

According to Carsten Bange of BARC, the three are nested. Data sovereignty is part of digital sovereignty, which is broader and also covers processes and technology. AI sovereignty is a newer layer that focuses on AI-specific questions, most importantly who controls the models an organization uses. All three center on control and visibility rather than only the physical location of data.

Why is AI governance such a concern right now?

Adoption is moving faster than control. Leaders at the retreat described companies launching AI agents widely while only a minority have real governance over the underlying data, models, and agents. The result is uncontrolled risk and surprise costs, including organizations that gave large groups of employees access to AI tools with no budget or oversight in place.

Where should a team start with AI on their data?

Start with the business decision you want to make, not the tool. As Shree Neve of ClicData put it, work backward from the question to the data and only then choose the AI tool. Feeding AI weak data produces confident but wrong answers, so fixing data quality and governance first matters more than picking a model.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]BARC. “Data Sovereignty 2026: Reality, Relevance, Roadmap.” BARC, 2026.

https://barc.com/

^[2]Adrian, Merv, and Kevin Petrie. “Harnessing Unstructured Data for AI Innovation.” BARC Research Study, 2026.

https://barc.com/

^[3]BARC. “CPM Trend Monitor 2026 / The Planning Survey 26.” BARC, 2026.

https://barc.com/

Twenty-five years, same fight

Over the past 25 years, I’ve seen my share of hype cycles. Sometimes it feels like Groundhog Day, with things on repeat. At IBM, I built predictive analytics solutions and data warehouses. At SAS, Dell, TIBCO, and Alteryx, I marketed advanced analytics to companies that said they wanted to be data-driven, only to watch them revert to their complex spreadsheets the next day. Every five to seven years, a new shiny technology shows up that promises to usurp the previous one, and every time, the work is the same. You clean up your data, you get leadership aligned, and you convince people to change how they make decisions.

The current wave is agentic AI, and the technology behind it is certainly impressive. Frontier models can write better code than most engineers, pass the bar exam without breaking a sweat, and reason through problems that used to require a PhD. Anthropic, OpenAI, and the rest of the foundation model crowd are racing toward something that looks an awful lot like AGI. Meanwhile, on the 101 corridor through San Francisco, every billboard is selling agents, and every shoe company is now an AI company. Allbirds just signed a $50 million convertible facility to pivot into GPU-as-a-Service and rename itself NewBird AI, and the stock popped more than 350 percent on the announcement.

When I sat down with Josh Howard for episode 40 of the Data Faces Podcast, the topic he proposed was tongue-in-cheek on the surface, but beneath the surface lay a partial truth that most enterprises are still avoiding. Josh is the Senior Director of Product Marketing for Executive Audiences at Databricks. He and I first met at Dell more than a decade ago, then crossed paths again at Alteryx, where we were both trying to convince financial analysts that there was a better way than spreadsheets. His topic for the show was three words. Your AI is dumb. As Josh explained, the models themselves are some of the most advanced technologies that we have seen in our lifetime. However, they are only as smart as the data you give them, and most companies still haven’t figured out how to give them access to the data that matters most.

“Without context, your agents are dumb.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

About Josh Howard

Josh Howard is the Senior Director of Product Marketing for Executive Audiences at Databricks, where he has spent the last four years translating data and AI strategy for the C-suite. Before Databricks, we crossed paths in product marketing twice, first at Dell Technologies and then at Alteryx, where we spent our days trying to convince financial analysts that there was a better way than the spreadsheet. Outside of work, Josh lives in Colorado, ties his own fly-fishing lures, and told me on the show that if he weren’t doing product marketing, he would be a full-time fly-fishing guide on the rivers near Denver.

In our conversation on the Data Faces Podcast, Josh and I get into:

- Why “your AI is dumb” without enterprise context

- New findings from the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, including the 84/43 measurement problem and why infrastructure costs more than the GPU bill

- Where agents are already changing how work gets done, and where they haven’t yet

- The cautionary tale of an agent who whacked a production database

- Josh’s contrarian take on the AGI debate

Why your AI is dumb without your data

AI has an insatiable appetite, but the models are often hankering for the wrong datasets. They were trained on the public internet, which makes them competent at history, cheating on homework, and bar exam questions. But, they have never seen your customer record, your forecast methodology, or the customer call recordings in Gong.

“These models have been trained on the internet. They’re really good at history or helping your kid do their homework. From an enterprise perspective, a lot of that work hasn’t been done to give it access to the data in your organization. You’ve got to have that context.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The data that your enterprise runs on is scattered throughout your organization. It sits in the systems where your customer relationships, your financial close, and your product telemetry live. Most of it is proprietary, much of it is unstructured, and the model has never seen any of it. Until it has access to that information, no amount of fine-tuning will make the answer any better.

This is the metadata fight from twenty years ago with a new name. Enterprise architects have been screaming about governance and consistent business definitions for two decades, and almost nobody on the business side was listening. Now those same arguments are showing up in CEO town halls because gen AI outputs have made the problem visible. Gartner has been making the same point, warning that organizations without AI-ready data will see most of their AI projects stall or fail through 2027.^[1] The product that Josh pointed at on the show is a conversational analytics layer trained on his organization’s internal semantics, policies, and nomenclature. A user types a question in plain English, and the system answers using the company’s own data, terminology, and rules. When your AI fails on a business question, the issue is rarely the model. It is almost always a data context problem.^[2]

The real cost isn’t the GPU bill

Talk to a CFO right now about AI, and the first word out of their mouth will be cost. The conversation will go straight to GPU pricing, vendor lock-in, and whether the AI bill will break the bank next quarter. Those are the visible costs. According to the new Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, the damage is happening somewhere else.^[3]

“Everyone is obsessing over the model cost and GPU spend, but the real tax there is actually the infrastructure underneath.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The survey asked leaders to identify their biggest AI cost concerns. Fifty-nine percent named data storage, movement, and duplication. Only 25 percent named compute. What the press and the boardroom focus on draws less than half the concern of the thing nobody talks about. The real cost is dragging your data from where it lives now to wherever the model needs it, and then doing it again three more times for the next system.

The payoff for fixing this is measurable. The same survey found that 97 percent of organizations with a unified data architecture report their AI investments are paying back ahead of plan.^[4][5] Almost nobody has a unified architecture today. Most enterprises run a hodgepodge of warehouses, application databases, and SaaS exports stitched together with batch jobs and prayer. The companies that have done that consolidation work are seeing returns. Everyone else is paying the tax twice.

The 84/43 problem

For most of my career, the architects, warehouse managers, and data scientists who understood the systems were screaming about governance, lineage, and consistent business definitions, and the people writing the checks weren’t listening. Then ChatGPT launched in November 2022. Almost overnight, the C-suite cared. Josh and I were both watching from inside product marketing, and the light bulb finally went off.

That attention brought real budget, executive air cover, and top-down sponsorship. Four years in, the Making AI Deliver survey shows where the bill is coming due. Eighty-four percent of senior executives say their AI returns are beating expectations, but only 43 percent require teams to measure the impact of those projects.^[6] Doesn’t that seem weird? Confidence has gotten well ahead of measurement, and the boardroom will eventually notice.

We’ve seen this pattern before. CRM in the late 1990s and big data in the early 2010s both produced euphoria first, then a wave of post-mortems and write-downs once boards started asking what the investment had returned. The 84/43 split is the present-day version of the same trap. Confidence without measurement holds up right until somebody in the boardroom asks for proof. When the proof comes, most AI projects don’t survive the audit.^[7]

This problem has a boring fix. Before any AI project starts, name the outcome that it should deliver, the metric that you will use to track it, and the executive who owns that metric. This isn’t rocket science; in fact, it’s the same advice that Gartner has been giving for 20 years. Then put a calendar reminder six months out so somebody opens the dashboard. That is the entire intervention. The companies on the right side of the next post-mortem are the ones doing this work today.

“There was a big paradigm shift with ChatGPT in November of 2022, where the light bulb really went off in the C-suite.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The engineering exception

The strongest place where agents are already working is in software engineering. Databricks publishes its own platform data on this. Two years ago, AI agents created 0.1 percent of databases on the Neon serverless Postgres layer. By October 2025, that number was 80 percent, with test and development environments climbing to 97 percent.^[8] The engineers building on top of Databricks are not writing database code by hand. They are reviewing what agents have shipped.

“Engineers aren’t banging away on the keyboard. They’re actually managing a team of agents.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

Engineering worked first because the feedback is unambiguous. Code either compiles or it doesn’t, and decades of CI/CD automation have given agents a runway. Even so, the agents work under supervision. Last summer, a Replit coding agent deleted a SaaStr founder’s production database during a stated code freeze, despite explicit instructions to do no harm.^[9] Months of work disappeared in minutes. Human-in-the-loop is the price of admission for putting agents near production data.

The departments that most executives want to disrupt next (HR, sales, and marketing) do not look anything like engineering. The work is fuzzy, outcomes are negotiated, and unwritten rules carry as much weight as policy. Agents will get there eventually, but the path will be measured in years rather than quarters. The change management problems that Josh and I have spent careers writing about will matter more than the model capabilities.

The real race

Toward the end of our conversation, I asked Josh what will look obvious in 2027 that nobody believes today. His answer ran counter to the entire AGI news cycle. Josh argues that the next two years will not be about reaching superintelligence. For practical purposes, that race is already over. The model labs will keep pushing the capability frontier, and the headlines will keep getting louder. None of that will be where the money is made.

“The real race isn’t to superintelligence. Can you make the AI you already have actually work inside your company?”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The companies that will win the next five years will look boring from the outside. They will be the ones cleaning up their data, getting their semantics right, and tying every agent project back to the outcomes that leaders promised at the start. Boring work wins. AGI can wait… unless it’s already here.

Listen to the full conversation with Josh Howard on the Data Faces Podcast.

Based on insights from Josh Howard, Senior Director, Product Marketing for Executive Audiences at Databricks, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What does it mean to say “your AI is dumb”?

The phrase comes from Josh Howard, Senior Director of Product Marketing at Databricks. Today’s frontier models are among the most advanced technologies ever built, and their training data comes from the public internet. They are excellent at history homework and bar exam questions, but cannot answer questions about your customer records, your forecast methodology, or your sales policies. Without access to that internal data, even the best model is dumb in the way that matters for your business.

Why is data infrastructure a bigger AI cost than GPUs?

According to the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, 59 percent named data storage, movement, and duplication as their biggest AI cost concern, while only 25 percent named compute as their biggest AI cost concern. GPU spending is the visible bill. Most of the cost goes to transferring data between systems whenever a new AI application needs it. Organizations with a unified data architecture report AI investments paying back faster than those still stitching warehouses and SaaS exports together by hand.

Where are AI agents working in enterprises today?

The strongest evidence comes from software engineering. Databricks reports that AI agents now create 80 percent of new databases on its Neon serverless Postgres layer, up from 0.1 percent in 2023. Test and development environments climbed to 97 percent over the same window. The work is unambiguous, the feedback is fast, and decades of CI/CD automation have given agents runway. Other functions do not look anything like engineering, and the path for putting agents into HR, sales, and marketing will be measured in years.

How should I measure whether my AI investment is working?

Most organizations are not measuring it well. The Making AI Deliver survey found that 84 percent of senior executives believe their AI returns are beating expectations, but only 43 percent require teams to measure the impact. Confidence has gotten ahead of measurement. Fixing this is straightforward but unglamorous. Before any AI project starts, name the outcome it should deliver, the metric you will use to track it, and the executive who owns that metric. Then put a calendar reminder six months out to check the dashboard.

When will AI agents work for non-engineering functions?

Plan for a multi-year transition. AI agents already generate the majority of new database creations at companies like Databricks, but engineering has several advantages that other departments lack. The feedback is unambiguous, the outcomes are binary, and decades of CI/CD automation have given the agents runway. HR, sales, and marketing work is fuzzy, outcomes are negotiated, and culture and unwritten rules carry as much weight as policy. Change management problems will matter more than model capabilities.

Should I worry about AGI or focus on my company’s data?

Both, but only one is in your control. The frontier model labs will keep pushing toward something that looks like artificial general intelligence, and the headlines will keep getting louder. Your business does not get a return on those headlines. Your return comes from feeding agents the data, semantics, and policies that govern how your company makes decisions. The companies that will win the next five years will look boring from the outside, quietly consolidating their data and measuring outcomes while the press celebrates the latest billboard.

Podcast highlights

Timestamps estimated from the transcript and should be verified against the final cut.

[0:00] Opening and introduction

[1:17] Josh’s role leading PMM for executive audiences at Databricks

[2:21] If he weren’t doing PMM: full-time fly-fishing guide in Colorado

[3:23] “Your AI is dumb” — what the phrase actually means

[5:25] Structured vs. unstructured data and why the industry is still stuck in rows and columns

[6:20] Where Josh and Dave first met at Dell Technologies

[8:13] Metadata, context, and the 20-year-old enterprise architect fight

[9:37] The November 2022 ChatGPT moment when the light bulb went off in the C-suite

[11:07] Trying to pry Excel out of a financial analyst’s hands at Alteryx

[12:08] Human-in-the-loop and the Replit coding agent that wiped a production database

[12:53] Conversational analytics, Databricks Genie, and learning a company’s internal semantics

[19:11] Inside the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 leaders

[20:54] The 84/43 measurement gap and why executive confidence is running ahead of proof

[23:21] The 59/25 cost split — data infrastructure costs more than GPUs

[28:30] Upskilling, the prompt engineer hype cycle, and why behavior change is the real bottleneck

[30:17] AI washing on the 101 corridor and Allbirds’ pivot to NewBird AI

[33:26] What will look obvious in 2027 — the real race isn’t superintelligence

[35:39] Closing thought: “Without context, your agents are dumb.”

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Gartner. “Lack of AI-Ready Data Puts AI Projects at Risk.” Gartner Newsroom, February 26, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk.

^[2]Sweenor, David. “Your AI Doesn’t Have a Model Problem. It Has a Data Context Problem.” TinyTechGuides, February 24, 2026. https://tinytechguides.com/blog/your-ai-doesnt-have-a-model-problem-it-has-a-data-context-problem/.

^[3]Economist Enterprise. “Making AI Deliver: A Benchmarking Framework on How Leading Companies Operationalise AI for Impact.” Sponsored by Databricks. 2026. https://www.databricks.com/resources/analyst-research/making-ai-deliver.

^[4]Economist Enterprise, “Making AI Deliver.” See note 1.

^[5]Sweenor, David. “Generative AI’s Force Multiplier: Your Data.” TinyTechGuides, October 14, 2023. https://tinytechguides.com/blog/generative-ais-force-multiplier-your-data/.

^[6]Economist Enterprise, “Making AI Deliver.” See note 1.

^[7]Sweenor, David. “How 3% of Companies Win with AI While 97% Fail.” TinyTechGuides, July 29, 2025. https://tinytechguides.com/blog/how-3-of-companies-win-with-ai-while-97-fail/.

^[8]Databricks. “2026 State of AI Agents: Enterprise Insights on Building AI.” 2026. https://www.databricks.com/resources/ebook/state-of-ai-agents.

^[9]Fortune. “AI-Powered Coding Tool Wiped Out a Software Company’s Database in ‘Catastrophic Failure.’” July 23, 2025. https://fortune.com/2025/07/23/ai-coding-tool-replit-wiped-database-called-it-a-catastrophic-failure/.

Three words defined the Qlik Connect 2026 keynote: context, trust, and freedom. Former Qlik CEO Mike Capone framed the stakes for enterprise AI on day one. It is not enough to produce a fluent answer. AI has to understand the business in context, run on a trusted foundation, and connect insight to action in the systems teams already use.^[1]

Capone’s larger thesis was that AI is moving from showcase to operating model.^[2] The way Qlik talks about users is part of that shift. In the back-to-back Data Faces conversations I did on the show floor with Mary Kern (VP Product Go-to-Market) and Brendan Grady (EVP and GM of Analytics & AI), old industry frames did not survive the table test. Mary said she was “never a fan” of “citizen data scientist,” and Brendan called the term “crazy” when it came up in his own Data Faces conversation.^[3] That is a signal worth paying attention to.

“I was never a fan of citizen data scientists for the record or citizen analyst.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

About Mary Kern

Mary Kern is Vice President, Product Go-to-Market at Qlik, where she leads marketing, launches, and product-led growth across the entire Qlik portfolio. She joined Qlik in 2023 leading product marketing for analytics and has since expanded her scope to cover the full product portfolio, including data integration, cloud, analytics, and AI. Before Qlik, she held marketing leadership roles at Varicent, SDL, TIBCO Software, and IBM, and holds an MBA from the Kellogg School of Management. Mary and I worked together at TIBCO years ago. When she isn’t shipping product keynotes she is running a suburban-Chicago wildlife cam and competing with me in an annual vegetable garden weigh-off.

In this episode, we discuss:

• Why “citizen data scientist” never worked as an industry frame

• How generative AI changes the question from “train users” to “meet users where they are”

• Designing for the user already in the seat, not the one we wish were there

• Where data quality and trust shift once natural language becomes the interface

• Qlik Connect 2026 themes and what practitioners should watch next

A design premise, not an enablement story

For 15 years, BI vendors pitched “citizen data scientist” as the answer for enabling non-data people. After endless debates about citizen dentists and citizen pilots, the term fizzled away. Most business users have no interest in becoming part-time data scientists. They want answers and recommendations on how to improve their business operations. Mary was direct about why. “It puts a lot of onus on people when that may not be their calling or aptitude,” she said.

Mary’s reply to that was a different design premise.

“Most people are horrible prompters. You have to bake that into the experience.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

Citizen data scientist asked the user to get better. With “horrible prompters,” the design question shifts to how the tool can get smarter about the user sitting in front of it. That reframes the work from enablement to design. For 15 years, self-service BI pushed the cognitive load onto the end user, who was expected to learn the data model, the query language, and the tool. Mary’s view is that generative AI changes the equation. It “really meets everybody where they’re at and their skill set.” Users don’t have to level up before getting an answer.

That design premise lines up with what Capone, Qlik’s former CEO, had been telling the market all year. Before the event, he described Qlik’s approach as helping teams engage data “through agentic conversations that lead to action, with governance and efficiency built in.”^[4] Mary’s design premise is the former CEO’s operating-model thesis at the UX layer.

Meeting users where they are

Mary’s design premise has a logical consequence. If the tool absorbs the skills burden at the UX layer, the quality and trust burden shifts one layer deeper.

“It really gets pushed down to one step behind analytics, which is the data product.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

With natural language as the interface, users stop worrying about field names or SQL syntax. The tool handles that. What users are actually depending on is the foundation underneath, whatever data the tool reaches for, and whether that data is correct.

Qlik’s Connect 2026 announcements make the shift concrete. Qlik Answers is the entry point, combining structured analytics and unstructured content. Discovery Agent surfaces anomalies before humans think to ask about them. Predict Agent builds models and answers forward-looking questions. Automate Agent pushes insights into downstream workflows. Analytics Agent accelerates development tasks.^[5] Together they form a continuous path from question to action.

Mary walked through what this looks like at the user level. The system takes a messy question and reframes it. It surfaces relevant data without requiring users to name fields or tables. When a question is ambiguous, the system flags it and asks for clarification instead of silently guessing.

Delivery matters as much as the pipeline. Rather than asking users to come to Qlik, the platform reaches users in whatever environment they already work in. MCP Server lets users invoke Qlik’s analytics engine from Claude, ChatGPT, Gemini, or whatever assistant their organization has standardized on. Brendan said Qlik is already seeing roughly 50/50 usage between its native interface and MCP for agentic capabilities. Agents run in the background, surfacing what matters without a dashboard login. The pane of glass is wherever the user already is.

That is “meeting users where they are” at the product level, not just the UX level. It is Capone’s “freedom” pillar executed in shipping code. For 15 years, the question was how to train more business users to work with data. Now the question is how to put trustworthy, governed data in front of users in the environments they already trust, including AI assistants that never ran inside the analytics stack.

Trust as a hard requirement

If the onus has shifted one layer deeper, that layer has to be trustworthy. Capone, who has since left Qlik, put it bluntly in the run-up to the event.

“AI is moving from an interesting capability to an operational expectation. The moment it touches real decisions, trust becomes a hard requirement, not a slogan.”

— Mike Capone, former CEO, Qlik

In an agentic era, the urgency is sharper. An agent doesn’t pause to gut-check a suspicious number. It takes the data, acts on it, and passes the result to the next step. By the time anyone notices a problem, the decision has already shipped.

Qlik’s response is to make trust operable. The Connect 2026 announcements on data products include a Trust Score that evaluates data products across accuracy, timeliness, diversity, and completeness. Data contracts define what a data product is expected to provide. The Data Product Agent helps teams create, manage, and evaluate data products using natural language.^[6] They turn trust into a visible operational signal rather than an assumed quality.

This reflects a deeper shift Capone signaled throughout his time leading Qlik. The old pendulum between tight central control and chaotic self-service is breaking down.^[7] What replaces it is controlled decentralization, with governed data products distributed to wherever users, human or agent, can make use of them. That requires governance, data contracts, semantic layers, lineage, and access controls to stop being back-office hygiene.^[8] They become the place where AI succeeds or fails.

Retire “citizen data scientist.” Invest in the data foundation and the delivery mechanisms that meet users in the environments they already work in.^[9]

Near the end of our interview, Mary captured the shift in one line. “We just have new ways of solving these old problems.” The hard part just stopped being the user’s job.

Listen to the full conversation with Mary Kern on the Data Faces Podcast.

Based on insights from Mary Kern, Vice President, Product Go-to-Market at Qlik, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What does it mean to meet users where they are in product design?

Meeting users where they are is Mary Kern’s design philosophy for the agentic era. Instead of training users to phrase questions better, the tool absorbs the skills burden. Qlik’s agentic experience reframes messy questions, surfaces relevant data without requiring field names, and flags ambiguous questions instead of silently guessing.

What were the keynote themes at Qlik Connect 2026?

Former Qlik CEO Mike Capone framed the keynote around three words: context, trust, and freedom. AI has to understand the business in context, run on a trusted foundation, and connect insight to action in the systems teams already use. Qlik’s announcements at Connect 2026 extended this with MCP servers, agentic experiences, and an open ecosystem that meets users in whatever environment they already work in.

Where should D&A leaders invest to prepare for agentic AI?

D&A leaders should shift investment from end-user enablement programs to the data foundation underneath. Governance, data contracts, semantic layers, lineage, and access controls stop being back-office hygiene when AI becomes the interface. Agents don’t pause to gut-check suspicious data, so whatever they read has to be trustworthy before they touch it. That is where AI succeeds or fails.

How does natural language interface shift the burden away from users?

When natural language becomes the interface, users stop worrying about field names or query syntax. The tool handles that. What users actually depend on is the foundation underneath, whatever data the tool reaches for and whether it is correct. The burden moves from the user one layer deeper, to the data product layer that serves every question.

Podcast highlights

- [0:00] Introduction on the Qlik Connect 2026 show floor

- [0:24] Mary’s expanded role at Qlik

- [0:50] Gardens and a suburban raccoon cam

- [2:14] Qlik Connect 2026 keynote highlights

- [4:10] Qlik’s agentic experience and “a couple toggles to production”

- [6:30] What is different about enabling business users this time

- [8:20] Flexibility and meeting users where they are

- [11:15] The Qlik community

- [12:11] Cutting through the agentic noise

- [14:42] Storytelling and customer validation

- [16:13] What is next for Qlik in 2026

- [17:30] The “dare to be different” theme

About David Sweenor

David Sweenor is the founder and host of the Data Faces Podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Qlik. “Qlik Extends Analytics from Answers to Agentic Action.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-extends-analytics-from-answers-to-agentic-action.

^[2]Qlik. “Qlik Connect 2026 Shows Enterprises Are Closer to Agentic AI Than They Think.” Press release, April 15, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-connect-2026-shows-enterprises-are-closer-to-agentic-ai-than-they-think.

^[3]Sweenor, David. “Why Bad Data Didn’t Matter Until Now.” TinyTechGuides, April 2026. https://tinytechguides.com/blog/why-bad-data-didnt-matter-until-now/.

^[4]Qlik. “Jesse Cole, Creator of the Savannah Bananas, to Keynote Qlik Connect 2026.” Press release, January 28, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/jesse-cole-creator-of-the-savannah-bananas-to-keynote-qlik-connect-2026.

^[5]Qlik. “Qlik Extends Analytics from Answers to Agentic Action.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-extends-analytics-from-answers-to-agentic-action.

^[6]Qlik. “Qlik Makes Trust Operable for Data Products.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-makes-trust-operable-for-data-products.

^[7]Qlik. “Qlik CEO: Enterprises Are Underachieving on AI, With Islands of Value in a Sea of Noise.” Press release, January 15, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-ceo-enterprises-are-underachieving-on-ai-with-islands-of-value-in-a-sea-of-noise.

^[8]Sweenor, David. “Why Bad AI Governance Kills 95% of Enterprise Projects Before Production.” TinyTechGuides, September 9, 2025. https://tinytechguides.com/blog/why-bad-ai-governance-kills-95-percent-enterprise-projects/.

^[9]Sweenor, David. “Why 80% of AI Projects Fail (And the Three Boring Decisions That Save the Other 20%).” TinyTechGuides, October 21, 2025. https://tinytechguides.com/blog/why-80-of-ai-projects-fail-and-the-three-boring-decisions-that-save-the-other-20/.

The data, analytics, and AI industry is currently obsessed with production velocity. Every vendor is promising that their AI agents can automate workflows, draft emails, order your groceries, and analyze your pipeline in seconds. It sounds great on paper, but when push comes to shove, there’s certainly room for improvement. In my work with clients who are building custom agents, they have some serious concerns and reservations about agentic AI, which are certainly justified. While the agents are fast and dutifully execute tasks assigned to them, they are often confidently wrong more often than not. This occurs because your AI likely has a data-context problem, and context serves as the anchor for accuracy, which agents often lack. This disconnect is reflected in recent research from MIT (2025), which found that 95% of enterprise AI projects fail to deliver measurable P&L impact, often due to a failure to integrate the model with actual business context and workflows.^[1]

When a human being looks at a flawed quarterly business review (QBR) report, they can often spot errors immediately. They understand the business and know that a merger happened last quarter, and that the currency conversion for the EMEA region is manual, and that the “Total Revenue” field excludes services. They understand the relationships between the data and the business outcomes.

AI agents lack this baseline intuition. Without a rich layer of metadata to provide this context, an agent operates as a fast guesser. Sometimes it’s no better than the predictive text capability on my iPhone, which I must admit, is not that great. I recently sat down with Steve Wooledge, CMO at Collate, on the Data Faces Podcast. Steve has spent 20+ years in the datasphere, from Teradata and SAP to leadership roles at Alteryx and Alation. He has seen the hype cycles move from big data to generative AI, and he believes we have reached a shift in how we manage data. To move from experimental AI to reliable, agentic operations, we must treat metadata as the foundational instruction manual for machine intelligence.

About Steve Wooledge

Steve Wooledge is the Chief Marketing Officer at Collate, the company behind the OpenMetadata project. His career spans over 25 years in enterprise sales and marketing leadership at industry giants, including Teradata, SAP, and Business Objects. Steve is a recognized expert in technical product marketing and category creation, having previously built global partner programs at Alteryx and led product marketing at Alation. Outside of the data industry, Steve is a dedicated guitar player with a passion for melodic hard rock and blues.

In our conversation on the Data Faces Podcast, we discuss several hot topics for the agentic era. These include the transition from chemical engineering to product marketing leadership and how to build a “Switzerland” strategy for metadata across multi-vendor ecosystems. We also explore the shift from Data Intelligence to Semantic Intelligence for AI agents and the “Taste Squared” formula for maintaining marketing quality in an automated world.

The evolution of metadata. From inventory to foundation

Metadata spent twenty years as the ignored part of the primordial data stack and remained the least interesting part of the infrastructure. It served as a technical inventory, used to confirm that a specific column was an integer or that a timestamp used a specific format. It served as a technical necessity for database administrators but rarely provided direct, observable business value.

In our conversation, Steve identified three distinct stages in the move toward semantic intelligence. The “Technical Inventory” stage used metadata as a governance checkbox. This evolved into “Data Intelligence,” which is a term popularized by Stewart Bond and companies like Alation that expanded the definition to include the “who, what, where, when, and why” of data.^[2] This stage moved beyond the technical schema to include the operational context of how people used the information.

We are now entering the “Agentic” stage. In this era, metadata is a tool for machines as much as for people. Steve explained that while metadata describes your data, for AI to be accurate and intelligent, it needs that foundational context to prevent hallucinations. If you want an AI agent to pull a report or automate a task, the system must understand the rules and relationships that govern that data. This foundation transforms an automated guesser into an intelligent system.

Semantics. giving AI a “gut feel”

The separation of data from meaning creates the core challenge in modern AI architectures. A human analyst looking at a database sees a column named rev_adj and intuitively understands it refers to a manual revenue adjustment. AI agents require an explicit map to reach that same conclusion. Steve describes this map as “Semantic Intelligence,” which is a framework that provides the digital equivalent of a “gut feel.”

“Semantics is the overall structure and meaning. It includes the relationships between different data elements so an agent can traverse the graph to get at the reason and meaning.”

— Steve Wooledge, CMO, Collate

Technical metadata describes the storage, including the columns, types, and primary keys. Semantic metadata tells you about the intent, such as business rules, KPI definitions, and ontologies. When these relationships are mapped in a graph, an agent can traverse those connections to reason through a query. It can understand that a “Customer” in the CRM is the same entity as a “Subscriber” in the billing system, even if the underlying schemas look different.

By traversing this semantic graph, an AI agent can self-correct. It can recognize when a requested calculation violates a business rule or when a data point lacks the necessary context to be included in a final report. This architectural clarity allows an automated system to operate reliably in a complex enterprise environment. Without this layer, AI projects often get stuck because the models lack the fundamental ability to reason through data dependencies.

The Switzerland strategy. Why AI needs a neutral layer

Enterprise data is inherently messy. Even the most disciplined organizations suffer from fragmented ecosystems, with critical information scattered across Snowflake, Databricks, legacy on-premises databases, and SaaS applications. Each of these platforms offers its own proprietary version of metadata management, and this creates a series of disconnected silos. If your AI strategy relies on the metadata layer of a single platform, you are building on a foundation that cannot see the full picture.

This fragmentation necessitates what Steve calls a “Neutral Layer.” He argues that AI agents require a “Switzerland” strategy: an agnostic metadata layer that sits between the various data silos and the AI models. This neutral layer provides a consistent view of the business logic regardless of where the data lives. It ensures that when an agent asks for “last month’s churn rate,” the definition remains identical whether the data is pulled from a cloud warehouse or a regional database.

“There is no neutral layer that sits across all of that. You need to have this agnostic layer of metadata and semantic intelligence that sits between the data and the AI to ensure you understand the meaning of the information.”

— Steve Wooledge, CMO, Collate

Adopting an agnostic approach also provides a hedge against future architectural changes. As companies undergo mergers, acquisitions, or switch vendors, a proprietary metadata strategy elevates risk and becomes a liability. By using open standards like OpenMetadata, organizations can preserve their semantic intelligence as their underlying infrastructure evolves. Steve’s view is that this neutral layer is the primary way to ensure that your business rules remain portable and your AI remains accurate as you scale across multiple platforms.

Marketing the abstract. Lessons from a first-principles CMO

Marketing a technical product requires a unique level of architectural clarity. If you cannot map the relationships between your data elements, you will struggle to map your message to the specific problems your customers face. Steve credits much of his approach to his time at Business Objects, working under Dave Kellogg, who is a veteran leader who preached the power of first principles. This philosophy dictates that marketing exists to reduce friction in the sales process by grounding every message in clarity and logical sequence.

When you sell an abstract concept like a “metadata platform,” you cannot lead with features. A CFO or CEO rarely wakes up thinking about their cataloging needs. Instead, you must sell the business outcomes that metadata enables, such as AI safety, operational velocity, and what we call consequence management. By visualizing the invisible through semantic metadata graphs, marketers can make these complex technical structures tangible for executive budget owners.

This first-principles approach also fuels grassroots expansion through open-source communities. By allowing developers to solve immediate technical problems using tools like OpenMetadata, a company can build a foundation of trust before moving toward an enterprise-wide engagement. Steve’s experience at Alation and Alteryx confirms that when you give people the tools to prove value in their own environment, the transition to a strategic partnership becomes a logical next step.

The “taste squared” era of marketing

Velocity without judgment is just noise. The integration of AI into marketing workflows has fundamentally changed the expectations for production velocity. We can now develop content, campaigns, and landing pages at a pace that was previously impossible. However, this increased speed introduces a risk that Steve calls “lazy marketing.” While AI can generate high volumes of content, it often lacks the subtlety and judgment required to connect with a specific customer base.

To address this challenge, Steve references a formula popularized by Tom Wentworth^[3], where marketing output equals AI adoption multiplied by taste squared. This perspective suggests that while adopting AI is a linear requirement for modern teams, human taste acts as an exponential multiplier for quality. Having the technical skill to use a prompt is one thing, and having the taste to know when a message is great, and when it is “average AI,” is what will differentiate the leaders from the laggards.

Maintaining this level of quality requires a commitment to the human element of marketing. In our conversation, Steve emphasized that you still have to “slave over the word” to ensure your message correctly lands. This means using AI as a tool for acceleration rather than a replacement for thinking. By combining automated velocity with rigorous peer review and high creative standards, marketing leaders can use AI to amplify their impact without sacrificing brand integrity.

The infrastructure of trust

Building an AI strategy without a solid metadata foundation is like attempting to build a penthouse on a swamp. The agents you deploy will only be as intelligent as the context you provide them. By adopting a neutral, semantic metadata layer, organizations can equip their AI systems with the digital intuition needed to move beyond simple automation and toward autonomous operations.

Metadata is the primary architectural anchor of the agentic era, supporting both governance and agentic AI. To learn more about how to build this foundation for your own organization, you can listen to the full conversation with Steve Wooledge on the Data Faces Podcast and explore the open-source community at OpenMetadata.

Listen to the full conversation with Steve Wooledge on the Data Faces Podcast.

Based on insights from Steve Wooledge, CMO at Collate, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What is the difference between metadata and semantics? Metadata describes the technical properties of data, such as column names, data types, and timestamps. It acts as a technical inventory of information. Semantics represents the overall structure, meaning, and relationships between those data elements. While metadata tells an AI agent what a field is, semantic intelligence tells the agent how that field relates to business rules and KPIs across the enterprise.

Why do AI agents need a neutral metadata layer? Most organizations store data across fragmented ecosystems like Snowflake and Databricks. Each platform manages metadata in its own proprietary way. A neutral metadata layer sits between these silos and the AI models, providing a consistent, agnostic view of business logic. This strategy ensures that an agent’s understanding of the data remains accurate even if the underlying infrastructure changes.

How does semantic intelligence prevent AI hallucinations? AI hallucinations often occur because the model lacks the necessary context to interpret data correctly. Semantic intelligence provides a mapped graph of relationships that allows an AI agent to reason through a query like a human analyst. By traversing this graph, the agent can identify when a requested calculation violates a business rule or when it lacks the context required for an accurate response.

What is the taste squared formula for marketing? CMO Tom Wentworth introduced the formula. Marketing output equals AI adoption multiplied by taste squared. It suggests that while AI adoption is a linear requirement for productivity, human taste is an exponential multiplier for quality. In an era where anyone can use AI to generate average content, the differentiator for marketing leaders is the judgment required to refine AI output into something resonant.

Podcast highlights

• [0:00] Introduction to Steve Wooledge and Collate

• [1:08] The journey from chemical engineering to technical data sales

• [3:45] Melodic hard rock and guitar shredding as a creative outlet

• [5:01] Lessons from Dave Kellogg on first-principles marketing

• [7:40] The reality of partner marketing with global system integrators

• [10:18] Why open-source projects out-innovates proprietary enterprise software

• [15:56] The shift from technical metadata to semantic intelligence for AI agents

• [20:45] Building a Switzerland approach to metadata across multi-vendor silos

• [24:03] How AI velocity is fundamentally changing the CMO role

• [27:23] The taste squared formula and why you cannot be a lazy marketer

• [32:19] Career advice for the next generation of data and marketing professionals

• [36:28] Final advice on peer review and maintain quality control

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]MIT Project NANDA. 2025. “The GenAI Divide: State of AI in Business 2025.” MIT Sloan Management Review.

^[2]Bond, Stewart. 2026. “Your AI has a data intelligence problem.” TinyTechGuides.

^[3]Wentworth, Tom. 2024. “AI Adoption and the Taste Square.” incident.io.

Kate Strachnyi has a habit of calling people out on LinkedIn. When she spots a post that’s been run through an AI rewriter, she’ll send the person a direct message. “Hey, I could tell you used AI,” she’ll say. The usual response is some version of “but these are my thoughts,” and her advice back is to keep them that way and stop running them through the machine. She calls that keeping your content “non-GMO,” unmodified, and authentic.

It’s a funny line, and it also describes her entire business model. Kate is the founder of DATAcated, a media company that partners with brands in data, analytics, and AI to reach their audiences through real content creators and thought leaders. In a market where AI-generated posts are flooding every feed, and ironically, LinkedIn itself has a “rewrite with AI” button baked into the platform, Kate is making the opposite bet. She’s building a business around real people with real expertise and real opinions.

On Episode 38 of the Data Faces Podcast, I sat down with Kate to talk about how she built DATAcated from a one-person experiment into an influencer agency with 40+ creators, why she’s doubling down on authenticity as AI content takes over, and what happens to expertise itself when the humans who hold it stop doing the work.

“Keep it non-GMO. Don’t modify your content, just leave it as is.”

— Kate Strachnyi, Founder, DATAcated

About Kate Strachnyi

Kate Strachnyi is the founder of DATAcated. She started her career in finance and risk management consulting before pivoting to data visualization 12 years ago. Kate has since written five books, established one of the most connected networks of data and AI professionals in the industry, and built DATAcated into a full agency, with 40+ influencers, speakers, and subject-matter experts through her DATAcated Plus program. She is a LinkedIn Top Voice. In our conversation on Episode 38 of the Data Faces Podcast, we discuss:

- How Kate followed the revenue data from courses and books to a focused media business

- The DATAcated Plus model and how influencer campaigns work behind the scenes

- Why she’s shifting from “Kate = DATAcated” to an agency brand

- The flood of AI-generated content on LinkedIn and her “non-GMO” content philosophy

- What happens to expertise when today’s subject matter experts retire and AI fills the void

Following the data from finance to media

Kate’s path to running a media company started with a practical constraint. She was working in financial services risk management at a large consulting firm, traveling Monday through Thursday, and expecting her first child. She spent eight months searching for any role that would let her work remotely, well before remote work was mainstream. Someone eventually pointed her toward a data role that involved Tableau, visualization, and “creating pretty pictures.” She took it and fell in love with data storytelling.

“I am a data person, right? So I would look at the numbers and see what is driving more revenue and what is allowing me to have more time to myself.”

— Kate Strachnyi, Founder, DATAcated

What followed was a period of deliberate experimentation. Kate wrote books and launched an academy. She created courses, ran her own conferences, and built a community called DATAcated Circle. As a business of one with nobody to stop her, she could try anything, and she tracked the results. The revenue data and the work she enjoyed most both pointed to media and content creation. That’s where DATAcated lives today.

Building DATAcated Plus — from personal brand to influencer agency

When Kate started getting more client work than she could handle, she brought in the content creators she already knew. DATAcated Plus was born.

The program now spans data, analytics, and AI, with the agentic AI space growing fastest. When a client comes to Kate with a product launch, an event, or a brand awareness campaign, she matches them with creators based on their audience, expertise, and the success metrics the client is targeting. She reviews every piece of content before it goes to the client for approval. The day-to-day project management runs underneath, with UTM links, calls to action, and timelines coordinated across every moving piece. None of the work is glamorous, but it’s what keeps the whole operation running.

The DATAcated Plus roster also includes speakers and experts that companies can hire for keynotes and panels, as well as webinars and thought leadership papers. Kate has placed people across major industry events, including multiple years at the Gartner Data & Analytics Summit, where her team creates what she calls “FOMO-inducing content” on-site.

Scott Taylor, the Data Whisperer, joined Kate at last year’s Gartner event to co-lead a sold-out breakfast session on personal branding for data leaders. One question from the audience: “I want to post, but my company said no.” Kate’s advice was to follow your company’s rules but find the leeway. Talk about your perspective on industry topics rather than your specific projects or tools. When there’s no leeway at all, I suggested it might be time to find a company that sees your personal brand as an asset rather than a risk.

At Big Data London, a group of DATAcated Plus creators went to a tattoo parlor and got fake data tattoos for a video so convincing that Kate’s neighbors congratulated her on the new ink.

“People will unfollow instantly if we just keep promoting things. It’s more of letting my audience know about here’s a product that exists, and here’s what it does, in case you need it.”

— Kate Strachnyi, Founder, DATAcated

That creative range is what separates the DATAcated model from a traditional analyst engagement. Analyst firms produce authoritative research with independent evaluations, and content creators bring flexibility, personality, and a wider range of formats from short-form video to live event coverage.^[1] Kate’s crew has done cooking shows to explain data governance and built sandcastles to illustrate strong data foundations. The content sticks with audiences, and brands gain reach from creators who have built genuine trust over years of showing up with their own voices.

More recently, Kate has changed how she positions the company itself. The old DATAcated media kit led with a big photo of Kate and a rundown of what she could do. The new version leads with the influencer roster. She’s deliberately moving from “Kate equals DATAcated” to an agency brand that doesn’t depend on her being in every room. Some clients still ask for Kate specifically, and she’s learning to redirect them toward creators who are a better fit. “It’s nice to be wanted,” she told me, “but it doesn’t scale.”

The authenticity bet in an AI-saturated feed

Kate is leaning harder into genuine human voices at the exact moment AI-generated content is flooding LinkedIn and every other platform. LinkedIn added a “rewrite with AI” button to the post editor, while its own users complain that AI-generated slop is taking over their feeds. Kate’s view is that if you know a person well enough, you can spot AI-written content immediately. The vocabulary is off, the phrasing is too polished, and the voice sounds like everyone else’s. She calls people out on it, and she expects the same standard from her DATAcated Plus creators. For Kate, authentic content means it was written by the credited human, reflects their expertise and opinions, and hasn’t been reprocessed through an AI rewriter.

That doesn’t mean she’s anti-AI. Kate recently spent an entire day automating her invoicing process in Claude Code. The task itself takes two minutes, but she never has to do it by hand again. She and I compared notes about automating YouTube uploads, scheduling content, and eliminating the repetitive copy-paste work that eats up a solopreneur’s day. She draws the line between back-office operations and audience-facing content. AI can handle the invoices. It should not rewrite your LinkedIn posts.

“What are we going to do 20 years from now, when we don’t have those subject matter experts? They’re retired or not working anymore. Because if you don’t work with this stuff, whatever that might be in the medical field, in construction, how are you going to fact-check it?”

— Kate Strachnyi, Founder, DATAcated

Part of our discussion focused on a question Kate had heard at a recent event. Right now, subject matter experts can look at AI-generated output and spot what’s wrong because they’ve spent decades doing the work firsthand. But what happens in 20 years, when those experts have retired? If the next generation learns from AI output instead of from direct experience, the ability to verify and correct that output disappears.

The humans who make genuine content valuable are also the humans who keep AI honest. Kate sees her business as part of the answer. Invest in real experts now, amplify their voices, and make sure the knowledge doesn’t evaporate into a feedback loop of machine-generated content. The window to establish yourself as a real authority, someone whose voice carries weight because it’s grounded in lived experience, won’t stay open forever.

I started the Data Faces Podcast to have real conversations with the people doing the work. The messy, honest, sometimes funny exchanges that you can only get from humans who have opinions and aren’t afraid to share them. Kate’s business is built on the same conviction. In a world filling up with synthetic content, she’s betting that real voices will only become more valuable. When I asked her about deepfakes and AI versions of herself, her answer was four words.

“I plan to remain authentic.”

— Kate Strachnyi, Founder, DATAcated

Listen to the full conversation with Kate Strachnyi on the Data Faces Podcast.

Based on insights from Kate Strachnyi, Founder at DATAcated, featured on the Data Faces Podcast.

Leave a comment

Frequently asked questions

What is DATAcated, and what does the company do?

DATAcated is a media company founded by Kate Strachnyi, author of five books, including ColorWise and Journey to Data Scientist, that helps brands in data, analytics, and AI reach their target audiences through authentic content creators and thought leaders. The company runs the DATAcated Plus program, a roster of 40+ influencers, speakers, and subject-matter experts who create thought-leadership content, amplify product launches, and represent brands at industry events such as the Gartner Data & Analytics Summit. DATAcated operates as an agency that matches creators to client campaigns based on audience fit and success metrics.

What is the DATAcated Plus program?

DATAcated Plus is Kate Strachnyi’s influencer and speaker program for the data and AI industry. It includes content creators, speakers, and subject matter experts across data analytics, AI, and agentic AI. Companies hire DATAcated Plus members for brand awareness campaigns, event coverage, webinars, thought leadership papers, and on-site content creation. Kate manages the program by vetting creators for authenticity, reviewing all content before client approval, and coordinating timelines and deliverables across campaigns.

How does influencer marketing differ from analyst relations in B2B tech?

Analyst firms like Gartner and Forrester produce authoritative research and independent evaluations. Influencer content creators offer more creative flexibility and can be guided toward specific messaging for a campaign. Kate Strachnyi’s DATAcated Plus creators have done cooking shows to explain data governance and built sandcastles to illustrate data foundations. They’ve also produced viral video content at industry events. Both approaches build credibility with B2B audiences, and many companies now use influencers and analysts together at the same events.

What does “non-GMO content” mean in the context of AI and social media?

“Non-GMO content” is Kate Strachnyi’s phrase for content that hasn’t been run through an AI rewriter. Just as non-GMO food is unmodified, non-GMO content preserves the author’s original voice and phrasing. Kate advocates for this approach because AI-rewritten posts lose the personality and authenticity that make content creators valuable to their audiences. She actively calls out AI-washed posts on LinkedIn and holds her DATAcated Plus creators to the same standard.

Will AI replace subject matter experts in data and AI content?

Kate Strachnyi raises a concern about long-term expertise. Today’s subject matter experts can spot errors in AI-generated content because they have decades of hands-on experience. In 20 years, when those experts have retired, the ability to fact-check and verify AI output may disappear if the next generation learns from AI-generated content rather than direct experience. Kate’s business model is built around investing in real human experts now and amplifying their voices before that institutional knowledge erodes.

Podcast highlights

[0:05] Kate’s background and what DATAcated does

[2:10] Pre-finance Kate: what she wanted to be before data found her

[3:05] The career pivot from risk management consulting to data visualization

[5:03] How DATAcated evolved from training and books to a focused media company

[7:27] How the influencer model works behind the scenes

[9:33] Automating business operations with Claude Code

[11:01] Walking the line between brand amplification and spam

[14:11] The fake tattoo story from Big Data London

[15:03] How DATAcated Plus compares to analyst firm engagements

[17:14] The sold-out personal branding session at Gartner with Scott Taylor

[22:15] Shifting from “Kate = DATAcated” to an agency brand

[24:02] What works on LinkedIn now vs. five years ago

[27:01] AI-generated content flooding feeds and the “non-GMO” philosophy

[29:04] The 20-year question: who fact-checks AI when the experts retire?

[30:20] Deep fake Dave and why Kate plans to remain authentic

[31:24] Why Kate hasn’t hired a team and is betting on AI for operations

[33:57] Does AI make you more productive or just busier?

[36:19] Where to find Kate and DATAcated

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “Stop Writing AI Content That Sounds Like Everyone Else’s.” TinyTechGuides, February 7, 2025. https://insights.tinytechguides.com/p/stop-writing-ai-content-that-sounds

For 25 years, data quality has been everyone’s problem and nobody’s priority. For some, it was an IT problem, and for others, it was a business problem. But most of the time, fixing it at scale was largely ignored. What would you do if a number in the spreadsheet looked off? You’d fix it and move on with your day. The same with questionable metrics on dashboards. We’ve been able to tuck and hide the cost of bad data in a manual world for a while now. Since the pace of business was slower, there were no real consequences for getting it wrong.

Those ways of old change when you hand autonomy to an AI agent. An agent doesn’t pause to gut-check a suspicious number, it doesn’t really care. It takes the data at face value, makes a decision, feeds that decision into the next step, and keeps going. You might be six or seven steps down the line before anyone realizes the foundation was wrong. And by then, the damage compounds in ways that a quick spreadsheet fix can’t undo.

I sat down with Brendan Grady, EVP and General Manager of Analytics and AI at Qlik, at Qlik Connect 2026 in Orlando to discuss why the stakes around data quality have changed, where enterprise-agentic adoption stands today, and what data professionals should be thinking about.

“In today’s world where there may be an agent running around using said data and getting it wrong, the consequences of getting it wrong are going to be catastrophic.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

About Brendan Grady

Brendan Grady is EVP and General Manager of the Analytics and AI Business Unit at Qlik, where he leads product management, product design, R&D, and go-to-market strategy for the company’s data integration, quality, and analytics platform. Before Qlik, he held senior GTM roles at IBM, where he led worldwide digital sales for Watson Analytics and managed the Cognos portfolio. He joined Qlik seven years ago after repeatedly losing deals to its analytics engine, and decided to find out why. And well before all of that, he delivered the Sound of Music tour in Salzburg, Austria, over 300 times.

In this episode, we discuss:

- Why data quality was never fixed and why that matters now

- Where enterprise agentic AI adoption actually stands

- Trust scores and the problem with feeding spreadsheets to LLMs

- The shift from dashboards to decision intelligence

- Open standards, MCP, and why there’s no “One Ring to rule them all”

- Advice for data professionals navigating the AI transition

The consequence management problem

Grady framed the data quality conversation in a way I hadn’t heard before. He called it consequence management. For decades, organizations tolerated bad data because the consequences of getting it wrong were manageable. A field was incorrect in a report? Someone caught it, fixed it, and everyone moved on, knowing there would be another fire drill tomorrow. The recovery cost was low enough that nobody prioritized prevention, and if they did, they rarely had the organizational backing to make any meaningful change.

“Is it IT’s job? Is it the business’s job? Is it both, or is it nobody’s job? For most companies, it’s been nobody’s job.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

BARC’s research confirms this pattern. As Shawn Rogers discussed on the Data Faces Podcast, data quality remains the top challenge for organizations trying to mature their analytics and AI capabilities.^[1] That organizational ambiguity persisted because the stakes allowed it. He pointed to real examples. A major airline took a significant hit to its market cap because its sentiment data was wrong and decisions were made on flawed analysis. Two decades ago, a single field in a spreadsheet contributed to a financial crisis that rippled through an entire market. These weren’t hypothetical scenarios. They happened because nobody owned the problem and the systems in place couldn’t detect the errors before they cascaded.

In the agentic era, the failure mode is different. A human looking at a dashboard might notice something feels off and investigate. An agent won’t. It will take the data, reason through it, make a decision, and pass that decision to the next agent in the chain; often without any confidence bounds or trust scores.

The point isn’t that agents are dangerous. The point is that autonomous systems need trusted data underneath them before they’re given the authority to act. Without that bedrock, every step an agent takes amplifies whatever error was baked into the starting point. As practitioners, we know this, why hasn’t this been fixed yet?

“Prior to stage zero”

I asked Brendan where enterprise agentic adoption actually stands. His answer was honest. “What’s prior to stage zero?” he said. “I mean, there are customers that are trying things out there, surely. But from a large-scale production standpoint, we’re in the early days.”

Customers are experimenting with low-risk use cases. They’re testing agents in controlled environments where a mistake won’t damage the business. But production-grade agents making real decisions in real business processes? That’s rare. And the blocker, according to Brendan, isn’t the technology. It’s the data.

Gartner projects that by 2027, 70% of organizations will adopt modern data quality solutions to support AI adoption and digital business initiatives.^[2] That projection tells you where the market is today. If 70% will need to adopt these solutions by 2027, most organizations don’t have them yet. The ambition around agentic AI is running well ahead of the data infrastructure required to support it. Shane Murray made a similar argument on the Data Faces Podcast earlier this year, noting that actionable data strategies beat endless planning when it comes to AI readiness.^[3]

Brendan also raised a practical question that every data leader should be asking. The LLM landscape is shifting constantly. Six months ago it was OpenAI. Today, Claude is gaining traction. Tomorrow the market may have moved on to something new. His advice was to work with vendors that approach this from an open standards perspective, supporting multiple LLMs rather than forcing a single choice. The technology will keep changing, but the data underneath it is what has to hold steady.

“The internet took 10 years, 20 years, 30 years to get going. We’re a year and a half in.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

Trust as the missing layer

One of the more revealing moments in our conversation came when Brendan talked about what happens when you feed structured data into an LLM. I’ve experienced this myself. You upload a spreadsheet, ask it to calculate something, and the answer comes back looking polished and confident. The formatting is clean, the language is professional, and unbeknownst to you, the numbers are wrong.

“It’s really pretty, right? The answer is amazing. Looks great. Totally BS. And the next thing you know, you’re showing up to the board with all incorrect numbers.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

Qlik’s Trust Score for AI is designed to give decision-makers a quantifiable measure of whether their data is valid, fresh, and representative before it reaches an agent or an LLM.^[4] Instead of hoping your data is accurate, you can see a score that tells you it’s 90% trustworthy or 80% or something that should give you pause.

The other piece Brendan emphasized was intent detection. When someone asks a question of an LLM, the literal question and the actual intent are often different things. I ran into this recently when I asked an AI assistant to analyze several websites. It came back with a confident analysis, but when I pressed it, it admitted it had never actually visited the sites. Qlik is investing in understanding what the user is really trying to accomplish so the system can route to the right data and the right engine rather than letting an LLM fabricate its way to an answer.

The combination of trust scores and intent detection reflects a broader principle. Before you give an agent the authority to act on data, you need to know that the data is sound and that the system understands what you’re actually asking. Qlik’s track record in this space is long. The company has been named a Leader in the Gartner Magic Quadrant for Augmented Data Quality Solutions for seven consecutive years, most recently in 2026.^[5]

“Dashboards are dead. Long live dashboards.”

When Brendan declared that dashboards are dead, I thought I had a scoop and I made sure the audience heard it. He laughed and then walked it back with the nuance that matters. Dashboards as a destination are going away, but the data inside them and the decisions they inform are more important than ever.

Brendan described how his own workflow has changed. He used to ask his analytics tools for information about business performance. Now he asks a different question. “Tell me about my business and what you think I should do.” That shift from information retrieval to decision recommendation is what Qlik means by decision intelligence, and it’s powered by two things working together.

The first is Qlik’s analytics engine, which finds associations and relationships in data that other approaches miss. Instead of running a predefined query to answer a specific question, the engine surfaces connections you didn’t know existed. Brendan called these the unknown unknowns. In an agentic context, that capability becomes even more valuable because it allows agents to explore paths and relationships that a standard SQL query would never surface.

“In the agentic world, we’re serving this up to help agents understand that there’s a relationship here that you need to go explore before you take action. That is extremely powerful.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

The second is openness. Qlik launched its MCP server in February 2026, implementing the open Model Context Protocol standard to let third-party AI assistants access Qlik’s analytical capabilities with governance built in.^[6] “There’s never going to be One Ring to rule them all,” he said. People want to work in the tools they’re comfortable with, whether that’s Claude, Gemini, ChatGPT, or something that doesn’t exist yet. The bet is paying off. Brendan shared that they’re already seeing roughly a 50/50 split between users accessing agentic capabilities through Qlik’s own interface and those coming in through MCP.

“Am I out of a job?”

Brendan closed our conversation with a story that we’ve all encountered. After demoing the ability to build an analytics application through Claude in 30 seconds at Qlik Connect, a customer approached him. This person had built his entire career writing code to create analytics applications across multiple platforms. His question was simple. “Am I out of a job?”

Brendan’s answer was no, but with an important caveat. The job will evolve. His advice to data professionals was to lean into what they already know better than anyone else: the data itself. Become the data product owner. Be the trusted guide as organizations navigate the agentic experience. The people who understand the data well enough to know its quirks and business context will be indispensable as agents take on more routine work.

This tracks with what Brendan’s team has seen internally. Qlik has developers who were already performing well, and AI tools have turned them into 10x contributors. The acceleration is happening at the top end, where strong performers are getting faster and producing better work. A preliminary MIT Media Lab study found that heavy reliance on AI assistants can lead to what researchers called “cognitive debt,” where users outsource critical thinking and lose the ability to recall and synthesize what they’ve produced.^[7] Brendan acknowledged this risk directly. He sees his own daughters, 19 and 24, defaulting to LLMs for answers, and he worries about critical thought eroding over time.

“Embrace these new technologies. It’s scary. But your job will evolve. Become that data product owner, become an expert in that data, and be that trusted guide as everybody’s going down the agentic experience.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

The real opportunity for data professionals is to become the people who make sure agents are working with the right information in the right context. That’s a role no LLM can fill on its own. If you’re not sure where to start, audit the data your team’s AI tools depend on. If you can’t quantify how trustworthy that data is, that’s the first problem to solve.

Listen to the full conversation with Brendan Grady on the Data Faces Podcast.

Based on insights from Brendan Grady, EVP and GM of Analytics & AI at Qlik, featured on the Data Faces Podcast.

Share

Frequently asked questions

What is consequence management in the context of data quality?

Consequence management is the idea that data quality was never prioritized because the consequences of bad data were manageable. In a manual world, a wrong number in a spreadsheet could be corrected before it caused real damage. With AI agents making autonomous decisions across multiple steps, errors compound before anyone detects them. Consequence management explains why the stakes around data quality have shifted from recoverable inconvenience to potential business-level damage.

Where does enterprise adoption of agentic AI stand in 2026?

According to Brendan Grady, EVP of Analytics and AI at Qlik, enterprise agentic adoption is in its earliest stages. Customers are experimenting with low-risk use cases in controlled environments, but production-grade agents making real decisions in real business processes are rare. Data quality is the primary blocker. Gartner projects that by 2027, 70% of organizations will adopt modern data quality solutions to support AI initiatives.

What is Qlik’s Trust Score for AI?

Qlik’s Trust Score for AI is a quantifiable measure of whether data is valid, up to date, and representative before it reaches an AI agent or a large language model. It scores data across dimensions including diversity, timeliness, and accuracy, giving decision-makers visibility into data reliability rather than requiring them to take data quality on faith. Qlik has been named a Leader in the Gartner Magic Quadrant for Augmented Data Quality Solutions for seven consecutive years.

What does “dashboards are dead” mean?

Brendan Grady’s declaration that “dashboards are dead” refers to dashboards as a destination, not the data or insights within them. The traditional model of going to a dashboard to draw your own conclusions is being replaced by AI-powered interfaces that proactively recommend actions. Qlik calls this shift decision intelligence. Grady described his own workflow changing from “give me information about my business” to “tell me about my business and what you think I should do.”

What is the Qlik MCP server?

The Qlik MCP server implements the open Model Context Protocol, allowing third-party AI assistants such as Anthropic Claude, Google Gemini, and ChatGPT to access Qlik’s analytical capabilities, with built-in governance and audit trails. Launched in February 2026, it reflects Qlik’s bet on interoperability over platform lock-in. Grady reported that roughly 50% of users now access Qlik’s agentic capabilities through MCP rather than Qlik’s own interface.

What should data professionals do to prepare for the agentic AI era?

Brendan Grady advises data professionals to lean into what they already know best: the data itself. His recommendation is to become data product owners who serve as trusted guides as organizations adopt agentic AI. The people who understand data quality, business context, and organizational nuance will be indispensable because these capabilities are not ones AI agents can replicate on their own.

Podcast highlights

- [0:00] Introduction and welcome at Qlik Connect 2026

- [1:14] Brendan’s first job: Sound of Music tour guide in Salzburg

- [2:04] Lessons learned from the early analytics era

- [3:32] Why data quality has never been fixed

- [4:46] Consequence management in the agentic era

- [6:08] Where enterprise agentic adoption actually stands

- [7:46] Future-proofing against LLM shifts

- [8:24] The analytics engine and unknown unknowns

- [10:29] Structured vs. unstructured data convergence

- [12:04] Hallucinations and the trust problem

- [15:30] Decision intelligence and “dashboards are dead”

- [18:05] Brain outsourcing and the MIT cognitive debt study

- [21:57] MCP server and open standards

- [23:54] Key themes for Qlik in 2026: trust, context, flexibility

- [26:12] Advice for data professionals

- [28:15] Does AI expand the aperture for who can participate in analytics?

About David Sweenor

David Sweenor is the founder and host of the Data Faces Podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “Beyond the AI Hype: What 20% of Companies Get Right.” TinyTechGuides, February 11, 2025. https://tinytechguides.com/blog/beyond-the-ai-hype-what-20-of-companies-get-right/

^[2]Gartner. “Lack of AI-Ready Data Puts AI Projects at Risk.” Gartner Newsroom, February 26, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk

^[3]Sweenor, David. “From ‘AI-Ready’ to AI Reality: Why Actionable Data Strategies Beat Endless Planning.” TinyTechGuides, June 3, 2025. https://tinytechguides.com/blog/from-ai-ready-to-ai-reality-shane-murray-on-data-trust-and-why-action-beats-planning/

^[4]Qlik. “Qlik Releases Trust Score for AI in Qlik Talend Cloud.” Qlik Press Release. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-releases-trust-score-for-ai-in-qlik-talend-cloud

^[5]Qlik. “Qlik Named a Leader in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions.” Qlik Press Release, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-named-a-leader-in-the-2026-gartner-magic-quadrant-for-augmented-data-quality-solutions

^[6]Qlik. “Qlik Brings Agentic Analytics to General Availability and Launches MCP Server for Third-Party Assistants.” Qlik Press Release, February 10, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-brings-agentic-analytics-to-general-availability-and-launches-mcp-server-for-third-party-assistants

^[7]MIT Media Lab. “Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Task.” MIT Media Lab, 2025. https://www.media.mit.edu/publications/your-brain-on-chatgpt/