Most enterprise AI programs have a data problem, but according to Dr. Stefaan Verhulst, co-founder of the Governance Laboratory (GovLab) at New York University, the deeper issue is not the data itself. It is the absence of the right questions to guide it. In a recent conversation with Perficient, Dr. Verhulst challenged a foundational assumption driving most AI investment: that more data, better models, and stronger governance are sufficient. Without question clarity, he argues, none of it delivers the decisions that actually matter.
1. Problem Definition Is a Discipline, Not a Starting Point
Most teams arrive at transformation initiatives with a solution already in mind. Dr. Verhulst's work with government innovation labs revealed that skipping rigorous problem framing is one of the primary drivers of failed programs. The pattern holds equally in the private sector.
"99% of people came to us with the solution they had in mind, but not with a problem.”
He advocates for structured methodologies that force teams to articulate the challenge before selecting tools or technologies, what he calls "do camps" focused entirely on problem definition.
Why it matters: In enterprise AI, this pattern is expensive. Teams that skip problem framing risk building models that optimize for the wrong outcomes and investing in data infrastructure that maps to no meaningful decision. Defining the problem first is not a delay. It is a forcing function for efficiency and smarter resource allocation.
2. The Value of Data Depends on the Value of the Question
Dr. Verhulst makes a well-grounded argument: the reason most data initiatives underdeliver is not a shortage of data. It is a shortage of good questions. Organizations accumulate vast datasets but rarely define in advance what decisions those datasets are meant to support.
"The value of data depends on the value of the question... as a society, we're actually really badly organized in terms of formulating questions.”
This insight led him to develop what he calls a "science of questions," a discipline as rigorous as data science but focused on question formulation, taxonomy, and sequencing before data is collected or analyzed.
Why it matters: Forrester research consistently identifies unclear business objectives as one of the top barriers to AI value realization. Dr. Verhulst's framework suggests that question clarity, not just data quality, is the missing variable in most enterprise AI programs. He applies a practical taxonomy to this work: descriptive questions establish a baseline, diagnostic questions identify causes, predictive questions model forward scenarios, and prescriptive questions drive action. Each level depends on the one before it, and skipping steps is a common source of analytical failure.
"We have data science. But we don't have question science.”
3. Data Literacy Alone Is No Longer Enough
Most enterprise organizations have invested meaningfully in data literacy programs. Dr. Verhulst argues that this investment is incomplete without a parallel commitment to what he calls "question literacy," the organizational capability to formulate precise, prioritized, and decision-relevant questions before reaching for data or tools.
"Data literacy, without question literacy, from my perspective doesn't really get you the value, doesn't get you the answers you need to make better decisions.”
Why it matters: As enterprises scale AI programs from pilots to enterprise-wide deployment, question formulation becomes a core organizational competency. The World Economic Forum has identified decision-making capability gaps as a systemic risk in AI adoption, and question literacy is a direct response to that gap. Training programs, hiring criteria, and team structures should reflect this shift.
4. A Data Winter Is Arriving at the Worst Possible Time
Just as AI demand for high-quality data has reached an all-time high, access to that data is quietly declining. Dr. Verhulst describes this convergence as a "data winter," driven by platform closures, rising data costs, and a broader pullback from open data sharing as organizations recognize the commercial value of what they hold.
"We are witnessing a decline in having access to data... many that had data and would provide it in the past have become convinced that they should hold it and use it for their own internal purposes.”
Why it matters: For enterprises building AI pipelines, this is a structural risk that technical investment alone cannot solve. IDC projects global AI spending will exceed $630 billion by 2028, but that investment is increasingly vulnerable to data access constraints that most organizations have not yet accounted for in their roadmaps.
5. Data Stewardship Is the Missing Function in Your Organization
Most enterprises have no designated role responsible for connecting data supply to decision demand. Dr. Verhulst calls this the "missing profession," the data steward who understands what questions matter, what data is needed, where it resides, and how to access it responsibly.
"Who in your organization has that data stewardship function?... having that clearly figured out and also having investment in upskilling those functions, especially now at a time of AI, would be a wise thing to do.”
This gap became starkly visible during the COVID-19 pandemic, when researchers needed rapid access to private-sector data but could find no one within those organizations with the authority to act on it.
Why it matters: Gartner has identified data stewardship as a critical capability gap in enterprise AI governance. As data sharing becomes a strategic function and not just an operational one, organizations that invest in formal stewardship roles will be better positioned to move faster, collaborate more effectively, and govern AI responsibly.
For enterprise leaders, Dr. Verhulst's message is a useful corrective to the prevailing instinct to acquire more data and build more models. The organizations that will get the most from AI are not necessarily those with the largest datasets. They are the ones disciplined enough to ask better questions first.
Explore how Perficient delivers AI-First Data Platforms solutions.
These insights come from Stefaan Verhulst’s appearance on Perficient’s What If? So What? podcast.