There is a version of this story that most executives have already heard: AI tools are producing more information than ever, much of it converging toward the same conclusions. What is less discussed is what’s missing from that convergence: the experts who can verify it, and the customers whose behavior it claims to predict. For investment committees underwriting risk and corporate strategy teams allocating capital on high-stakes bets, that absence isn’t an abstract concern. It shows up directly in consumer and client-facing markets, where misreading a shifting landscape has immediate and measurable consequences for enterprise value.

The gap between what models know and what markets are doing

AI adoption has accelerated sharply. A global study of 1,594 large companies conducted by Dialectica and the Athens University of Economics and Business found that 84% now use AI technologies, up from 70% just two years ago. The tools are in place. The results are not.

Only 5% of companies in the study have reached what the research calls “AI Transformed” status, meaning AI has produced measurable, strategic business impact. Meanwhile, 30% have adopted AI without seeing a single concrete business benefit. And when success does occur, it is almost always narrow: 64% of companies that report any win cite productivity gains, meaning time saved, not decisions improved or growth accelerated.

McKinsey’s State of AI report corroborates the pattern. Most organizations have yet to see organization-wide, bottom-line impact from Gen AI, and only 1% of company executives describe their rollouts as mature. A RAND Corporation meta-analysis of 65 documented enterprise AI projects found that more than 80% failed to deliver their promised business value, a figure Gartner confirmed in April 2026 with comparable failure rates across IT and operations teams.

The decision-support gap is particularly telling. Improved decision-making was among the most anticipated outcomes executives expected from AI investment, yet only 37% of companies in the Dialectica study report having achieved it. That is not a technology failure. It is a verification failure: most organizations-built AI pipelines that no expert ever stress-tested against what’s happening in the market. And as AI output becomes harder to interrogate, the judgment to challenge it is becoming one of the most valuable and necessary capabilities an organization can have.

According to a recent survey from my company, Prosper Insights & Analytics , four in ten consumers name the need for human oversight as their top concern about AI. That figure is virtually unchanged among executives who already use the technology daily.

And when it comes to consequential decisions in banking, healthcare, and travel, more than three quarters of consumers still prefer a live person over an AI program, a preference that holds across every service category tested. These are the kinds of signals that most retrospective models would have flagged only after the window to act had already closed.

This is where synthetic consensus does its most damage to investment committees and corporate strategy teams: not in the research phase, but in the planning phase, when resource allocation decisions, pricing calls, and market entry bets are made against a picture of the consumer that stopped being accurate some time ago. By the time the discrepancy surfaces in the numbers, the decision has already been made.

The question leaders aren’t asking

Most organizations have become reasonably good at asking whether their data is current. Fewer are asking whether it’s verified: whether an expert with direct, current exposure to that market has actually looked at it, challenged it, and confirmed it holds up against what customers are saying and doing right now.

Gartner flagged the same gap from a governance angle at its Data and Analytics Summit in June 2026, warning that as AI agents take on more strategic and operational decisions without adequate oversight structures, organizations face mounting legal, operational, and reputational exposure. The issue is not that the tools are making decisions. It is that no one has defined the process for checking them.

A model updated daily can still be structurally wrong about a category if it draws from sources that share the same blind spot: they document what consumers or clients say in aggregate rather than what individual customers actually do, and what they intend rather than what they will actually choose under real conditions. Closing that gap takes two things most AI systems skip: an expert who knows the category well enough to flag when a trend has already turned, and direct access to the customer’s own read on why they’re behaving that way

Enhanced customer experience was one of the most cited strategic goals for AI investment in the Dialectica research, yet it remains one of the weakest areas of actual achievement, with only 44% of companies reporting meaningful progress. The expectation existed. The insight to act on it did not.

As George Tsarouchas, Founder and CEO of Dialectica , puts it, “The organizations struggling to close the gap aren’t short on data. They’re short on verified intelligence, the kind that comes from people who are actually inside the markets they’re trying to understand. AI can surface patterns, but it takes human judgment to know which patterns are real and which ones are already obsolete.”

Rethinking what “good research” looks like

The companies that have crossed into genuine AI transformation share a pattern that goes beyond tool selection. Dialectica’s study found that AI-transformed companies are more than five times as likely to have built proprietary, in-house AI systems compared to companies still in the adoption phase. And among those planning future investment, custom AI infrastructure now outpaces off-the-shelf solutions by nearly two to one.

The leaders aren’t simply buying better tools, they’re building systems around inputs off-the-shelf AI can’t generate on its own: practitioners with frontline expertise, and customers who can speak to their own decisions in their own words.

In practice, this means treating expert interviews and customer conversations not as a supplement to the research process but as a quality control mechanism within it. Before any major strategic decision, the question should be whether anyone on the team has spoken recently to an expert with firsthand exposure to that market, and separately, to the customers the decision effects. When either signal conflicts with the aggregated data, it deserves serious weight; often more than the model’s output.

There is institutional comfort in a well-formatted report. There is also real friction in building the kind of access; to vetted experts, to customers willing to talk candidly, that makes human verification possible at speed. But the organizations that have made that investment consistently find the same thing: the confidence gap narrows. The hesitation between having information and being willing to act on it shrinks when decisions are grounded in firsthand answers, not statistical inference

Tsarouchas puts it plainly: “AI has made information even more accessible. What it hasn’t changed is the value of knowing which information to trust. If anything, that has become harder to find and more important than ever.”

What this means for the next decision on your desk

Consumer markets are moving faster than the models designed to track them. The executives navigating this well share a common orientation: they are less interested in having more data and more interested in having insights they can trust.

That shift from volume to insights and validity is not a technology upgrade but a strategic one, and it is available to any organization willing to make it. The tools already exist. What is still missing, for most, is an expert to catch what those tools cannot see.

Disclosure: The consumer sentiment study referenced above was conducted by my company, Prosper Insights & Analytics . This is the same dataset used by the National Retail Federation, and available from Amazon Web Services, Databricks, and the London Stock Exchange Group for economic benchmarking.