The gap between spending and readiness
Dun and Bradstreet surveyed 10,000 businesses across 32 countries and found that only 8 percent describe their data as fully ready to support AI. Another 33 percent call their data mostly ready, and 44 percent, the largest single group by a wide margin, call it only partially ready, with the remainder reporting even less confidence than that. That distribution means more than half of surveyed businesses are running active AI initiatives on data they themselves would not describe as fully reliable, which reframes a lot of disappointing AI results across the industry as a predictable outcome of that gap rather than a mysterious surprise that nobody could have anticipated in advance.
The return figures attached to each readiness tier make the stakes concrete. Among companies with fully AI-ready data, 81 percent report strong AI returns. Among companies with only partially ready data, that figure drops to 12 percent. That gap describes a cliff rather than a gradual decline across readiness tiers, and most enterprises responding to this survey are standing somewhere below that cliff edge rather than safely above it, closer to the 44 percent calling their data only partially ready than to the 8 percent who can honestly call it fully ready.
Spending is accelerating in the wrong places
Gary Kotovets, Dun and Bradstreet's chief data and analytics officer, says the next phase of AI adoption will be defined less by model access than by the data foundation underneath it, a point this survey backs with hard numbers rather than opinion alone. Businesses in the Americas illustrate the problem precisely: they are among the regions most likely to be increasing AI spending right now, actively growing their budgets quarter over quarter with little sign of slowing down, yet they report the second-lowest rates of both strong ROI and ready or fully ready data among all the regions this survey covered, a mismatch that should concern any finance committee signing off on next year's growth in that line item without first asking where the underlying data actually stands today.
That combination describes an investment pattern working against itself in a fairly visible way once the two figures are placed side by side. Enterprises accelerating budget into AI without first closing the data-readiness gap are effectively spending faster into a bottleneck they have already been warned about by their own survey responses. Kotovets's reading of the data is direct: organizations whose data is ready for AI are seeing broad returns at a much higher rate than everyone else, and increasing spend on top of an unready data foundation does not substitute for the harder, less visible work of actually closing that gap before the next round of budget gets approved.
This diagnosis is already widely accepted
What makes this finding more actionable than most survey results is that it is not a contested diagnosis inside the industry. Collibra's research, cited alongside Dun and Bradstreet's, found that 72 percent of AI decision-makers already identify poor data foundations as the root cause of their enterprise AI shortfalls. That is close to consensus among the people actually running these programs, which removes the usual excuse that leaders simply did not know where the problem was.
The honest question, then, is why so few organizations have closed the gap despite recognizing it. Part of the answer is that data readiness work is unglamorous and hard to attach to a single project's budget, the same dynamic that makes Heineken's narrowly scoped, decision-specific approach to data readiness notable rather than obvious. Most enterprises default to generic data-quality initiatives instead, which this survey's results suggest do not move the readiness needle the way scoped, decision-tied data work does.
Scaled agents remain rare for the same reason
A Google Cloud and MIT Technology Review Insights report from August found that only 10 percent of companies have scaled AI agents widely across their operations. That figure sits suspiciously close to Dun and Bradstreet's 8 percent fully-ready-data figure, and the overlap between them is not coincidental once the two surveys are read side by side rather than filed away as two unrelated data points from two unrelated vendors. Agents operating with meaningful autonomy need reliable, current, and well-connected data to make decisions safely without constant human correction at every step, and an enterprise that has not solved data readiness for its simpler, human-reviewed AI use cases has no realistic path to scaling autonomous agents responsibly, no matter how capable the underlying model it has licensed actually is.
This is worth stating plainly because so much of the current enterprise AI conversation treats agent deployment as a separate, more advanced initiative sitting apart from ordinary data readiness work, as though it were a later, more sophisticated phase rather than a direct continuation of the same unfinished project. The numbers argue against that framing fairly clearly. The organizations that will actually be able to scale agents in 2027 are overwhelmingly going to be the same organizations that closed their data-readiness gap first, well before they went shopping for an agent platform, rather than the ones that simply bought the most capable platform on the market and hoped the data underneath it would somehow sort itself out.
What this means for the transformation roadmap
For CIOs sequencing a 2027 transformation roadmap, this survey argues for treating data readiness as a gating investment rather than a parallel workstream. Funding a flashy agent pilot before closing the readiness gap is backwards, given that the return difference between ready and partially ready data is a seven-fold gap, not a marginal one. Budget committees approving next year's AI spend should ask for a data-readiness assessment before approving the agent project that depends on it.
The governance implication pairs directly with the cost-pruning discipline enterprises are already being asked to build elsewhere. A project built on partially ready data is a project that should carry a lower confidence score in any portfolio review, regardless of how promising its use case sounds. Enterprises that fund data readiness first, even if it delays a visible AI launch by a quarter, are the ones most likely to be in the 81 percent reporting strong returns rather than the 12 percent explaining why they are not.



