The AI ROI deadline passed. The budget survived; the question got harder.
In February, 71% of IT leaders said their AI budget would be cut or frozen if it hadn’t proved itself by mid-2026. Mid-2026 came and the money mostly stayed — but the pressure to show a return moved from a date to a standing question. What that means for the next review, and why the answer is not the next platform.
Enterprise AI budgets largely survived the mid-2026 reckoning their own owners predicted, and the pressure to prove a return did not go away with it — it turned from a deadline into a standing question. The practical consequence: you no longer get to defer the ROI answer to a date, and buying the next platform is now the least convincing version of that answer.
What happened to the mid-2026 AI ROI deadline?
It passed without the reckoning. In February 2026, 71% of CIOs said AI budgets were likely to be cut or frozen if targets weren’t hit by mid-2026 (Harris Poll for Dataiku, n=600 CIOs, global, published 12 February 2026). By August, CIO.com’s reporting on the aftermath (Grant Gross, 18 August 2026) described a reprieve rather than a reckoning — with scrutiny tightening, not relaxing.
Two caveats before these figures go anywhere near a board pack: the samples are US and global, never European or Czech, and Dataiku sells an AI platform, which makes some of its findings self-serving. Where a figure cuts against the publisher’s own interest, it is worth more — and one of them does. We will come to it.
Why do AI budgets keep growing without a proven return?
Because the spending decision and the proving decision sit in different places, and only one of them has a deadline. The gap is visible in the numbers CIO.com assembled: 71% of organisations plan to increase AI spending this year, while only 27% expect a near-term return (TEKsystems). Separately, 40% of IT leaders report that only between 30% and 70% of their AI initiatives are meeting ROI goals (CIO.com *State of the CIO*), and nearly half of organisations have delayed, stopped or scaled back AI projects on budget grounds (KPMG).
Seven in ten spending more; fewer than three in ten expecting anything back soon. That gap is not a forecasting error. It is a bet that the return will arrive on its own once enough capability is bought — and that is precisely the assumption the last two years have tested.
Do CIOs regret their AI platform decisions?
Yes, and more than most vendors would like published. Nearly three in four CIOs regret a major AI vendor or platform decision made in the previous 18 months, and 62% say they have been directly challenged by their CEO over that selection (same Harris Poll for Dataiku, reported by CIO Dive, 12 February 2026).
The second figure is the one to keep. A platform vendor publishing "you regret your platform choice" has an obvious motive; a platform vendor publishing "your CEO is already questioning you" has rather less. The CEO-challenge number cuts against the publisher’s interest, which is exactly why it is the sturdier of the two.
It is also the most useful sentence in the dataset for anyone deciding what to do next. Your buyer already knows they bought wrong once. That changes what a credible proposal looks like.
Why buying the next platform doesn’t close the gap
Because the gap was never a capability gap. If three-quarters of the market regrets a purchase made in the last 18 months, the constraint is not the quality of what is available to buy — it is what happens after the invoice.
The pattern is old and predates AI: a tool arrives, a pilot succeeds with the people who volunteered for it, and the rollout stops at the edge of that group. Nobody redesigned the work around it. Nobody made a named person accountable for a workflow. The licences renew anyway, and the following year the proposed remedy is a better tool.
The uncomfortable arithmetic is that the second purchase inherits every unfixed condition of the first, and adds a migration.
What actually closes the ROI gap
Two moves, in this order.
Find what you already own and are not using. Not as a cost-cutting exercise — as the funding source. Most organisations are carrying idle seats, over-provisioned tiers, duplicated tools and orphaned subscriptions, and the money is recoverable at renewal. That number is also the first honest ROI figure in the whole programme, because it is traceable to invoice lines rather than projected.
Spend part of it on getting people to use what remains. On their real workflows, with champions drawn from your own staff, measured in weekly active use on real work rather than course attendance. This is the half that turns the remaining spend into a return instead of merely shrinking it.
Done in that order, the adoption work does not need a new budget line — which removes the argument that stalls it. It also produces the thing the February survey says your CEO is already asking for: a number with a source attached.
What to bring to your next budget review
- The spend-versus-return gap, in your own numbers. What went up this year, and what you can currently evidence coming back.
- A utilisation figure produced independently of the vendor whose renewal is at stake. If the only usage data you have came from your supplier, you have their account of your business.
- One recoverable euro figure, traced to invoice lines — small and real beats large and modelled, every time.
- A named owner per workflow, not per tool. Tools do not adopt themselves and licences do not have accountability.
An extension is not an answer. The budget surviving means the question comes back next quarter with a sharper edge — and the one thing that will not answer it is another platform.
Three-quarters of your peers already regret the last major AI purchase. That is an argument for finishing one, not for starting another.