Corporate enthusiasm for artificial intelligence has met its auditor. A striking 83 per cent of chief financial officers say that most of their companies’ AI projects have not produced measurable results, according to survey findings that capture a widening gap between AI spending and AI returns. After two years of boardroom urgency, hastily approved pilots and soaring vendor bills, finance chiefs are asking the question that defines this phase of the AI boom: where, exactly, is the money?
How the spending outran the scoreboard
The pattern is now familiar across industries. Under competitive pressure, companies launched chatbots, copilots and automation pilots in parallel, often several departments at a time, frequently without agreed baselines against which to judge success. Projects were declared promising on the strength of demonstrations rather than outcomes, licences were bought for headcounts that never adopted the tools, and cloud and data costs crept upwards in the background. When CFOs came to consolidate the picture, they found portfolios full of activity and thin on evidence: few projects could point to saved hours, reduced error rates, faster cycles, new revenue or avoided costs. Analysis from firms such as Gartner has repeatedly warned that a large share of generative AI projects would be abandoned after proof of concept, citing poor data quality, unclear value and inadequate risk controls.
What separates the successes
The same surveys that expose the failures also illuminate the minority of projects that pay. Winners tend to share a profile: a specific, painful, measurable problem; clean data connected before the model arrives; workflow redesign rather than simple tool deployment; and an owner accountable for a number, not a narrative. Successful finance teams insist on baselines, run controlled rollouts, and count total cost — licences, integration, training, oversight — against benefits with the same rigour they would apply to any capital project. Reporting on enterprise technology by The Wall Street Journal has documented this shift from experimentation to accountability as CFOs take a firmer hand in AI governance.
The ROI problem is partly a measurement problem
Some of the 83 per cent reflects genuine waste; some reflects value that is real but hard to attribute, such as better decisions, avoided mistakes or employee time scattered in small savings across thousands of tasks. CFOs are responding by narrowing portfolios: fewer, larger bets; kill criteria agreed in advance; and staged funding that releases money only when milestones in adoption and outcome are met. Vendors, in turn, are being pressed for usage data and outcome-based pricing rather than per-seat promises. None of this signals that AI is a bubble about to vanish from the enterprise. It signals maturation — the moment a technology stops being bought on fear of missing out and starts being bought like everything else in the budget: on evidence.
Boards are changing the questions
The shift is visible in boardrooms. Directors who once asked how much AI a company had deployed now ask which three processes it has measurably improved, what the tools cost per active user, and which projects were stopped and why. That last question matters culturally: in the first wave, cancelling an AI project felt like falling behind, so weak pilots lingered. Finance chiefs are normalising the kill decision, treating it as portfolio management rather than failure. Analysts expect the next round of surveys to show a smaller total number of projects but a higher share reporting results — the signature of a technology moving from spectacle to infrastructure inside the enterprise.
What comes next
For employees, the new discipline may actually help: projects that survive scrutiny tend to be the ones that remove drudgery rather than generate slideware. For smaller firms, the lesson is encouraging — you do not need the largest budget to win, only a clear problem and an honest scoreboard. The 83 per cent figure is less an indictment of AI than of how it was purchased. The organisations that slow down just enough to measure properly are, quietly, the ones now pulling ahead. Follow continuing business coverage in our business section and at News Headlines Daily.
