The world’s finance chiefs have delivered their verdict on the AI spending boom, and it is more confession than celebration. Survey after survey of CFOs finds the same pattern: the large majority of enterprise AI projects cannot yet demonstrate measurable financial return, even as budgets for them continue to climb. Research houses such as McKinsey’s AI practice have documented the adoption surge in detail; the returns column, our business coverage notes, has been slower to fill in.
This is not a story about AI failing. It is a story about how the spending was approved — on narratives, competitive fear and board pressure — and about a measurement gap that finance departments themselves now admit.
Approved on Stories, Audited on Numbers
The typical enterprise AI programme of the past three years began with a demonstration, not a baseline. Vendors showed what the tools could do; executives extrapolated; budgets followed. What rarely happened was the unglamorous preliminary: measuring how long the target process currently took, what it cost, and how often it erred. Without that baseline, every subsequent claim of impact is unfalsifiable. Productivity that might have arrived anyway is indistinguishable from productivity the AI delivered.
CFOs, to their credit, are saying this out loud. The surveys show finance leaders distinguishing sharply between a minority of instrumented projects — where baselines existed and returns can be shown — and the long tail of deployments justified by strategic language. The minority, not coincidentally, are the ones reporting real numbers.
Where the Returns Are Hiding
The pattern of genuine results is instructive. Returns concentrate where AI is pointed at a costed, bounded process: document processing with known error rates, customer-contact deflection with known per-contact costs, code assistance with known cycle times. They evaporate where the promise is diffuse — “transformation”, “augmentation”, enterprise-wide copilots whose benefit is measured in sentiment surveys.
There is also a timing defence that honest CFOs offer: infrastructure spending precedes returns in every technology cycle, and some of today’s unmeasurable platforms are the substrate for next year’s measurable products. The counter-argument, gaining ground in boardrooms, is that this defence has now been running for three budget cycles.
The Discipline Arriving Now
The correction visible in the newest surveys is procedural: stage-gated funding, baseline instrumentation before deployment, benefits tracked in the same ledger discipline as any capital project, and kill criteria agreed in advance. AI is being moved, in other words, from the innovation budget to the capital-allocation process — a demotion in glamour and a promotion in credibility.
What the Honest Minority Did Differently
The survey’s most actionable finding concerns the minority of companies reporting clear returns. Their programmes share a profile: narrow initial scope, a named business owner rather than a technology owner, baselines logged before deployment, and benefits audited by finance rather than asserted by the project team. Notably, their spending is not smaller — it is simply accounted for like money.
Several CFOs in the survey describe the same turning point: the meeting where AI proposals were first subjected to the questions every factory or fleet proposal answers automatically. What does it replace? What does it cost per unit of work? When do we check? The technology did not change in that meeting. The governance did — and the returns, where they exist, date from it.
For boards reading the latest results, the conclusion is bracing but useful: the AI returns problem is largely a management problem, and management problems, unlike technology problems, can be fixed by decision.
For investors, the signal to watch is language. Companies that report AI returns with baselines, denominators and timeframes are telling you something about management quality that transcends AI entirely. The boom’s first phase rewarded spending announcements; its second phase will reward accounting. Our business section will track which companies make that transition credibly.
