Behind the confident keynote language about artificial intelligence sits a colder assessment from the people who sign the cheques. Survey evidence from finance leaders indicates that around 83 per cent of chief financial officers report their AI projects have yet to produce measurable results — not necessarily no activity, no pilots or no enthusiasm, but no gains they can isolate, quantify and defend in a budget review. For a technology absorbing capital at historic speed, that gap between spending and demonstrated return is now the central argument in enterprise AI.
It is important to read the figure precisely. CFOs are not typically claiming the tools are useless; many see genuine promise in specific workflows. Their complaint is measurement: benefits are asserted by vendors and project sponsors in language — efficiency, insight, transformation — that finance cannot reconcile against costs that are painfully concrete, from licences and cloud compute to the staff time consumed by implementation. When the invoice is exact and the return is a narrative, scepticism is a professional duty. More business analysis appears on our homepage.
Why AI returns are so hard to isolate
Several structural problems recur across companies. Pilots are launched without a recorded baseline, so afterwards nobody can prove what changed. Benefits leak: an hour saved in one team dissolves into general busyness instead of appearing as output, cost reduction or revenue. Data quality, cheerfully assumed in the sales process, turns out to need months of unglamorous cleaning before models perform. And integration — connecting new tools to the ageing systems where value actually lives — routinely costs more than the AI itself.
There is also a portfolio problem. Organisations scattered small experiments across every department, each too small to move a measurable number and too numerous to govern. Gartner’s finance research has repeatedly urged leaders to tie technology spending to a small set of owned, quantified outcomes, while McKinsey’s AI work finds value concentrating in companies that redesign workflows around the technology rather than bolting it onto existing habits. The pattern is consistent: the constraint is rarely the model, and almost always the organisation around it.
What separates the successes from the stalls
The minority reporting real returns share recognisable habits. They chose narrow, high-volume processes where success is countable — document processing, customer-contact summarisation, forecasting support, code assistance — and they measured before and after with the finance team inside the project from day one, not invited at the end to validate a success story. They also budgeted for the unglamorous majority of the work: data preparation, staff training, process redesign and the temporary productivity dip while people learn.
Governance distinguishes them too. Successful programmes kill weak pilots early, publish their criteria, and scale only what clears them — treating AI spending like any other capital allocation rather than an innovation tax exempt from scrutiny. Their CFOs are not opponents of the technology; they are its most useful allies, because a return that survives finance’s questions can be scaled with confidence and defended to the board without embarrassment when the hype cycle turns.
What this means for the next budget round
The likely consequence is not a collapse in AI spending but a change in its shape: fewer scattered pilots, larger commitments to proven use cases, and contracts scrutinised for measurable commitments rather than feature lists. Vendors should expect harder questions about baselines, adoption support and total cost, and internal sponsors should expect to arrive with a finance partner already persuaded. Companies that built measurement discipline early will find this environment congenial; those that spent two years on unmeasured enthusiasm face an awkward reckoning.
None of this makes the 83 per cent figure an obituary for enterprise AI. Electricity and enterprise software both passed through periods where spending vastly outran demonstrated productivity — the famous productivity paradox — before organisational change let the gains surface. The CFOs‘ message is better read as a deadline than a verdict: the experimental grace period is ending, and the next phase belongs to projects that can show their working. More coverage sits in our Blog category.
The bottom line
A technology can be genuinely important and still fail individual business cases; both things are true across most portfolios today. The finance chiefs reporting no measurable results are describing their own projects honestly, and the companies listening hardest are the ones most likely to join the successful minority.
AI’s enterprise story will ultimately be written in the same place every other technology’s was — not in the demo, but in the accounts. The organisations that understood that earliest are already pulling away.
