Somewhere in most large companies, a finance chief has recently asked a lethal question: show me the return on the AI spending. A new survey suggests the room usually goes quiet. Eighty-three per cent of CFOs report that most of their organisations’ artificial intelligence projects have not produced measurable results — a statistic now ricocheting around boardrooms as companies plan their next round of technology budgets.
How the Money Got Ahead of the Measurement
The pattern the survey exposes is sequencing. Under competitive pressure, companies bought platforms and launched pilots before defining what success would look like in financial terms. Pilots were celebrated for usage and novelty; baselines — what the process cost before AI arrived — were rarely recorded, making later proof impossible. Finance leaders, arriving with a spreadsheet’s scepticism, are not necessarily declaring the technology useless; they are declaring the bookkeeping absent. Our AI news coverage has watched this reckoning approach since the first wave of enterprise deployments.
The Anatomy of the Projects That Pay
Contrasting cases are instructive. The AI investments CFOs do certify tend to share four traits: a named business owner accountable for the number; a pain point with a pre-measured cost, such as invoice cycle time or forecast error; a scope small enough to reach production quickly; and savings or revenue tracked by finance itself rather than estimated by the project team. Research organisations including Gartner have converged on the same observation — deployment discipline predicts AI value better than model choice does.
Some Value Hides From the Ledger
Balance demands acknowledging the measurement trap. Benefits that manifest as avoided events — the fraud not committed, the outage foreseen, the customer who did not churn — are real but resist attribution. A CFO’s “no measurable result” can therefore mean two different things: no result, or no ruler. Mature organisations respond by instrumenting workflows before automating them, so that both kinds of value leave evidence.
Vendors Face a New Kind of Contract
The survey’s pressure lands on the supply side too. Enterprise AI vendors built their sales motion on demonstrations and pilot enthusiasm; CFO-led procurement is replacing it with baselines, milestones and outcome-linked pricing. Contracts increasingly include measurement clauses — agreed metrics, review points, and fees that flex with delivered results. Vendors confident in their products advertise this as differentiation; those who thrived on ambiguity are discovering that the buyer’s spreadsheet has a longer memory than the sales cycle.
Consolidation is the predictable sequel. Companies that ran a dozen overlapping pilots will standardise on the few platforms that produced evidence, and the long tail of undifferentiated tools will be acquired, absorbed or abandoned. None of this diminishes the technology’s trajectory — if anything, the discipline is what converts a promising capability into durable infrastructure. The CFOs behind the 83% figure are not AI’s opponents. They are its admissions committee, and the standard they are setting — prove it, in numbers, on a deadline — is the one every mature technology eventually has to meet.
Sorting, Not Stopping
The likely consequence is a great sorting of corporate AI portfolios: programmes with owners, baselines and evidence attract the next budget round, while demonstration projects quietly expire. That is not the failure of enterprise AI; it is its graduation from fashion to infrastructure, a passage every general-purpose technology has made under finance’s supervision. The 83% headline will shrink as the discipline spreads — and the CFOs delivering it may turn out to have been AI’s most useful critics. More business and AI reporting is on our homepage.
