The question boards are asking has shifted from “should we use AI?” to “is our investment in AI producing measurable value?” It is the right question, and for many organisations the honest answer is: nobody checked.
Where it pays
In finance functions, the dependable returns cluster in unglamorous places. Document-heavy processing – invoices, statements, reconciliations – where volume is high and judgement is low. First-draft work: management commentary, variance narratives, board-pack assembly, where a system produces the draft and a professional owns the conclusion. And anomaly detection across transactions, where software attention is cheap and human attention is scarce.
What these have in common is a clear baseline. You know what the month-end close costs today, so you can measure what it costs after. Value that cannot be measured against a baseline is a story, not a return.
Value that cannot be measured against a baseline is a story, not a return.
Where it quietly adds risk
The failures cluster too. Systems given judgement rather than drafting – approving, valuing, deciding – without a human owning the outcome. Data leaving the organisation through tools nobody assessed. Automation layered onto a broken process, which simply produces the wrong answer faster. And “pilot sprawl”: a dozen experiments, none measured, none retired.
The discipline that governs any other capital allocation applies here unchanged: a stated use case, an expected return, a measurement point, and the willingness to stop. AI does not exempt an investment from evidence. It is one more proposition that has to earn its keep – and the organisations getting real advantage are the ones that insist it does.
