What finance teams should count as AI return

The organizations reporting credible AI returns in this year's research share one trait more than any technology choice: they defined what would count as a return before deployment, not after. IBM's framing of average-versus-best-in-class ROI, and Wharton's adoption research on productivity, profitability, and throughput as distinct metrics, both point the same direction — pick the unit first.

In practice that means choosing, in advance, whether a program is being measured on cost avoided, revenue lifted, throughput gained, or risk reduced, and assigning a named owner who is accountable for reporting it — not a vendor dashboard defaulting to whichever metric makes the deployment look best. This site's glossary entry on cost per outcome is the sharpest available unit for this: it prices the business result the organization actually wants, not a proxy like tokens or seats.

Finance teams that skip this step end up reconciling vendor-reported "AI ROI" figures after the fact, against assumptions they never chose. The fix is procedural, not technical — and it is the one recommendation nearly every source in this site's research index converges on independently.