Find the changed work before you count the return.

You can tell AI is paying off when access has turned into changed work, and changed work into a result the business measures. That chain is how we read every AI investment.

Seats, logins, and training hours sit on the first link. They’re worth reporting, and none of them is a result on its own.

The chain, one link at a time

Access

Tools bought, seats granted, training completed. Worth reporting, never an outcome on its own.

Changed work

A specific workflow runs differently, with a named owner, and quality holds.

Result

Something the business measures moved: revenue or cost, risk, customer or employee experience, or capacity. A deliberate stop counts too: capital returned.

Put each claim at the highest link its evidence supports, and say how firm it is. A result that only moved at the same time as the change still belongs in the report, labelled as that. “Is it live?” and “Is it paying back?” are different questions, read at different times. Nobody should read a result in week two.

Results also need someone who owns them. KPMG found that where the CEO is accountable for AI outcomes, 14 percent of organizations report established ROI, against 4 percent where the CEO is less or not accountable.

So: for each AI investment, write down the result it was meant to move, the workflow that has to change first, and who owns each. Report every claim at the link its evidence reaches, and set the date you read the result.

Sources checked October 2026