An operations team finishes its weekly report faster with an AI assistant. The investment proposal multiplies the hours saved by staff cost and presents the total as a return. Yet the payroll is unchanged, the team completes the same work and nobody has identified an expense that can now be avoided. The report may be easier to prepare. The financial claim still needs an explanation.
Time saved is evidence about a task. To support an AI investment, it needs a route into lower spending, more valuable work completed or a loss avoided. Deloitte and BCG/Consumer Goods Forum survey evidence from retail and consumer goods in 2026 points to difficulties measuring AI returns. Those reported measurement gaps do not tell us whether a particular assistant earns its cost. That requires following the saving into the operating decision.
Ask which spending will actually fall
A salary divided into hours gives a way to value staff effort. It does not establish that saving an hour reduces the salary bill. A spending claim needs a specific expense that the business can remove: paid overtime, contracted support or recruitment that would otherwise be necessary. The budget owner should identify when that change can happen and what work will still need covering.
Small savings scattered across a team can be especially difficult to convert. A few free minutes between requests may make the working day less pressured without removing a shift or a supplier invoice. Record that capacity honestly. Where a future hire could be avoided, show the workload that would have required it and distinguish a forecast from spending that has already fallen. An unfilled vacancy with no approved need is a weak basis for claiming an avoided cost.
Find the work that can use the freed time
The next test is whether output rises. Name the work waiting to be done and ask whether the people whose time is released can do it. Skills, customer demand and another team's capacity may limit the answer. Saving preparation time has little effect on completed work if every item still waits for the same scarce approval. Count work accepted at the required quality, rather than drafts produced by the assistant.
Extra output needs a financial explanation too. More completed orders can add value only if they can be fulfilled economically; more internal reports may add none. Keep the costs of the extra work alongside its benefit. Where the same released hour supports additional work, claiming it again as a payroll saving would count the benefit twice.
Trace the value of an earlier commitment
A third test concerns timing: does a commitment get made earlier, with better evidence? Preparation can be faster while the decision still waits for a meeting or an authorised person's confirmation. Measure when checked information reaches that person and when they act. An earlier report alone leaves the claimed benefit unresolved.
For avoided loss, identify what the earlier decision changes and the evidence for its cost. Review comparable past work to establish how often delay mattered and what happened because of it. A projected reduction in loss should remain labelled as an estimate; a period without an incident cannot establish that the assistant prevented one. Nor should an earlier commitment be rewarded if essential checks have simply been skipped.
| Conversion | The test question | What to measure |
|---|---|---|
| Spending falls | Which expense can the budget owner remove? | Overtime, supplier costs or an evidenced hiring need avoided; when spending changes. |
| More valuable work gets done | Can the team use freed time to complete valuable work? | Extra work accepted at the required quality; its benefit and delivery cost. |
| A commitment arrives earlier | Does checked information lead to an earlier action? | When information is checked, when the owner acts and the cost of delay. |
Include checking in the measurement
Measure the whole task before and after the change, through to an answer someone can act on. Include finding source records, checking them, correcting errors and obtaining missing confirmations. Separate staff effort from elapsed waiting time. If checking moves from the preparer to a manager, include both people's work. A faster first draft can consume its entire saving in verification.
Use comparable requests over a stated period and record how difficult they were. Keep incomplete answers and failed attempts in the measurement. Report the number of tasks, the checking method and what the comparison excludes, then distinguish trial findings from a forecast of wider use. Running and support costs remain part of the investment, even when net staff effort falls.
Before approving wider spending, ask the operating owner and budget owner to agree which benefit they will verify, how they will measure it and by when. Approve where the evidence supports it, defer where a specific dependency remains untested, or reject where the saving has no credible route to value. Available capacity may still be worth buying, but the decision should name what the business expects to gain from it.