Record a baseline you can explain
Measure a representative set of completed tasks. Include the time spent checking, correcting, and chasing missing information. Note changes in volume or task complexity. An estimate from memory is a useful starting hypothesis, not a measured result.
Name the type of value
Time released, costs removed, additional gross profit, and better service are different outcomes. If a salaried person saves two hours and uses them elsewhere, that is additional capacity. It becomes a cash saving only when actual expenditure is reduced.
Include the unglamorous costs
Count implementation, subscriptions, model or messaging usage, human review, failed runs, support, and future changes. A cheap successful run tells you little if someone spends every Friday repairing the workflow.
Use scenarios for sales impact
For a lead workflow, an illustrative model is: eligible leads × change in booking rate × attendance rate × close rate × gross profit per sale. Every factor is an assumption until it is observed. Use conservative and optimistic scenarios rather than presenting one forecast as a promise.
Separate attribution from coincidence
A stronger offer, a new campaign, a different salesperson, or a seasonal change can move the results. Compare like-for-like segments where possible and document what else changed. Small samples deserve restrained conclusions.
Agree the next decision before the pilot
Define what would justify continuing, improving, or stopping. A process that reduces manual handling but produces unacceptable errors is not a success. A process with little revenue impact may still be worthwhile if it reliably removes an expensive operational bottleneck.
Report observed operational changes, actual costs, and estimated business impact separately.