What counts as evidence after a pilot?
Check time, service quality and handoffs before extending an operational AI pilot.
Audeamus Group · Practical guidance · 21 October 2026
Define the claim
Describe the workflow and the period observed. Record the baseline, sample size, data source and person responsible for checking it. A percentage without a denominator or comparison period is difficult to interpret.
Separate three kinds of statements
Measured means observed under a stated method. Projected means estimated from explicit assumptions. Qualitative means a documented experience or judgment. Keep those labels visible in the decision record and any later case study.
Check the whole workflow
Include human review, correction, exceptions and operating costs. Check whether an apparent improvement in planning moves work into deployment or customer care. Compare quality and service alongside speed, and record what remains unknown.
Ask whether the result travels
Document what changed, what was excluded and whether the result is likely to hold at normal volume. Use the evidence to choose whether to extend the test, change the approach or stop.