Define the task from start to finish
Choose a unit of work that staff recognise, such as one completed enquiry or one reviewed document. State where timing starts and ends. A fictional pilot comparing draft preparation should not count only the AI generation step while ignoring the time a person spends checking and editing it. Include the stages needed to reach the same usable outcome in both methods. Separate active staff effort from elapsed waiting time, because reducing one does not necessarily reduce the other.
Observe representative current work
Collect examples from the actual process with permission and appropriate handling of business information. Include ordinary cases and relevant exceptions rather than selecting only the easiest items. Record the period and method of observation. If staff provide estimates, retain them as estimates and explain their basis. A short observation window may miss month-end peaks or unusual work. Do not hide that limitation. The baseline should describe the work you observed, not imply it represents every team, season or customer type.
Measure quality alongside effort
Define what a usable result must contain and how errors or omissions will be identified. Record rework, escalation and missing information using the same definitions before and during the pilot. A faster output that requires more correction may not improve the whole task. Ask the people using the result to assess it against the agreed standard. Keep evidence for material findings rather than relying only on a satisfaction score. Where an outcome cannot be checked independently, report the uncertainty instead of treating confident wording as correctness.
Make the comparison fair
Use comparable input types, complexity and completion criteria. If the pilot uses cleaner data or a narrower task than the baseline, state that difference and avoid attributing all improvement to AI. Record changes in staffing, source quality or workflow design that could affect the result. Do not compare one exceptional manual case with an average automated case. Preserve the raw counts and timing observations so another reviewer can understand the calculation. Summaries should explain the comparison, not replace the underlying evidence.
Separate saved effort from financial benefit
Measured time reduction is an operational result. Turning it into a money figure requires additional assumptions about how the capacity is used and what costs actually change. Keep those assumptions visible and do not describe theoretical capacity as cash saved. Include ongoing review, support and operating effort when discussing the overall process. If the business wants a financial assessment, use its actual information and the appropriate responsible reviewer. This guide provides a measurement process, not a forecast or a guaranteed return.
Report results with their limits
State the number of cases, observation period, task boundaries and checks performed. Show the measured difference alongside quality findings and unresolved issues. Avoid multiplying a small pilot result across a whole year without clearly labelling the projection and its assumptions. Revisit the baseline if the process changes materially. A useful report supports a specific decision: continue testing, improve the workflow or proceed within the demonstrated scope. It should remain understandable when read later by someone who did not participate in the pilot.