Set the baseline. Then test the case.
Agree relevant KPIs before recommending implementation: cycle time, staff effort, error or rework rate, throughput, cost per task, or another measure tied to the desired result. Record definitions, data sources, sample limitations and the accountable owner.
If AI is a candidate, define task-specific quality tests, acceptable failure conditions and the human review required. Compare any pilot against the baseline, including the effort to review and maintain it.
Potential value is a hypothesis to test. Savings, revenue, AI accuracy and implementation results are not guaranteed. If the baseline or evidence is inadequate, the recommendation may be to gather better evidence first.