An editorial framework for discussion. This note does not report empirical findings or client outcomes.
Name the decision
Describe the work: triaging an exception, preparing a quote, reviewing a document. Identify who acts on the output and what a better result changes.
Establish the baseline
Measure the current outcome and its variation. Check whether inconsistent inputs, unclear approval rules, or missing data explain the problem. Repairing these foundations may be the right first step.
Make the test interpretable
Specify the comparison, review process, and stopping rule before building. Include errors, rework, adoption, and oversight effort. A promising demonstration is evidence to investigate, not a basis for automatic rollout.
Stable workflow + usable data → bounded pilot
Unclear ownership → process clarification
High error consequence → human control