Write the stop condition before extending the pilot

If you are deciding whether an AI pilot gets another extension, write what would make the team pause, narrow or stop it. “We will review progress later” leaves the continuation decision open without saying which evidence will change the choice.

Begin with the decision the pilot is meant to inform. If the question is whether a support-drafting workflow reduces total handling effort, define what evidence would answer it: baseline effort, drafting effort, review effort and exception work for a comparable set of cases.

Name a decision owner and the evidence they need. State the boundary of the pilot: users, data, workload and allowed actions. Then write the trigger. It might be an unacceptable failure pattern, missing evidence, costs above a stated limit or a useful result that appears only for a narrower set of cases. Use thresholds appropriate to the real work; do not copy a fictional example as policy.

A narrow decision is still a decision. The outcome can be to fund a bounded next step, restrict the use case, defer until a missing dependency is resolved, or stop. Recording the alternative and its reason makes the next discussion more useful than treating every pilot as a commitment to scale.