Revisit the original question
Read the pilot brief before reviewing the results. Did the experiment answer its original question, or did the task change midway? Both can be useful, but they are different findings. If configuration, users or data changed, explain which observations belong to which version.
Compare results with the acceptance criteria agreed at the start. Avoid lowering a threshold after seeing the outcome without documenting why. A changed threshold may be reasonable, but it should become a new decision rather than an invisible adjustment to make a pilot appear successful.
Choose one of three paths
The evidence does not justify continuing
The value is weak, required controls are unavailable, or failures are too consequential for the proposed use. Record what was learned and remove temporary access using the organization's normal process.
The question needs a narrower test
Limit the use case, improve the evaluation set or change the workflow. State what the next experiment will establish that this one could not. Give it a new owner-approved scope and review date.
A defined next step is justified
Specify the additional users, task volume or data involved. Recheck the risks introduced by that expansion. A successful controlled pilot is not automatic approval for organization-wide deployment.
Use a decision record, not a score alone
Capture the observed results, known limitations, unresolved risks, chosen path and accountable decision-maker. Include dissenting views or conditions attached to approval. A single readiness score can obscure a serious control gap, so the planner here does not generate one.
For example, a hypothetical document-triage pilot might reduce review effort on routine cases while performing poorly on unusual requests. The sensible next step could be a restricted routine-case workflow with manual routing for exceptions, rather than a broad rollout. That is a decision pattern, not a claimed case study.
Keep operating obligations visible
Before expansion, identify who maintains the reference material, monitors output quality, reviews permission changes and handles user reports. Budget for this work. An experiment supported by a highly engaged project team can behave differently when transferred into ordinary operations.
Record the conditions that trigger another review, such as a changed model, new data source, expanded audience or unexpected failure pattern. Keep an ordinary fallback available where feasible. Governance continues after the pilot rather than ending at the final meeting.
A compact meeting agenda
- Restate the intended use and boundary.
- Review quality, effort, exceptions and control failures.
- Name what remains unknown.
- Choose stop, adapt or a bounded expansion.
- Assign owners, conditions and the next review date.