How to run an after-action review that improves the next decision

Prepare the record

Collect the original objective, decision brief, options, assumptions, evidence, expected outcome, commitments, trigger conditions, and minority warnings. Include timestamps. Ask participants to write individual observations before the meeting so that the first confident narrative does not become the whole account.

Define the review boundary. A product experiment, negotiation, launch, hiring process, or strategic quarter may each need a different horizon. State which result is mature enough to examine and which effects remain too early to judge.

Compare intent and reality

Ask four plain questions: what did we intend, what happened, what explains the difference, and what will we change? Begin with observable events. Separate the timeline from interpretations. Capture results that were better than expected as carefully as failures; positive surprises can also reveal a faulty model.

Compare the predeclared success signal with the actual measure. If the team changed the metric during execution, record why. A changed measure may be sensible, but it should not disappear into the final narrative as if it had always been the goal.

Separate decision, execution, and environment

Evaluate the decision using information available at the time. Then evaluate execution against the chosen plan. Finally identify external changes or other actors’ responses. These categories interact, but separating them prevents every poor result from being blamed on execution and every good result from being credited to strategy.

Ask whether implementation exposed a missing assumption that should have been in the brief. Ask whether the trigger conditions fired and whether anyone acted. A trigger that everyone ignored is a governance problem; a trigger that created noise may need redesign.

Review counsel and dissent

Map the result back to council contributions. Which seat identified a mechanism that mattered? Which warning was accepted but not mitigated? Which advice was followed without adequate evidence? Where did a mentor corpus have insufficient coverage? Preserve the citations used at the time rather than retrieving new passages to justify the outcome.

Do not score mentors by one result. A valid risk may not materialise, and a confident prediction may be lucky. Evaluate whether the contribution improved framing, evidence, contingency, or action. Update the seat description or source requirements if the same limitation recurs.

Convert learning into system changes

Choose a small set of changes with owners: add a field to future briefs, change a threshold, gather evidence earlier, alter a council seat, clarify ownership, instrument a process, or obtain professional review. General lessons such as “communicate better” are not operational until they change a routine or artifact.

Schedule a check that the change occurred. Then close the review. The purpose is not exhaustive explanation; it is sufficient understanding to improve the next relevant decision. Preserve unresolved questions separately so they can be revisited when later evidence arrives.

Distinguish a local correction from a reusable lesson. A one-off supplier failure may require a contingency for that relationship; a repeated failure to test dependencies belongs in the standard decision brief. Write the proposed rule, where it will live, who maintains it, and the cases to which it should not apply. Revisit the rule after several uses. Without that boundary, teams either forget the lesson or overgeneralise from a vivid event and burden unrelated work with controls that add cost without reducing the relevant risk.