Key Takeaway
Blameless post-mortems produce better action items because participants share information freely when they know the focus is on systemic improvement rather than individual fault.
When to Use This Template
Use this agenda after every significant AI system incident: model quality degradation, safety filter failures, data leakage, availability outages, or cost overruns. Schedule the post-mortem within one week of incident resolution while details are fresh. Invite everyone involved in the incident response, the system owner, and a facilitator who was not directly involved in the incident.
Meeting Flow
- 1
Ground Rules (5 min)
Facilitator establishes blameless norms: we are here to improve systems, not assign blame. Focus on what happened and why, not who did what wrong. Emphasize that speaking up about mistakes and near-misses is valued and protected. Set the expectation that action items will target systemic fixes.
- 2
Incident Summary (5 min)
Present the incident factually: severity level, duration, users affected, financial impact, and current status. This establishes shared context without debate. Use the written incident report as the starting point.
- 3
Timeline Reconstruction (15 min)
Build a collaborative timeline on a shared document or whiteboard. Start from the first anomaly signal and walk through detection, response, escalation, mitigation, and resolution. Each participant adds their perspective. Focus on events and decisions, not judgments about those decisions.
- 4
Root Cause Analysis (15 min)
Apply the five-whys technique starting from the incident trigger. Identify contributing factors across categories: monitoring gaps, testing gaps, process gaps, knowledge gaps, and tool limitations. Map contributing factors to systemic issues that can be addressed to prevent recurrence.
- 5
What Went Well / What Could Improve (10 min)
Explicitly identify effective responses and practices that limited the incident impact. Then identify areas for improvement: detection speed, response coordination, communication, documentation, and tooling. This balanced review prevents post-mortems from being purely negative.
- 6
Action Items (10 min)
Generate specific, measurable action items for each improvement area. Categorize as: immediate (this week), short-term (this quarter), long-term (next quarter). Assign an owner and deadline to each. Prioritize actions that prevent recurrence of this specific incident and similar classes of incidents.
Facilitation Guidance
The facilitator should be someone who was not directly involved in the incident response to maintain neutrality. If the discussion drifts toward blame, redirect with: 'Let us focus on what we can change in our systems and processes.' Ensure quieter participants have space to contribute, especially those who may have observed things others missed. After the meeting, publish the post-mortem document to the broader organization to maximize learning value.
Track post-mortem action item completion rates as a team metric. Post-mortems that generate action items that are never completed erode trust in the process. If completion rates drop below 80%, reduce the number of action items per post-mortem and focus on the highest-impact systemic fixes.
Version History
1.0.0 · 2026-03-01
- • Initial incident post-mortem agenda template