Retrospective and lessons-learned synthesis
A retro or lessons-learned session generates a pile of sticky notes and comments, and the value leaks away when nobody synthesizes them into patterns and concrete improvements. AI can theme the raw feedback quickly, separating a broad systemic issue from one loud voice. Because the input is honest team feedback, anonymizing it first is not optional.
You are a facilitation assistant. Synthesize the anonymized retrospective feedback below into themes and actionable improvements. Do not exaggerate sentiment or invent issues that are not in the input. Project context: {{project_context}} Anonymized retro feedback (what went well / what didn't): {{retro_feedback}} Produce: 1. What went well — themed, with rough frequency ("about half of comments"). 2. What did not go well — themed the same way. 3. Signal vs. noise: which themes are broad patterns versus one or two strong individual comments. Do not overweight a single vivid remark. 4. 3-5 concrete, owned improvement actions for next time, phrased so they can go into a plan. Rules: - Base themes only on the feedback provided. Do not attribute a comment to any named person, team, or role. - Preserve dissent and critical feedback honestly — do not smooth it into something more positive. - Report frequencies as approximate, never as false precision. - Flag any comment describing a personnel, harassment, or conduct issue separately so it can be routed to the right process, not treated as a retro theme.
Fill in your details and the prompt updates live — then copy.
What went well (about half of comments): Daily standups kept blockers visible; strong vendor relationship. What didn't (about a third): Unclear ownership of data cleanup; testing squeezed at the end. Signal vs. noise: "Testing got squeezed" is a broad pattern across the team; one comment about meeting-room booking is a one-off, not a theme. Improvement actions: - Assign a single data-quality owner at kickoff (owner: PM). - Reserve a fixed testing window in the plan, protected from scope pressure. Flag for separate handling: One comment describes a conflict between two team members — route to their managers, not the retro summary.
The full workflow
- Collect the retro input and strip names and identifying detail before pasting
- Run the synthesis in an approved enterprise tool, not a personal account
- Spot-check the themes against the raw notes so nothing is overstated or lost
- Turn the improvement actions into owned items and carry them into the next project
Watch out for
Retro feedback is candid because it is confidential — pasting identifiable comments into a consumer AI account breaks that trust and can expose personnel matters. Anonymize first and use an approved enterprise tool with no-training and retention controls.
AI can smooth over or overstate sentiment. Verify the themes against the raw feedback before sharing, and route any conduct, harassment, or HR issue to the proper process rather than leaving it as a retro line item.
Where this comes from
Every use case on this site is grounded in real reports from working project managers — not invented by us.