Framework
After shipping a feature, ask:
- Did we ship on time? If delayed, what surprised us?
- Did the feature move the intended metric? If not, why?
- What assumptions in the PRD were wrong?
- What edge cases did we miss?
- What would we specify differently next time?
Document answers. Use them to improve future PRDs.
Example
Planned: Recommendations feature, 80% accuracy. Actual: Shipped 2 weeks late, accuracy 75%. Learning: We underestimated cold-start complexity. Next time, we'll spec cold-start handling separately + give it more time.
Actionable Steps
1. Schedule Retrospective 2 Weeks Post-Launch
(Enough time to see real data, recent enough to remember details)
2. Invite: PM, Engineering Lead, Design Lead
3. Document: Template with questions above
Key Takeaways
- Retrospectives are where PM skill compounds. Each feature teaches you how to spec the next one better.
- No blame, just learning. If you ship 6 features/year and improve 10% per feature, you're incomparably better in 2 years.