Run a Launch Retrospective
Structures an honest post-launch review of what the feature actually achieved, what the process cost, and what to change next time — separating the outcome from the execution so a well-run launch of the wrong thing gets named as such. Use it two to four weeks after shipping.
0 likes
0 dislikes
Sign in to rate this prompt
Prompt
You are facilitating a retrospective. Your job is to get to what's true, not to what's comfortable, and to keep blame out of it without letting that turn into vagueness.
What we launched and when: {{launch}}
What we predicted would happen: {{predictions}}
What actually happened — metrics, adoption, feedback, incidents: {{results}}
How the process went — timeline, slips, surprises, friction: {{process}}
What the team has said so far: {{team_input}}
Produce:
**Outcome versus prediction.** Compare {{results}} against {{predictions}} directly, metric by metric. Where we were wrong, say by how much and in which direction. Resist the pull to reframe a miss as a learning — first record it as a miss.
**Why.** The most plausible explanations for the gap, separated into: we built the wrong thing, we built the right thing badly, we built it fine but nobody found it, or the metric was never going to capture it. These have completely different fixes, and teams routinely apply the wrong one.
**What we got right.** Specific and attributable, not morale-management. If something worked because of a deliberate decision, name the decision so we repeat it.
**Process.** From {{process}}: where time went that we didn't expect, what was discovered late that should have been known early, and which handoffs were rough. Look particularly for problems visible in the PRD or planning stage that only surfaced during build.
**The uncomfortable finding.** The thing in {{results}} or {{team_input}} that people are talking about privately and not in the review. Name it directly.
**Decisions, not action items.** For each finding, what actually changes: a specific change to how we work, with an owner. Cut anything that's really just "be more careful."
**What's next for the feature itself.** Given {{results}}: invest, iterate, leave it, or sunset it. Make a recommendation. Features that quietly underperform forever are a real cost.
**The prediction to record.** What we now expect over the next quarter, written down so the next retro can check it.