All insights
ProcessOperations · 5 min read
Why incident retros lie to themselves
Most incident retros end with action items that everyone agrees to, almost nobody does, and nobody follows up on. The format is mostly to blame.
Read a year of incident retros and you’ll see the same patterns: same root causes, same contributing factors, same action items. The format produces the illusion of learning without the learning itself.
What’s broken
- Action items get assigned but not prioritized. They lose to feature work.
- Root cause analysis stops at the first plausible explanation.
- The same five people attend, so the same five perspectives get heard.
- Nothing checks whether the previous incident’s actions actually happened.
How to fix it
Read the previous retro at the start of every new one. Verify the actions actually happened. If they didn’t, that’s the first finding, and a more interesting one than whatever broke this week.
Retros don’t fail because the analysis is wrong. They fail because nobody reads the last one.