Everyone knows you should write postmortems. Nobody writes them, because the template demands a novel. Ours used to take three hours — blameless culture or not, nobody has three hours at 5am after a bad night. So we cut the template to five fields and gave it a hard 15-minute timer. Completion rate went from ~20% to near 100%, and the incident-review meeting shrunk from an hour to 12 minutes.
The five fields
- What the customer saw — one sentence, plain language. "Checkout returned errors for 23 minutes." If you can't write this sentence, you don't understand the incident yet.
- The trigger — the proximate event. "Deploy 4471 changed the session TTL."
- The contributing conditions — the boring reasons the trigger mattered: stale dashboards, missing alert, on-call in meetings, undocumented dependency.
- The one fix — singular. The change that stops this class of incident, not just this instance.
- The check — how you'll verify the fix worked, with a date. "Restore drill passes on staging by Friday."
Why the timer matters
Fifteen minutes forces small findings. Small findings get done. A finding like "add a 30-minute disk-usage forecast alert" ships in a day. A finding like "redesign the deployment pipeline" never ships — it just lives in the doc making everyone feel thorough. When someone proposes a six-week fix, we ask: what's the 15-minute version that prevents 80% of the damage? There almost always is one.
The rules that keep it honest
- Written by the on-call, not the manager. Managers write narratives; on-calls write facts.
- No names, ever. Systems failed, not people. The moment a postmortem names a person, the next incident gets hidden.
- Read aloud in the weekly ops sync — 12 minutes for all of last week's incidents. Postmortems nobody reads are just diary entries.
Two years of these five-field reviews is worth more than any monitoring upgrade we've bought. The template is free. The habit is the product.
Every drill in this post ships in the Ops Starter Kit ($29). All five kits in one download: Ops Mega Bundle ($49). Or let the always-on ops desk run them for you.
Top comments (0)