Every alert that fired and nobody owned taught us the same lesson: monitors don't fail because the tool is wrong, they fail because nobody's name is attached to the response. We audited a quarter of pages and found 30 alerts where the runbook said "investigate" and stopped. Thirty invitations to shrug.
The fix that worked
Every alert must answer three questions in its own description:
- Who gets paged - a name, not a team alias. Aliases distribute responsibility to nobody.
- What "handled" means - one sentence. "Error rate back under 1% for 15 minutes." If you can't write it, the alert is a vibe, not a monitor.
- The first three commands to run - copy-paste, not links to a wiki that moved in March.
What we deleted
After adding the three answers, 40% of existing alerts couldn't be completed by their own owners. Those got deleted. Noise down by half in a week - not by tuning thresholds, but by forcing every remaining alert to have a person and a definition of done.
Ownership is the cheapest monitoring upgrade. It costs one doc review per alert and returns every 3am page you never had to send.
All drills ship in the Ops Starter Kit ($29). Every kit in one download: Ops Mega Bundle ($49). Want it run for you, 24/7? Always-on ops desk.
Top comments (0)