Preventing Burnout: The Metrics of Fatigue

You cannot manage what you do not measure. On-call health is no different. The following metrics, tracked per shift and per rotation cycle, tell you whether your on-call program is sustainable or quietly destroying your team.

Key Metrics to Track

1. Pages Per Shift
The most direct measure of on-call burden. Count every page that woke or interrupted an engineer during their shift.

• Target: Fewer than 2 actionable pages per 12-hour night window.
• Red flag: More than 5 pages per night, sustained over multiple shifts. At this level, engineers are not sleeping. Cognitive function degrades. Mistakes multiply.

2. Alert Actionability Rate
For every page, ask: did the engineer have to do anything, or did they just look and go back to sleep?

• A page that requires no action is called a non-actionable alert or false positive.
• Non-actionable alerts cause alert fatigue — the gradual conditioning of engineers to treat all alerts as probably-nothing. This is how real incidents get missed.
• Rule: Any alert that fires three times without requiring action should be deleted or fundamentally redesigned. Silence is better than noise.

3. Sleep Interruption Rate
Track pages specifically between 11 PM and 7 AM local time for the on-call engineer. This is the metric that most directly maps to health impact. A service that pages twice at 2 AM is more damaging than one that pages ten times at 10 AM.

4. Mean Time to Acknowledge (MTTA)
How long does it take the on-call engineer to acknowledge a page? A rising MTTA is a signal that engineers are exhausted, have alert fatigue, or are in a timezone where the page arrives at an unreasonable hour.

Human Policies That Prevent Burnout

The Post-Incident Recovery Mandate: If an engineer worked a SEV1 from 2 AM to 5 AM, they are not expected at standup at 9 AM. This should be written policy, not a manager's discretionary favor. Sleep deprivation after a major incident is a safety issue — a tired engineer making decisions about a fragile system is a liability.

Minimum Rotation Size: A rotation needs at least 4 people to be survivable, and 6–8 people to be healthy.
- With 3 people: on-call every 3rd week. That is 17 weeks of on-call per year. This is not a rotation; it is a chronic condition.
- With 6 people: on-call roughly once every 6 weeks. Engineers have time to recover, forget the stress, and return to the rotation without dread.
- With 8 people: on-call once every 8 weeks. This is the range where engineers stop dreading the rotation and start treating it as a manageable part of the job.

The Rotation Freeze: When a team drops below 4 people (due to attrition, leave, or hiring lag), freeze new feature work and treat rotation health as a P0. A team of 3 on a weekly rotation will lose a fourth member within months.