Designing on-call rotations that don't burn out your team

On-call burnout is usually blamed on the schedule, but the schedule is rarely the actual cause. It's being paged for things that never should have paged a human in the first place — a flaky check that self-resolves, a threshold set too tight, an alert with no runbook attached so every page becomes an investigation from zero. Fix the noise before you touch the rotation, because a perfectly fair schedule of being woken up for nothing is still a schedule of being woken up for nothing.

Most pages shouldn't be pages

Before redesigning a rotation, audit a month of pages and sort them into three buckets: genuinely required a human right now, could have waited until morning, and shouldn't have fired at all. That third bucket is usually the biggest one, and it's the highest-leverage fix available — every alert that gets tuned, batched into a daily digest, or deleted outright is one fewer 3 a.m. wake-up for every future rotation, forever. This is the same discipline behind error budgets: decide in advance what's actually worth an interruption, and let everything below that threshold accumulate as a metric someone reviews in daylight instead of a page someone answers at 3 a.m.

Teams that skip this step and go straight to redesigning the rotation are optimizing the wrong variable. A fairer distribution of noise is still noise — it just spreads the exhaustion around more evenly instead of removing it.

A schedule people can actually see and trust

A rotation only works if it's unambiguous: whose week it is right now, when it started, when it hands off, visible on the same tool where the actual incidents live rather than a separate spreadsheet or calendar invite that quietly drifts out of sync. The moment someone has to ask 'wait, is it my week?' the schedule has already failed at its one job. Keeping the rotation on the same system as the incident issues it produces also means a page automatically has an owner it can be assigned to without a manual lookup — one less step between an alert firing and someone actually looking at it.

Compensate the time, don't just thank people for it

Being reachable at 3 a.m. is real labor even on a quiet week, and treating it as a favor someone does for the team rather than work that gets accounted for is how resentment builds quietly under a rotation that looks fine on paper. It doesn't have to be extra pay — protected time off the morning after a page, or a lighter workload the day after a bad night, is often enough to make the rotation feel respected rather than extracted. The teams where on-call burns people out fastest are almost always the ones where a rough night is followed by a normal 9 a.m. standup as if nothing happened.

The handoff is where the time actually gets saved

Most of the wasted effort in a bad rotation isn't the incidents themselves — it's the next person re-discovering context that already existed. A real handoff, even a short one, changes that: what fired this week, what's still open and being watched, what pattern to keep an eye on. Without it, every new on-call week starts from zero, and the team pays the investigation cost of the same incident twice — once when it happened, and again a week later when a different engineer rediscovers it the hard way. This connects directly to running a retrospective that actually changes something — a good handoff is a retro's raw material, written down while the details are still fresh instead of reconstructed a week later from memory.

Escalation tiers beat a single flat rotation

A single person carrying every possible page, at every hour, for an entire week is a heavier load than the same volume of pages spread across a tiered structure: a primary who gets the first attempt, a secondary who's paged automatically if the primary doesn't acknowledge within a few minutes, and a wider escalation for anything that's genuinely severe enough to need it. Tiering doesn't reduce the total number of pages a team absorbs, but it does mean any one person's week is bounded — they're not the last line of defense for every category of failure, just the first attempt at the routine ones, with real backup for the rest.

It also makes the schedule itself less fragile. A flat rotation where the one person on-call is unreachable for an hour means every incident in that window goes unanswered; a tiered one degrades gracefully instead of failing outright, which is worth the modest extra setup cost on its own.

What a healthy rotation looks like in practice

Put together, a rotation that doesn't burn people out has a small, tuned set of things that are actually allowed to page a human; a visible, trusted schedule living next to the work it produces; escalation tiers so no single person is the sole point of failure for an entire week; a real acknowledgment — time, not just thanks — that the week costs something; and a handoff that hands the next person context instead of a clean slate. None of these fixes require a bigger on-call team, just a more honest one about what deserves an interruption. And on the input side, AI triage catching and routing the incoming noise before it ever reaches a human pager is one of the more direct ways to shrink that first, noisiest bucket without touching the schedule at all.