On-call rotations are one of those systems that quietly determine engineering retention more than most teams realize. A well-designed rotation is barely noticeable. A poorly designed one shows up in resignation letters months after the actual pattern of bad nights started, by which point the damage is already done. A few structural choices tend to make the biggest difference between the two outcomes.
Rotation length matters more than rotation frequency
A common instinct is to spread on-call thin, more people in the rotation means each person is on call less often. This helps, but it interacts with rotation length in ways that aren't always intuitive. A one-week rotation with eight people in the pool means each person is on call roughly every two months, which sounds reasonable, but a full week of being interruptible, including nights, is a meaningfully different cognitive load than the same total hours spread across shorter, more frequent shifts.
Some teams find that shorter rotations, three or four days, with a slightly larger pool, produce less cumulative fatigue than longer rotations with a smaller pool, even when the total on-call hours per person over a quarter work out similar. The difference seems to come down to how much a full week of disrupted sleep and attention compounds compared to shorter stretches with more recovery time between them.
Alert quality determines whether on-call is sustainable at all
No rotation schedule survives contact with a noisy alerting system. If on-call engineers are being paged for issues that don't actually require immediate human intervention, the schedule itself becomes almost irrelevant, because the actual problem is alert fatigue, not rotation design.
A useful practice is tracking, for every page, whether it required real-time action or could have waited until business hours. Alerts that consistently fall into the second category are strong candidates for downgrading to a non-paging notification, or for fixing the underlying issue that's generating them in the first place. Teams that review this data monthly tend to see page volume drop significantly over a few quarters, simply by removing alerts that were never actually actionable at 3am.
Compensation and recognition need to be explicit, not implied
On-call work is real work, and treating it as an unstated expectation of the job rather than something explicitly compensated, whether through pay, time off in lieu, or another mechanism, tends to create quiet resentment that doesn't show up directly in complaints but does show up in attrition and in reluctance to volunteer for the rotation.
Teams that handle this well tend to be explicit and consistent: a fixed on-call stipend, or a clear policy of comp time for any incident response outside working hours, removes the ambiguity and signals that the disruption is recognized rather than assumed as a baseline expectation.
Escalation paths need to actually work, not just exist on paper
A common failure mode is an escalation policy that looks complete in the documentation but has never been tested in practice. The primary on-call person doesn't respond within the expected window, and the secondary escalation either doesn't trigger correctly or nobody remembers who's supposed to pick it up. This gap is usually invisible until an actual incident exposes it, at the worst possible time.
Periodically testing the escalation path, not just reviewing it on paper, catches configuration drift and staffing gaps before they matter during a real incident.
Protecting recovery time after a bad on-call shift
A rotation that technically ends on schedule but doesn't account for a rough night, several pages, disrupted sleep, still leaves someone expected to be fully present the next morning. A policy that allows for a delayed start or a lighter workload the day immediately following a disruptive on-call night, without requiring the engineer to justify or negotiate it individually, removes a source of quiet burnout that pure schedule design doesn't address on its own.
The signal that a rotation needs redesign
Volunteer rate for on-call duty is a more honest signal than survey responses. If engineers are actively avoiding the rotation, negotiating out of it, or the same small subset of people keep ending up covering more than their share, that's a more reliable indicator that something structural needs to change than any satisfaction score collected after the fact. Schedules that are actually sustainable tend to have engineers rotating in without needing to be convinced, because the system, page quality, compensation, recovery time, has been designed around what a person can reasonably sustain rather than around minimum coverage requirements alone.
Top comments (0)