Standing up a new team's first rotation is where most of the bad habits get baked in permanently. Here is the checklist we now use on day one.

A rotation's worst habits are set in its first month and are brutally hard to unwind later. We built a day-one checklist after fixing three rotations that had already calcified badly.

Size it from expected page volume, not headcount you have

We estimate expected weekly pages from comparable services before deciding rotation size, rather than just splitting whoever is on the team evenly. A team of four with an expected two pages a week doesn't need a formal secondary shift yet; a team of four expecting fifteen does, even if that means borrowing responders from an adjacent team temporarily.

from pages
| where service in (comparable_services)
| summarize weekly_pages = count() by bin(_time, 7d)
| summarize avg(weekly_pages), p90(weekly_pages)

Set the escalation policy before the first page, not after

New rotations often discover their escalation policy is wrong during the first real incident, which is the worst possible time. We now require a documented, tested escalation chain — including a named backup for the backup — before a service goes on a paging rotation at all, and we test it with a synthetic page in business hours before it ever fires for real.

Budget onboarding time into the first shift

A new engineer's first on-call shift should not be their first exposure to the runbooks. We pair a new responder with a shadow shift — full visibility into pages, zero responsibility to act — before their first solo shift. Skipping this step was the single biggest predictor of a bad first incident in the rotations we audited.

Write the rotation's first retro before the first incident

We schedule a 30-day rotation retro on the calendar the same day the rotation goes live, before there's necessarily anything dramatic to discuss. Waiting for a bad incident to justify a retro means the small annoyances — a confusing dashboard link, an alert that fires slightly too eager — never get raised until they've compounded into something that actually hurts.

  • Size the rotation from expected page volume of comparable services, not headcount.
  • Document and test the full escalation chain before the first real page.
  • Require a shadow shift before anyone's first solo on-call rotation.
  • Schedule the first rotation retro on day one, not after something goes wrong.
  • Revisit rotation size after the first quarter of real page data.