On this page5 sections
Two on-call rotations receive the same number of pages. In this hypothetical comparison, one handles short daytime incidents. The other interrupts several nights, keeps calling the same specialist and leaves cleanup for the following day. The counts match. The mornings do not.
To plan the work, follow it beyond the moment the pager goes quiet. Who helped? What is still unfinished? Which commitment will move to make room for recovery? Telling someone to rest while leaving all their deadlines in place is a kind thought attached to an impossible calendar.
One page can occupy several people
A page is a notification, not a unit of effort. An illustrative thirty-minute investigation with two fully engaged responders uses half an hour on the clock and one person-hour of work. If both were off duty, two people lose part of their night. Keep those measures separate; they describe different costs.
The work can continue after the incident. One person writes a handoff, another replaces a temporary repair, and a specialist answers questions while trying to return to planned work. Recording only the primary responder misses those hours. Averaging across the whole team can hide an expert who carries most of the difficult cases.
Google's discussion of on-call connects operational load with the engineering capacity available to improve a service. The useful local question is whether this team can carry this service commitment alongside its other work. A staffing ratio borrowed from elsewhere only helps when its assumptions fit.
Read diagram description
Conceptual workload map, not measured data. Track interruptions, investigation, specialist assistance and follow-up separately so a shorter primary shift does not hide work moved to someone else. Diagram labels: Page arrives: Timing and interruption matter; Primary response: Investigation and recovery; Specialist assistance: Work outside the scheduled shift; After the incident: Cleanup, handoff and recovery time; Capacity plan: Change commitments to reflect the work.
A kinder rota can still leave the specialist awake
Suppose two rotations merge. Each engineer now takes fewer primary shifts, but the same specialists support both families of services, and the primary has more systems to understand. Fewer shifts are a benefit. They do not establish that the total work fell.
Follow a difficult incident through the primary, secondary, specialist and next day's cleanup. Check whether the available skills and recovery time fit the combined service. If the work no longer fits, the review must change an allocation: defer a delivery commitment, adjust coverage or give recurring problems engineering time. A trend line cannot supply the hours the plan left out.
Look where the relief was supposed to happen
Alert cleanup should reduce unnecessary night interruptions while preserving failure detection. Training should let the primary finish more investigations without waking the specialist. Added staffing should make recovery possible and reduce the pile of unfinished work. The rota describes the plan; those outcomes show whether it helped.
Compare periods with similar work and note changes in service scope, incident mix and reporting. A quiet week may be welcome luck. Ask responders about help that never reached the incident record, too. Keep voluntary wellbeing information separate from performance scoring: the task is to repair the operating plan, not infer someone's health.
Give temporary coverage an ending
A rare incident may justify extra coverage. Name who arranges relief, which commitments move, when the arrangement will be reviewed and what permits ordinary coverage to resume. If that return keeps receding, revisit the service commitment. Calling the same workload temporary every week does not make it smaller.
An unstaffed dependency, a known permission gap or repeated calls to an off-duty specialist need not wait for a large statistical sample. Limited evidence should temper claims of improvement, not prevent repair of a demand everyone can identify.
The rotation design guide covers staffing and qualification, while the overload response guide addresses immediate relief. Where alert cleanup is the remedy, the alert acceptance method keeps failure coverage in view. This review connects those choices to work the organization will fund, defer or stop.
Bring the hidden hours into the delivery plan
| Field | What to record |
|---|---|
| Service commitment | State covered services, response expectations, and known skill dependencies. |
| Observed demand | Separate interruptions, disturbed nights, elapsed duration, combined effort, and follow-up. |
| Hidden transfer | Record work absorbed by secondary responders, specialists, or the following shift. |
| Capacity conflict | Name the delivery or recovery commitment that demand displaces. |
| Decision | Fund repair, change coverage, defer work, narrow scope, or accept a time-bounded surge through the proper owner. |
| Acceptance evidence | Define what should decrease and which independent service signal must remain acceptable. |
| Reversal condition | Name the workload or coverage evidence that would invalidate the change. |
| Accountable owner | Record who can change the plan and when they will revisit it. |
Return to the merged rotations. If the primary calendar improved but specialist nights did not, show both results. Put the calls, hours and unfinished repairs into the plan, and make room through training, repair work, changed coverage or recovery time. At the next review, follow those same pieces of work. Relief should reach the people doing them, including the person who was never listed as primary.
Sources & context
Sources linked in this article. Read alongside the author’s analysis; a citation does not independently verify a publisher’s claims.
- Google’s discussion of on-callsre.google