Measuring on-call health quantitatively

WHAT THIS TESTS Whether you can make on-call pain measurable and turn those measurements into a business case for engineering work, rather than relying on anecdotes.

A GOOD ANSWER COVERS Useful metrics: the number of pages per shift, broken out by business hours versus out-of-hours and especially overnight, since sleep-disrupting pages are the strongest burnout driver; the actionability rate, the share of pages that genuinely required human action, which exposes noise; time to acknowledge and time to resolve, indicating both responsiveness and incident severity; the interrupt load on the on-call engineer, including non-page interrupts; and the distribution of load across people, to catch a few engineers absorbing most of the pain. Google SRE suggests a rough budget, on the order of no more than about two pageable events per shift, as a sanity check. You use these by tracking trends over time and per service: a service generating many low-actionability overnight pages is a clear, data-backed candidate for reliability work; rising MTTR justifies better runbooks or tooling; uneven distribution justifies rotation changes. Presenting these numbers to leadership converts felt pain into prioritized, fundable engineering tasks.

COMMON WRONG ANSWERS Counting raw incidents only, ignoring whether they were actionable or off-hours. Treating all pages as equal regardless of timing. Ignoring load distribution. Having no target or budget to compare against. Failing to connect the metrics to a concrete prioritization decision.

LIKELY FOLLOW-UPS What is a reasonable page budget? How do you weight overnight pages? How do these metrics feed sprint planning?

ONE CONCRETE EXAMPLE Over a quarter, a service averages six pages per shift, 40 percent overnight, with only 20 percent actionable and an uneven load where two engineers took most shifts. You present this: most pages are non-actionable overnight noise. The data justifies a sprint to retune alerts and fix the top recurring cause, and the next quarter pages drop to under two per shift, demonstrably reducing burnout.