On-Call for a Five-Person Team (Without Pretending You're Google)
Incident commanders and severity matrices are for companies with an SRE department. The small-team version of on-call: one primary, an escalation chain instead of a calendar, nights that stay quiet, and a weekly proof it all still works.
On-Call for a Five-Person Team (Without Pretending You're Google)
Search "on-call best practices" and you'll find runbooks written for companies with an SRE department: incident commanders, severity matrices, blameless postmortem rituals, follow-the-sun rotations. Useful — if forty people carry the pager.
Most businesses that need on-call have five people. A shop, a clinic, a small agency, a two-developer SaaS. The failures are just as real — the site dies on a Saturday, a payment run fails at night — but the org chart is you, maybe a partner, and whoever answers their phone. Here's the version of on-call that actually fits.
The three roles that matter (and the five you can skip)
You need exactly three things decided before an alert fires:
- A primary — the person alerts reach first. Not "the group chat." One name.
- A backup — where the alert climbs if the primary doesn't acknowledge in N minutes.
- A final catch — usually the owner. Not because they fix things, but because an alert must have nowhere left to die.
You can skip: incident commanders, comms officers, severity taxonomies beyond "urgent or not," war rooms, and any rotation spreadsheet complicated enough to need its own manual. Those exist to coordinate forty responders. You're coordinating three.
Let escalation BE the rotation
The enterprise model asks "who is on call right now?" and maintains a calendar to answer it. The small-team model asks a better question: "who acknowledged it?" — and lets the chain handle the rest. Alert → primary. Five quiet minutes → backup. Five more → owner.
This absorbs real life in a way calendars don't. Primary at a wedding? The chain doesn't care — it climbs. Nobody has to remember to swap a shift, and yet nothing is ever unowned. A calendar-grade schedule is worth adding only when you have genuine shifts (a weekend person, an evenings person) — and then it should feed the same chain, not replace it.
Respect sleep, structurally
Small-team on-call fails when it burns people, and it burns people through non-urgent noise at 2 a.m. Two structural fixes:
- Quiet hours for the humans: routine alerts hold until morning; only the genuinely urgent tier makes sound at night.
- Distinct sounds for the tiers: the urgent tone should be recognizable from across a room — and its arrival should mean something, because everything non-urgent respects the night.
If everything pages loudly, people mute everything; the system self-destructs politely. Priority discipline is what keeps the pager credible.
Test it like you mean it
An untested pager rots silently: phones get replaced, notification permissions drift, sound profiles change, someone's token expires. The fix costs one alert a week — a scheduled drill through the real routing chain that must be acknowledged on a real phone. If Monday's drill goes unacknowledged, you've learned the pipeline is broken during a rehearsal instead of a fire. (This is exactly what FlowAlert's weekly test alert does, and why we run it on ourselves.)
The whole setup, concretely
A five-person version takes an evening: route urgent things (site down, failed payments) to the primary with acknowledgement required and a five-minute escalation to the backup; route everything informational at normal priority where quiet hours apply; add a catch-all so nothing new can arrive unrouted; enable the weekly drill. FlowAlert was built for exactly this shape — flat price for the whole team, escalation and acknowledgement included, no per-seat math punishing you for adding the Saturday person — but the shape matters more than the tool: one name first, somewhere to climb, nights that stay quiet unless it's real, and a weekly proof the whole thing still works.
