What is an escalation policy (and how many levels a small team needs)

People conflate two different things.

An on-call rotation answers: whose phone is supposed to ring this week?
An escalation policy answers: what if that phone does not?

Without the second, a missed primary page becomes a silent outage. You thought someone was on it. Nobody was.

The job of escalation

Escalation is ordered patience:

1. Page the primary.
2. Wait a defined timeout for an acknowledgment.
3. If nobody acks, page the next level.
4. Stop climbing as soon as someone owns it.

Acknowledgment is the important word. Escalation should cancel when a human says "I have it," not when the underlying monitor recovers — recovery might be luck while the root cause is still on fire.

How many levels you need

Two levels for most small teams:

• Level 1 — primary on-call.
• Level 2 — secondary / backup (next in rotation, or a teammate who can take over).

A third level only if you have a real destination for it: engineering manager, whole-team channel, or a louder path for true SEV1. If level 3 is "page everyone every time," you do not have a policy — you have a panic button wired to the smoke alarm.

More than three levels on a ten-person team usually means unclear ownership, not maturity.

Timeouts

Start at five to ten minutes for customer-impacting pages. Shorter and you escalate while someone is still finding their laptop. Longer and a real outage burns while you wait politely.

Different severities can use different policies. A SEV1 might escalate in five minutes. A SEV3 might not page at night at all — it waits for business hours. That is still an escalation policy; it is just honest about urgency.

What escalation cannot fix

Noisy monitors. If level 1 is false half the time, level 2 will learn to ignore the policy. Fix alert trust before you add more rungs.

Wrong channels. Escalating into a Slack channel that everyone mutes is theater. See which channel wakes you at 3am.

No rotation underneath. Escalation policies route to schedules and people. Empty schedules escalate into the void.

A minimal policy you can write today

• Level 1: primary schedule, ack in 5 minutes.
• Level 2: secondary schedule, ack in 5 minutes.
• Level 3 (optional): manager or #incidents for SEV1 only.

Wire it once. Test it on purpose. The first time you need it for real, you will not want to be debugging the ladder.