how to page the right person: escalation policy template

TL;DR

Write down who gets paged first for each service, who they escalate to after 10 minutes, and what happens at the 30-minute mark for a SEV1. A one-page escalation policy kills the "who do I call?" scramble at 3am. Templates beat tribal knowledge because the incident commander follows a list instead of guessing.

Error / query

how to page the right person: escalation policy template

You want pages to reach the right person fast, without improvisation, even when the primary responder is asleep.

Use this skill when

• pages routinely land on the wrong person or on nobody
• youre standing up an on-call rotation for the first time
• alerts have no clear owning service or team
• you need a documented escalation chain for an audit or compliance review

Not for this skill when

• you need the paging tool itself (PagerDuty/Opsgenie setup is a separate task)
• you need on-call scheduling software configuration
• the question is compensation or rotation length (see the on-call norms skill)

Steps

1. Map every service to a primary responder and a backup

List services and who owns them. Nobody pages "the backend team"; pages go to a named rotation.

cat > /tmp/escalation-map.csv <<'EOF'
service,primary_rotation,backup_team,sev1_owner
payments-api,payments-oncall,platform-team,engineering-manager
web-frontend,frontend-oncall,platform-team,engineering-manager
postgres-primary,data-oncall,platform-team,cto-oncall
EOF
cat /tmp/escalation-map.csv

Expected: a CSV with one row per critical service, each with a named primary rotation, a backup team, and a SEV1 owner.

2. Write the time-based escalation ladder

Define exactly what happens at 5, 10, and 30 minutes of silence. Minute marks remove all judgment calls.

cat > /tmp/escalation-ladder.md <<'EOF'
# Escalation ladder

- T+0: page the primary rotation for the affected service
- T+5: no ack -> page primary again and notify the team channel
- T+10: no ack -> escalate to the backup team rotation
- T+15 (SEV1): auto-escalate to the engineering manager regardless of ack
- T+30 (SEV1 unresolved): page the CTO-level on-call, pull in a second responder
EOF
cat /tmp/escalation-ladder.md

Expected: a documented ladder with concrete minute marks, so nobody improvises escalation timing mid-incident.

3. Store it where the incident commander can reach it in 30 seconds

Put the policy in the runbook repo and link it from the team channel topic or a pinned post.

cp /tmp/escalation-map.csv ./runbooks/escalation-map.csv
cp /tmp/escalation-ladder.md ./runbooks/escalation-ladder.md
git add runbooks/ && git commit -m "Add escalation policy" && git push

Expected: policy committed to the runbooks repo; anyone can open it from a phone browser during an incident.

Variant phrasings

who gets paged first when a service goes down

Same fix: the escalation map from step 1 answers this per service. If there is no map, nobody knows, and writing the map is the first job.

escalation policy example for a small team

Same fix: shrink the ladder to two levels (primary, then everyone) but keep the minute marks. A 5-person team still needs written timing.

how to escalate an incident to management

Same fix: the T+15 manager escalation rule defines when management gets pulled in, which stops both premature escalations and late ones.

Why it happens

Pages go to the wrong person because ownership lives in peoples heads, not in a doc. The 10 minutes spent figuring out who owns a service are the most expensive minutes of any incident, so writing it down once pays off every single time.

Edge cases and pitfalls

• Cross-team services with two owners: pick one primary. "Both own it" means nobody responds.
• People who cant carry a pager (contractors, some regions) need a named fallback, not a silent gap.
• Update the map when teams reorganize. A stale escalation policy is worse than none because people trust it.
• Time zones matter for global teams: the ladder should name rotations, not people, so it works while someone sleeps.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_jcOPJlwymunMpA-whwbruQ