Alert escalation policies: who gets paged, and when
How to design an escalation policy: severity levels, acknowledgment timeouts, on-call rotations and backups, with an example you can adapt to a small team.
· 6 min read · By the Spot Downtime team
An alert that nobody sees is the same as no alert. An escalation policy is the written answer to a simple question: when something breaks, who is told first, and who is told next if the first person does not respond?
Small teams often skip this because "everyone is in the Slack channel". That works until the outage happens at 3 a.m., or the one person who watches the channel is on a plane. This guide covers how to build a policy that is simple enough to follow under stress.
The parts of an escalation policy
- Trigger. What starts it: a confirmed outage, a failed cron job, a certificate about to expire.
- Severity. How bad it is, which decides how loud the notification is.
- Primary responder. The person or rotation notified first.
- Acknowledgment window. How long they have to say "I have this" before the next level is notified.
- Next levels. A backup, then a lead or manager, each with their own window.
- Channels. How each person is reached: push, SMS, phone call.
Start with severity, not with people
If every alert pages someone, people stop trusting alerts. Decide first which events deserve to wake a human. A short scale is enough:
- Critical. Customers cannot use the core product: the site is down, checkout fails, the API returns errors. Page immediately, any hour.
- Major. A feature is broken or badly degraded, but there is a workaround. Notify during working hours, page if it lasts.
- Minor. Something needs attention but nothing is broken yet: a certificate expiring in two weeks, a slow endpoint. Goes to a channel or ticket queue, never a phone.
Fewer false pages
Design the levels
A common shape for a small team, with timings you should adjust to how fast you must respond:
Critical alert fires
0 min Notify primary on-call (push + SMS)
5 min No acknowledgment -> call primary on-call
10 min Still none -> notify secondary on-call (push + call)
20 min Still none -> notify engineering lead (call)
Major alert fires
0 min Post to #incidents (and notify primary in working hours)
30 min Not acknowledged -> escalate to criticalKeep acknowledgment windows short
Five minutes for the first level is typical for customer-facing outages. Longer windows mean longer outages. Shorter windows mean more needless escalations when someone is simply typing a reply, so pick a number your team can meet.
Change the channel as you escalate
A push notification is easy to miss. If the first level does not respond, the next step should be something harder to ignore, such as a phone call. Each person should have at least two ways of being reached.
Always have a final fallback
The last level should be someone who is accountable for the system, such as a lead or owner. The goal is that no alert can fall through all levels without a human seeing it.
On-call rotations
- One primary and one secondary at a time. Two people means one can be in a lift or asleep without the alert being lost.
- Hand-offs at a fixed time. Weekly rotations that change on the same weekday and hour avoid gaps.
- Swaps are allowed. Holiday, illness and travel will happen. Make swapping easy and visible.
- Compensate and limit. Being on call is work. Rotate fairly and give time back after a bad night.
If you are a two- or three-person team, the rotation can be as simple as a shared calendar plus a rule that the primary acknowledges within five minutes.
What happens after acknowledgment
Acknowledging stops the escalation, but it starts a responsibility. Write down what the responder does next:
- Acknowledge, so the team knows it is owned.
- Confirm the problem is real and note the time it began.
- Post a first update to customers if it is user-facing. See incident status updates.
- Pull in help early if you are not making progress. Escalating is not a failure.
- Record what happened for a review. Use a post-incident review template.
Review the policy
After every real incident, ask: did the right person get the alert, how long did it take to acknowledge, and did any level add delay without adding value? Remove noisy alerts and adjust windows. A policy that is never reviewed drifts away from how the team actually works.
Setting this up with Spot Downtime
When a check fails, Spot Downtime confirms it from several regions and then notifies the alert channels in your workspace. It sends one alert when a monitor goes down and one when it recovers, rather than repeating while it stays down. Channels include Slack, Microsoft Teams, email, SMS, voice calls and PagerDuty, and you can send a test from any channel. See alerts for the full list.
Spot Downtime does not run timed escalation levels itself. If you need acknowledgment timeouts and rotating schedules, route alerts into a dedicated on-call tool through the PagerDuty channel or a webhook and define the policy there. If you do not, a simpler setup works for many small teams: a chat channel for visibility plus SMS or voice-call channels for the people who must respond.
Downtime cost calculatorEstimate what an hour of downtime costs, to help decide how fast you need to respond.Get new guides by email
Practical posts on monitoring, outages, error fixes and on-call, when they're published. No spam.
Keep reading
- Alerts · On-callUptime alerts your team won't ignore: how to cut false alarmsWhen alerts cry wolf, people stop reading them, and then miss the real outage. Seven practical ways to make every uptime alert worth acting on.October 2, 2026 · 4 min read
- Alerts · MonitoringWebsite monitoring apps: getting outage alerts on your phonePush notifications, SMS, calls or chat apps: the ways a website monitoring app can reach your phone, which ones wake you up at 3 a.m., and how to set them up so you don't miss an outage.November 3, 2026 · 3 min read