How to set up an on-call rotation for a small team
Decide who is paged this week, who backs them up, and what happens when nobody answers, without buying an incident management suite.
Run the response · 4 min read · Updated Oct 12, 2026
In a team of three, alerts usually go to everyone. That feels safe and works badly. At 3 a.m. each person assumes one of the others has it. Or the same person always picks it up, and after a few months they are the one who is tired and looking for another job.
On-call is the agreement that fixes this: at any moment one named person owns the alert, everyone else can sleep, and the duty moves round.
How it works
Three things together make a rotation.
- A schedule says who is on call and when.
- A backup says who is asked next.
- An escalation says what happens when an alert is not acknowledged: after a set time, the next person is paged.
The third part is what makes it safe for one person to carry the phone. If they are in the shower or out of signal, the alert does not stop with them.
Step 1: Agree the rules before the tool
Decide these in a conversation, not in a settings screen:
- How long a shift is. A week is common. For two people, half-weeks are easier to live with.
- When it changes hands. In working hours, on a weekday.
- What is worth a page. Only what needs action tonight. Everything else goes to chat and waits for morning.
- How fast "answered" is. Fifteen minutes is a reasonable promise. It decides how far from a laptop the person can be.
- What on-call gives back. Time off after a bad night, at the least. A rotation nobody is thanked for does not last.
Put the handover in working hours. A change of person at 09:00 on a weekday is noticed; one at midnight is not.
Step 2: Draw the rotation
Write down who is first and who is second for every day of the week, so that nobody holds both on the same day.
With only two people, each is the other's backup. With three or more, the backup is a good place for the person who was first last week: they know what is fragile.
Step 3: Decide what happens when nobody answers
An alert that waits on one person is a single point of failure. Give it a path:
The delays are a trade. Short ones wake the backup for things the first person was about to answer. Long ones leave an outage unattended. Fifteen minutes a step is a sound start.
The path only works if answering is explicit. Acknowledging an alert means "I have it", stops the steps after it, and is a promise to look, not a claim that it is fixed.
Step 4: Make sure a page wakes someone
Email and a chat message do not wake anyone. Before their first shift, each person needs one channel that does, such as a phone call, a text message, or a push notification allowed through their phone's do-not-disturb, and needs to have heard it go off.
This is each person's own setup, on their own phone. It is the step most often skipped.
Step 5: Plan for holidays and swaps
People swap shifts, fall ill and go on holiday. A rotation needs a way to say "this week, someone else" without rebuilding the schedule. Agree how a swap is recorded, and that the person covering confirms it.
Test it
Do this in working hours, with the team warned.
- Trigger a real alert, for example with a check on a page that does not exist.
- Do not acknowledge. See the first person paged, and check that their phone made a noise.
- Still do not acknowledge. See the backup paged.
- Acknowledge, and remove the check.
Common mistakes
- Paging for things that can wait. Every unnecessary page teaches people to ignore the phone.
- Nobody on call for a day. Check the week ahead for gaps, especially around holidays.
- Relying on email to wake someone. It will not.
- Putting someone on call who cannot act. They need the access to fix things, or at least to acknowledge and call for help.
- No handover. Five minutes at the change of shift on what happened and what is fragile saves an hour at night.
Doing this in saysOK
The On-call page holds two schedules, Primary and Secondary. Each assigns one person to each day of the week, with a shift start and end, and repeats weekly. An override hands a schedule to someone else for up to 14 days, which covers holidays and swaps.
A schedule alone pages nobody. In an alert rule, an Escalation step set to Primary on-call or Secondary on-call pages whoever holds that level, 15 minutes after an alert that is still not acknowledged; each further step waits another 15. Acknowledge on the incident stops the steps that remain.
A page arrives as an email and a notification. Each person can add push notifications, and a phone call or WhatsApp message to their own number, under Your account. Two things to know: a viewer can be paged but cannot acknowledge, and an incident declared by hand notifies channels, not people.
See pricing for how many seats each plan includes, and for calls and messages.
Next
Paging the right person is half of handling an incident. The other half is learning from it: see how to write an incident postmortem. And if scheduled jobs are among the things that page you, see how to monitor a cron job.
Set this up in saysOK
Start on the free plan. The guide says where a step needs a paid one.