Advice about on-call generally assumes an organisation with enough people to run follow-the-sun coverage. If you are four engineers keeping a product alive, most of it does not apply.
Weekly is usually right
Daily rotations fragment context badly and keep everyone in a state of low-grade readiness. Weekly gives you one week of disruption and three weeks of genuine freedom. Set the handover on a weekday morning, not Friday evening — the outgoing person is awake to explain what happened, and nobody inherits an unexplained mess at the start of a weekend.
The single-expert problem
Small teams almost always have one person who knows the payment integration. If they are not on call, whoever is escalates to them anyway; if they are always on call, they burn out. The fix is to reduce the concentration: write the runbook during the incident, pair on first response for unfamiliar areas with the expert not touching the keyboard, and rotate who takes the postmortem.
Escalation needs a human backstop
A minimum viable chain has three or four levels and ends with someone different: a quiet channel immediately, an SMS after five minutes, a phone call after ten, a call to a second person after fifteen. That last level covers the case where the primary is asleep with their phone on silent — without it, a single unreachable person means nobody responds.
Acknowledging is a commitment, not a receipt
If your team treats acknowledgement as "I have seen this," someone will acknowledge from their phone while walking into a meeting, the chain stops, and the incident sits untouched. Make it explicit: acknowledging means "I am working on this now."
Compensate it
At four people, on-call is a meaningful imposition on a quarter of everyone's life. A flat payment per week, time off in lieu after a disturbed night, or a policy that anyone woken after midnight starts the next day at noon. The last costs almost nothing and does more for morale than most perks.
Keep the paging surface tiny
Every false page consumes a meaningful fraction of a small team's tolerance. The test: is there something a human must do right now? A small team with five well-chosen paging alerts will respond better than a large team with two hundred.
Write down the boring things
Where the production logs are. How to get VPN access at 3am. How to roll back a deploy. Who to call at the payment provider. Every one of these is trivial when you know it and a twenty minute detour when you do not.