On-call means one named person is responsible for responding when something breaks, at any hour if needed. Escalation is what happens when that person doesn't answer. Without either, alerts land in a shared channel that nobody owns and problems wait until a customer complains.
Rotations
A rotation passes on-call duty around the team on a schedule, so nobody carries the pager indefinitely. A weekly handover is enough to start. Larger teams run more than one layer, such as a primary and a backup, or a follow-the-sun schedule across time zones so whoever is on call is awake.
Escalation policies
An escalation policy is the list of steps an alert goes through until someone acknowledges it. A typical one pages the on-call person. If they don't acknowledge within a few minutes, it pages them again or pages the backup. If there is still no answer, it pages a wider group or a manager. Each step has a wait, so a phone left in another room doesn't leave an incident unanswered.
Acknowledgement
Acknowledging an alert tells the system someone has it, and the escalation stops there. That's what makes it safe to page quickly: later steps only fire if nobody has responded.
Severity and channels
A minor slowdown doesn't need a phone call at 3am. Route by severity: a major outage can call and text, while a small degradation might only go to a chat channel or email.
Coverage
People go on holiday, so let someone pause their own paging while the rest of the rotation covers for them. Look for gaps where the schedule has nobody on call, and make sure an unacknowledged page eventually reaches a person. Every page that wasn't worth waking up for makes the next real one easier to ignore.