The backup nobody tests
Almost everyone has backups. Far fewer have ever restored one. An untested backup is a hope with a filename — you find out whether it works on the worst possible day.
Set up on day one
When the agent provisions a system, backups and monitoring go in with it, not "later". Scheduled backups, pulled off the box so a server failure can't take them with it. Health and uptime checks that tell you before your customers do.
Rehearse the recovery
The step that makes it real is the restore drill: stand the backup up somewhere clean and confirm it actually comes back. We reboot-tested and restore-tested our own recovery for exactly this reason — a runbook you haven't run is fiction.
Boring on purpose
None of this is exciting, which is the point. Monitoring, off-site backups and a recovery you've actually rehearsed are what turn "the server died" from a catastrophe into an inconvenience.