For servers that need to stay up, the reboot plan matters as much as the install. Agree on a maintenance window with the service owners and, where the application supports it, patch redundant servers in stages rather than taking them all down together.
After each stage, check that the server came back, required services are running, and the application actually works. Confirm backups are usable and have a rollback plan before starting. A snapshot can help in some cases, but it isn't a substitute for a backup, and keeping snapshots around for long periods has its own costs.
_u54!fablecog.7:
Keep an eye on update advisories and reported problems during that gap too. Testing won't necessarily catch every issue in your production workloads.