Backup strategies are usually configured once during setup and then trusted indefinitely. The failure modes are quiet: an excluded table, an expired credential, a job that has been failing for months into a channel nobody reads. The only way to know is to restore.

Start with two numbers

How much data can the business afford to lose, and how long can it afford to be down? Those two figures determine everything else — backup frequency, whether you need point-in-time recovery, and how much you should spend. Without them, backup design is guesswork dressed as configuration.

Point-in-time recovery matters more than people expect

Nightly snapshots protect against infrastructure failure. They do not protect well against the far more common scenario: a bad migration or an accidental deletion at 3pm, discovered at 4pm. Point-in-time recovery lets you restore to just before the event rather than losing the day.

Keep a copy somewhere separate

A backup in the same account with the same credentials as production is not protection against a compromise of that account, and ransomware operators specifically target backups. Keep at least one copy in a separate account or provider, ideally with immutability or object lock so it cannot be deleted within its retention window.

Test the restore, and time it

  • Restore into a clean environment on a schedule — quarterly is a reasonable minimum.
  • Verify the data is complete and the application actually runs against it.
  • Record how long the whole process takes, end to end.
  • Write down the procedure so it is not dependent on one person's memory.

Alert on failure, not success

Success notifications are ignored within a fortnight. Alert only on failure, and separately on the absence of a successful backup within the expected window — a job that stops running produces no failures at all, which is the failure mode people miss.

Watch what backups contain

Database backups contain everything, including personal data that retention policies were supposed to have removed. Include backups in your retention thinking, set expiry, and ensure copies restored into staging are treated with the same care as production data.