Most businesses only discover their backup strategy was broken the day they need it. A real backup strategy follows the 3-2-1 rule, picks the right backup type for the job, runs on a schedule that matches how fast your data changes, and is tested by actually restoring it.
The 3-2-1 rule, in practice
The 3-2-1 rule is the baseline every serious backup plan should meet: keep three copies of your data, on two different types of storage, with one copy offsite. It sounds simple, but it's the difference between a bad day and a business-ending one.
- 3 copies: the live data plus two backups — never rely on a single copy of anything.
- 2 media types: for example, local disk snapshots and cloud object storage, so one failure mode can't take out both.
- 1 offsite: a copy outside your primary data center, so fire, theft or a regional outage can't destroy everything at once.
Full, incremental and differential backups
Not every backup needs to copy everything. Full backups capture the entire dataset and are the simplest to restore from, but are slow and storage-heavy if run often. Incremental backups save only what changed since the last backup of any kind — fast and compact, but a restore has to replay the whole chain in order. Differential backups save everything changed since the last full backup — a middle ground: bigger than incrementals but a restore only needs the last full plus the last differential.
Most practical setups combine them: a weekly full backup with daily incrementals or differentials in between, which keeps both storage costs and restore time reasonable.
How often, and where
Backup frequency should match how much data you can afford to lose, not a generic schedule. A marketing site that changes monthly is fine with daily backups. An order database that takes payments every minute needs much tighter intervals — sometimes continuous replication plus hourly snapshots.
Untested backups are not backups
A backup job that "completes successfully" every night can still be useless — corrupted archives, missing dependencies, or credentials that quietly expired are common and often invisible until restore day. We treat an untested backup as an assumption, not a guarantee, and schedule periodic real restores into a separate environment to confirm the data and the services on top of it actually come back to life. That's paired with reliable, redundant infrastructure underneath, so backups are a safety net, not your only line of defense.