Articles · Backups

2026-11-05 11 min EN / FR

Backups that actually restore

A backup you have never restored is a hope, not a backup. Most people discover whether theirs works at the worst possible moment, and a surprising number discover that it does not.

The ways backups fail

  • The backup job stopped months ago and nobody noticed
  • The backup is stored on the same server that just failed
  • The backup is incomplete (database yes, uploaded files no)
  • The files are corrupted or empty
  • Nobody knows the encryption password
  • Restoring takes three days, and the business cannot wait that long
  • The backup contains yesterday's ransomware infection
  • The only person who knew how to restore left the company

The 3-2-1 rule

A simple standard that covers most risks:

  • 3 copies of your data (the live one plus two backups)
  • 2 different types of storage (for example, a server's disk and cloud storage)
  • 1 copy off-site, in another location and under a different account

Modern advice adds: keep at least one copy that cannot be modified or deleted from your main systems (immutable or offline), so ransomware or a compromised account cannot wipe it.

Step 1: Decide what needs backing up

Think in terms of data you cannot recreate:

  • Databases (the most important item for most businesses)
  • Uploaded files and documents
  • Configuration: server settings, environment variables, secrets (stored securely)
  • Source code (it should already be in version control, which counts as a copy)
  • Email, if you host it yourself
  • SaaS data: many people assume their cloud tools back everything up for them. Often they offer limited recovery windows, and not for your mistakes. Export what matters (see leaving a SaaS without losing your data)

Also decide what you do not need, such as caches, temporary files and things that can be rebuilt from code.

Step 2: Set two numbers

Two questions turn "we have backups" into a real plan:

  • How much data can we afford to lose? If the answer is one hour, daily backups are not enough. This is the recovery point objective.
  • How long can we be down? If the answer is four hours, a restore that takes two days is a failure. This is the recovery time objective.

For a small business the answers might be "one day" and "one day", and that is fine. What matters is that they are known and the setup matches.

Step 3: Automate and monitor

  • Backups run automatically, never from memory
  • Each run reports success or failure, and failures reach a real person
  • A "no news" setup is dangerous: configure an alert when a backup has not run, not only when it errors
  • Check backup size over time. A sudden drop to nearly zero is a red flag

Want a restore drill?

We can review what you back up, where copies live, and run a restore in a test environment with you.

Step 4: Protect the backups

  • Encrypt them, and store the key separately from the backups (in your password manager, with at least two people able to access it)
  • Use a separate account or credentials for backup storage, so a compromise of the main system cannot delete them
  • Limit who can delete backups, and enable retention rules
  • Keep several generations: daily for a few weeks, weekly for a few months, monthly for longer. A problem is often noticed days after it started

Step 5: Test the restore (the part everyone skips)

At least twice a year, and after any major change:

  1. Choose a backup at random, not the latest one every time.
  2. Restore it to a clean, separate environment, never over production.
  3. Follow your written procedure, ideally performed by someone who did not write it.
  4. Verify the result: can you log in, are recent records present, are files and images intact, do key features work?
  5. Time it. Compare against your recovery time target.
  6. Write down what went wrong and fix the procedure.

The first test nearly always reveals a gap. That is the point. It is much cheaper to find it now. The same idea sits in an exit plan: document backups, then prove the restore.

Step 6: Write a one-page restore procedure

It should be usable by someone stressed and in a hurry:

  • Where the backups are, and how to get access
  • Where the encryption keys are
  • The exact steps to rebuild a server and restore data
  • Who to call, and who decides
  • Expected duration
  • How to verify success
  • Date of the last successful test

Keep a copy somewhere that does not depend on the system that might be down. A procedure stored only on the failed server is not much use.

Special cases

Databases: copying database files while the database is running can produce an unusable backup. Use the database's own export or backup tool, or a method that guarantees a consistent snapshot.

Virtual machine snapshots: convenient, but usually stored with the same provider and sometimes in the same place as the machine. Treat them as one layer, not the whole plan.

Hosting provider backups: useful, but check how long they keep them, where they are stored, whether you can download them, and what happens if your account is suspended. When you switch hosts, set backups and a restore test on the new side before you cancel the old one.

Ransomware: make sure at least one backup copy is out of reach of your normal credentials, and test that you can restore from it.

A small-company setup that works

  • Daily automated database export and file backup
  • Copy sent to a second provider, under a separate account
  • Weekly copy kept for two months, monthly copy for a year
  • Alerts if a backup is missing
  • A test restore every six months, with the date recorded
  • A one-page procedure, known to at least two people

None of this is expensive. It is mostly a matter of doing it once, writing it down and checking it regularly.

Quick audit

  • I know what is backed up and what is not
  • At least one copy is outside my main provider
  • Someone gets an alert if a backup fails or does not run
  • I can find the encryption key without the person who set it up
  • I have restored a backup successfully in the last six months
  • I know how long a full restore takes

Every "no" is the next task on your list.

Related

From hope to a dated restore test

If you want a second pair of eyes on RPO, RTO and where copies live, write us.