Security2026-11-16·8 min read

How Often Should You Test Your Backups?

Restore one real file every quarter and rehearse the full recovery once a year. What a realistic backup test looks like for an Australian small business.

Restore something real once a quarter, and rehearse the whole recovery once a year. That is the honest answer for a business of one to thirty staff, and it has three parts:

Restore something real once a quarter, and rehearse the whole recovery once a year. That is the honest answer for a business of one to thirty staff, and it has three parts:

  • Every quarter, restore one file that matters. Not a test document. This quarter's invoices, the customer list. Open it on a different machine, and time it. Fifteen minutes.
  • Every year, rehearse the full recovery. The ACSC's Small Business Cloud Security Guide is explicit that a real test covers the entire backup rather than a subset of data or a spot check.
  • After every change. A new laptop, a new provider, a reorganised folder. Change is what silently breaks coverage, and the backup software will not mention it.

The ASD's Essential Eight maturity model asks, even at its lowest level, that restoration is tested as part of disaster recovery exercises rather than against a calendar. A backup job that reports success proves the job ran. Only a restore proves the data comes back.

Why is there no single correct interval?

Because the interval is set by what you can afford to lose, not by a rule of thumb. The ASD states the requirement as backups being performed and retained "in accordance with business criticality and business continuity requirements" (ASD Essential Eight maturity model, 2026). That is deliberately not a schedule. It is a dependency.

Work it backwards. If losing a day of your accounting file would be survivable but losing a week would not, your backup needs to run nightly and you need to know, more than once a year, that nightly restore actually works. If the only thing you would truly miss is a folder of photos, quarterly is generous. Those two numbers have names, and they are the whole basis of the decision: they are explained in RPO and RTO in plain English.

Here is the part that gets missed. The test interval is not the same question as the backup interval, and most owners set the first by copying the second. A nightly backup tested once every three years is a nightly hope. The test interval answers a different question: how long are you willing to be wrong for? Our view is that a quarter is about as long as that should be, because a backup can fail quietly for a quarter without anything in a small business giving anyone a reason to look.

What does a realistic test look like for 1 to 30 staff?

Small enough to actually happen. In our experience backups go untested mostly because the owner has imagined a test that takes a weekend, and so it never gets booked.

The quarterly one, fifteen minutes. Pick a file you would genuinely miss. Restore it from the backup, not from the recycle bin and not from the sync folder. Open it on a machine other than the one it came from. Write down the date, how long it took, and where you got stuck. That last one is the point of the exercise.

The annual one, half a day. Restore enough to prove you could keep trading: a mailbox, the accounting data, the shared folder your team works out of. Check the data is complete and usable rather than merely present. This is the test the ACSC is describing when it says the entire backup should be tested, and it is the one that catches the failures nobody sees coming.

The unscheduled ones. Any time you change device, provider, or folder structure. Ten minutes, same drill.

Book them. A test that lives in the calendar happens; a test that lives in an intention does not. There is a one-screen version of this cadence you can hand to whoever will actually run it: how often should I test my backups.

What goes wrong that only a restore ever finds?

These are the failure modes we see, and the reason the interval matters at all. Every one of them looks like a working backup right up until the moment it is needed.

The backup that stopped months ago. A credential expired, a drive filled, an update changed a setting, and the job has been failing quietly ever since. Failure notifications are usually configured to go to whoever set the system up, which is often a person or an address the business no longer has. The green tick you remember seeing was real. It was just a while ago.

The backup that faithfully copied the damage. If the arrangement is a synchronised folder rather than a backup, it mirrors your mistakes at the speed of your internet connection. Delete a file and the copy goes; encrypt it with ransomware and the encrypted version syncs up. This one deserves its own answer, because it is the misunderstanding we see most often: does cloud storage count as a backup. The defence is a copy the attacker cannot rewrite, which is what an immutable backup means.

The backup nobody can log in to. The console is tied to a personal account belonging to someone who left, or to an authenticator app on a phone that has been replaced. The data may be perfectly intact and completely unreachable. This is not an exotic scenario; it is one of the ordinary ways small businesses lose access to their own systems, and it costs nothing to fix before it happens.

The backup that comes back incomplete. The ASD puts this plainly in its Small Business Cloud Security Guide: there can be synchronisation issues or critical interdependencies that are unknown until a full backup is tested (ASD, Technical example: Regular backups, 2026). The files arrive. The system that made them mean something does not.

What does your backup dashboard actually tell you?

It tells you the job ran. That is a genuinely useful thing to know and it is not the thing you need to know.

The same gap shows up in the admin console you probably trust most. Microsoft's own documentation is clear that when you restore a deleted Microsoft 365 user "within 30 days of deleting it, the account and all associated data are restored" (Microsoft Learn, Restore a user, 2026). Read that as the promise it is: thirty days, then the data is gone. Nobody is hiding it. It is simply that a deletion discovered in week six reads exactly like a deletion discovered in week one until you go looking. The same shape of limit applies to Google Workspace and to your accounting platform, which is why Microsoft 365 being backed up by Microsoft is a different claim from your data being recoverable.

A restore test is how you find out which of your systems have a window like that, and how wide it is.

What if your setup is different from that?

The cadence holds, but what you restore changes.

If everything is in the cloud, you still test, and the thing you are testing is your third-party backup of those platforms rather than the platforms themselves. If you do not have one, the annual rehearsal is where you discover the retention window is your real recovery policy.

If an IT provider handles it, ask them for the date of the last successful restore test and what was restored. Not the backup report. The restore. A good provider will have the answer and will be pleased to be asked; if the answer is vague, you have learned something for the price of an email.

If you are one person with a laptop and a drive, the quarterly restore is more important, not less, because there is nobody else who would notice the job failing. Fifteen minutes, four times a year.

If you have a contractual or regulatory recovery obligation, the interval is whatever that obligation says, and testing is evidence you met it rather than a nice habit.

Restore testing is one item on a longer list of things that only look fine until they are checked, which is the same reason the Essential Eight puts regular backups alongside patching and multi-factor authentication rather than at the end.

When is this answer not enough?

When downtime is measured in hours rather than days. A quarterly file restore is a floor. If your business cannot trade for even half a day without material loss, you need a documented recovery procedure, a rehearsed one, and probably someone to help you design it, because the gap between "we have backups" and "we can be trading by lunchtime" is filled with decisions rather than products. Start with what goes in a business continuity plan, and read the whole picture in our guide to business continuity for small business.

It is also worth being clear about what we do not cover. The Red Bridge Cyber scan looks at your public-facing posture, which means your domains, your email authentication and your website. It cannot see inside your backup system, and it does not test, monitor or verify that your data can be restored. No external scan can. That test only exists if someone in your business does it.

So do the small one. Pick a file you would miss, restore it today, and write down how long it took. Our founder, Adam Burgess, has written about what an untested backup is really worth, and the short version is that you own a green tick and a feeling until the day you have done this once.


Sources

#security#australian-business#small-business