Testing That Your Backups Actually Restore
A backup you have never restored is a guess. Here is a 30-minute restore test a small business can run every month, and the failures it reliably uncovers.

The backup console had shown a green tick every morning for fourteen months. When the business finally needed it, the restore ran for six hours and produced a folder tree with every file present and every file zero bytes long. The job had been completing. It had been completing successfully on nothing.
Variations of this happen constantly, and they all share a cause. Nobody had ever taken data out. A backup system is only half a system until you have proved the second half works, and the only way to prove it is to restore something and open it.
What a green tick does and does not tell you
A successful backup job means the software ran, reached its destination and wrote what it believed it was asked to write. It does not tell you whether the right folders were in scope, whether a database was captured in a consistent state, whether your retention window goes back far enough, whether anyone still knows the password to the backup console, or how long a real recovery would take.
Many products offer a verification or integrity check. Turn it on, by all means. It confirms the stored data is readable and matches a checksum. It does not confirm that the data is what your business needs, or that your people can get at it under pressure.
The four questions a test has to answer
A useful restore test is not a technical exercise so much as four questions with recorded answers.
- Is it there? Can you find the file, mailbox or database you went looking for, at the date you expected?
- Does it open? Not just exist. Open the spreadsheet, read the email, query the database.
- How long did it take? From deciding to restore to having usable data. Write the number down.
- Could someone else have done it? If the answer depends on one person's memory, that is a finding.
The monthly test: thirty minutes
Keep this small enough that it survives a busy month. Pick something different each time so you gradually cover everything.
- Choose a target in advance. A folder from the shared drive, a user's mail from two weeks ago, a single accounting file. Rotate through your systems month by month.
- Pick a date, not the latest copy. Ask for the version from, say, 21 days ago. This is what exercises retention, which is the setting most often wrong.
- Restore to a separate location. A new folder, a spare laptop, a test mailbox. Never over the live data.
- Open the result. Check file sizes look sane, open two or three documents, confirm the content is from the date you asked for rather than today.
- Log four lines. Date, what you restored, how long, what went wrong.
- Delete the restored copy. A test restore sitting in a shared folder for a year is a data problem of its own.
Thirty minutes a month will catch almost every silent failure that matters.
The quarterly test: restore something whole
Once a quarter, go bigger, because small restores hide the problems of scale. Rebuild a full user mailbox into a test account. Restore an entire project directory. If you run a server, restore its image to spare hardware or a virtual machine and see whether it boots.
This is where the expensive surprises show up: a restore that would take four days over your office internet connection; an application that will not start because the licence is tied to hardware that no longer exists; a database that restores but will not open because the backup was taken while it was running; an encrypted backup whose key was only ever stored on the machine being recovered.
Each of those is cheap to fix when discovered on a Tuesday afternoon with nothing at stake. Each is close to unfixable when discovered during a ransomware incident.
Compare two numbers. How long you can afford to be without a system, and how long the restore actually took. If the second is bigger than the first, you have a decision to make about faster recovery, a local cached copy, or a lower expectation. Pretending the gap is not there is the one option that fails on the day.
The annual test: pretend you have lost the office
Once a year, run the scenario rather than the task. Assume the office network is untrusted and every device has been reimaged. Now ask: where is the backup, who has the credentials, which system comes back first, who tells customers, and how do you work in the meantime?
You do not have to perform the whole recovery. Walk through it on paper with the people who would be involved, which takes about ninety minutes and surfaces the gaps between your backup system and your incident response plan. The NCSC publishes free tabletop exercise material for exactly this, and a ransomware scenario is the obvious one to use.
The recurring finding is not technical. It is that the recovery order was never agreed. Three people each restore the thing they think is most urgent, and the business spends its first day of recovery working on the wrong systems.
Build the log, because the log is the control
Keep one simple table: date, system tested, restore point requested, time taken, outcome, notes. One row per test. After a year you have twelve rows, and they tell you things no console does. Which systems have never been tested. Whether restore times are getting worse as data grows. Whether the same problem keeps recurring.
That table is also the artefact other people want. Insurers ask whether backups are tested. Larger customers ask it in supplier questionnaires. A dated log answers the question in one line, which is a surprisingly good return on thirty minutes a month. See measuring security progress for how it fits alongside your other simple measures, and cyber insurance for small business for how insurers phrase it.
Who should do it
Rotate. The person who built the backup system knows the shortcuts and will unconsciously avoid the rough edges. Have someone else attempt a restore from the written instructions at least twice a year. If they cannot, the instructions are the problem, and fixing them now is far cheaper than discovering it while your backup owner is on a flight.
Had a suspicious email during a recovery? Attackers follow up incidents with fake support and fake invoices. Paste anything doubtful into CyberWatch AI Scan for a free check.
Testing only makes sense on top of a sound design, so start with a backup strategy you will keep to and the fuller guide to backup and recovery. The rest of the picture is in our complete guide to small business cybersecurity.
Frequently asked questions
How often should we test a restore?
Restore something small every month, and do one larger test of a whole system or mailbox every quarter. Monthly is frequent enough that a silent failure gets caught within weeks rather than discovered during an incident.
What counts as a real test?
Taking data out of the backup and opening it. A green tick in the backup console means the job ran, not that the data is usable. Verification features check integrity, which is useful but not the same as proving you can get your work back.
What is a recovery time objective?
The longest outage you can tolerate for a given system before the damage becomes serious. You compare it against how long your restore actually took, and if the real number is larger, either the plan or the expectation has to change.
Should we restore to the original location?
No. Restore to a separate folder, a spare machine or a test account. Restoring over live data during a test risks overwriting good files with older versions.
What do restore tests usually find?
Missing systems that were never in scope, backups running but excluding a key folder, retention windows shorter than assumed, nobody knowing the credentials for the backup console, and restores that take far longer than anyone expected.
Who should run the test?
Occasionally someone other than the person who set the backups up. If only one person can perform a restore, your recovery plan depends on that person being reachable on a bad day.
Do we need to test cloud backups too?
Yes, and particularly those. Restoring a mailbox or a shared drive from a cloud backup service involves steps and permissions nobody has rehearsed, and the first attempt is always slower than you expect.
What should we write down after a test?
Date, what was restored, who did it, how long it took, and anything that went wrong. Four lines. That log is also the evidence insurers and larger customers ask for.
Sources
- SP 800-34 Rev. 1, Contingency Planning Guide for Federal Information Systems, US National Institute of Standards and Technology
- #StopRansomware Guide, US Cybersecurity and Infrastructure Security Agency
- Backing up your data, UK National Cyber Security Centre
- Offline backups in an online world, UK National Cyber Security Centre
- Exercise in a Box, UK National Cyber Security Centre
- SP 800-61 Rev. 3, Incident Response Recommendations and Considerations for Cybersecurity Risk Management, US National Institute of Standards and Technology
- Small Business Cybersecurity Corner, US National Institute of Standards and Technology


