Every IT engineer has met the office whose backup drive was full, whose backup job died months ago, or whose backup folder was encrypted by the same ransomware as the server. The backup was never the control. The restore is. This page explains the difference and the small routine that proves your safety net actually holds.
Backup software is optimistic by design: it shows green ticks for jobs that copied something. None of that answers the only question that matters on the bad day - can we actually get working data back, and how long does it take? A test restore answers it while the stakes are low. Restore a handful of real files to a spare folder, open them, time it. Do the same once for the accounting data and once for a full system image if you keep one. This guide covers what a small office backup must include, where copies must live so one incident cannot take them all, and the quarterly test that separates a real backup from a comforting icon.
The classic rule still works: keep three copies of important data, on two different kinds of storage, with one copy off-site or offline. In a small office that translates plainly: the live data on the server or PCs, a nightly backup to a local NAS or external drive, and a third copy in cloud backup storage or on a rotated drive that goes home or to a safe.
The offline copy is the ransomware answer. Modern ransomware looks for connected backups and deletes or encrypts them first. A copy the network cannot reach - unplugged drive, cloud backup with its own separate login and MFA - survives the worst day intact.
Once a quarter, pick real files at random - not the same test file every time - and restore them to a test folder. Open them. Check the accounting file opens in the accounting software. Time the whole exercise. Then write one line in the log: date, what was restored, whether it worked, how long it took.
Once a year, go further: restore the server or a key PC image to spare hardware or a virtual machine and confirm it boots. Businesses are routinely shocked by the honest answer to how long a full rebuild takes - finding out during a drill is education; finding out during an incident is disaster.
The backup console and the NAS admin page are high-value targets. Give them unique passwords in the password manager and MFA where offered, and never use the same credentials as the daily Windows admin account. If the person who manages backups leaves, change those passwords the same week.
Also watch capacity and failures with fresh eyes monthly: a backup drive that filled up in June has been silently failing since June, no matter what the icon looked like in May. Most backup tools can email a failure alert - turn that on and send it to someone who reads it.
File backup does not automatically include everything an office assumes. Cloud email and documents may need their own backup product, because a deleted mailbox is only recoverable from the provider for a limited window. A file-level backup of the server does not restore the server itself quickly - that needs an image backup.
Write down, in plain language, what is backed up, where, how often and how far back you can go. If any answer surprises the owner, the time to fix it is now, in the calm, not during the incident.
Every essential in this guide is explained with full knowledge and installation in the IT Security Essentials video.
Watch Full VideoA small file-level test every quarter is realistic for most offices, plus one full system restore drill a year. Test immediately after any change too - new backup software, new server, new storage - because change is when silent failures start.
Cloud backup gives you the off-site copy and survives fire, theft and ransomware on the local network. But restoring a whole server from the cloud can take a long time on an ordinary connection. Most offices pair it with a fast local copy for everyday restores and keep the cloud copy for the true disaster.
It proves the job ran and copied data. It does not prove the data opens, the credentials still work, or the restore finishes in a usable time. Only an actual restore proves those - which is why the test, not the green tick, is the control.