
The 3-2-1 backup rule: which copy lives where, and why it needs a 0 on the end
The 3-2-1 backup rule fits on a beer mat: three copies, on two types of storage, one of them off-site. What does not fit is which copy that actually is in your organisation, once mail, files and Teams live in Microsoft 365 and a few servers are still running in the comms cupboard. This is that translation, with the two digits that ransomware has added to the end.
Three copies means the original plus two backups, on two types of storage that fail for different reasons, with one of them in another location.
Microsoft 365 is production, not a backup: whatever an account with the right permissions deletes or encrypts disappears everywhere at once, including from the safety net inside your own tenant.
The extra 1 is a copy nobody can erase, not even your own administrator, and the 0 is a restore test you have actually carried out and timed.
What the 3-2-1 backup rule actually says
The rule counts copies of the same data. Three: the original you work with counts, so that is two backups alongside production. Two types of storage: not two disks in the same server, but storage that fails for a different reason, such as a local backup appliance and object storage in a data centre. One off-site: in a different location from the rest, so that a fire, a flood or a break-in does not take everything at once.
The idea behind it is independence. Each copy should survive whatever makes the other two unusable. So the question for each copy is not “is it there?” but “what would have to happen to lose this one and the previous one together?” If the answer is the same for two copies (same power circuit, same admin account, same sync), then for the purposes of this rule you have one.
The rule dates from a time when a failed disk or a fire was the main risk. It still holds, but it says nothing about someone who is inside your network with your passwords. That is why it has been extended.
From 3-2-1 to 3-2-1-1-0
The two extra digits cover what the original rule leaves open:
- The extra 1: one copy offline or immutable. A copy that cannot be deleted or encrypted for a fixed retention period, not even by an administrator. Ransomware goes for the backups first, so a copy reachable with the same login as the systems it protects does not count here.
- The 0: zero errors in the last restore test. Not zero errors in the backup job log, but a restore you actually ran and that worked.
How immutable storage works technically, and what you need to configure, is covered in our article on immutable backup. Here the point is the place that copy takes in the whole.
What it looks like with Microsoft 365 and a few servers
Take a typical setup: mail, OneDrive, SharePoint and Teams in Microsoft 365, and in the server cabinet a file server, a server running the accounts package or ERP, and a domain controller. Laptops work through OneDrive. This is how the copies divide up:
- Microsoft 365, copy 1: the data itself, at Microsoft. That is production, not a backup.
- Microsoft 365, copy 2: a backup service that copies mail, OneDrive, SharePoint and Teams at least daily to storage outside your own tenant, with its own admin accounts and a locked retention period. That copy is both your off-site copy and your immutable 1.
- Servers, copy 2: a local backup appliance in another part of the building, not joined to the domain and with its own login. This is your fast restore: a deleted folder or an entire server comes back from here within hours.
- Servers, copy 3: a copy of those backups to object storage in a data centre, with a retention period nobody can shorten. This is the copy for the day the building or the network itself is the problem.
- Laptops: no backup of their own, provided the files really are in OneDrive, Desktop and Documents included. Anything that only lives locally falls outside every copy.
For Microsoft 365 that gives you two copies rather than three. To meet the rule to the letter you add a second destination; for most offices, an immutable copy outside the tenant is the step that makes the difference.
And do not forget the configuration. Firewall and switch settings, the Entra ID setup and the policies in Microsoft 365 are not files on a server, but without them a restore becomes a rebuild. Export them after every change and keep them with the rest.
Why Microsoft 365 is not a backup in itself
Microsoft keeps the service available and makes sure your data survives a failed disk or a data centre outage on their side. Their own shared responsibility model also spells out the other half: the data, and who has access to it, are yours. Whatever you, a colleague or an attacker with your permissions deletes, encrypts or overwrites is faithfully applied everywhere at once.
There is a safety net: recycle bins, version history and the option to roll a OneDrive or site back over a limited period. But that net has a window of weeks to a few months, depending on the component. It is designed for a mistake, not an attack, and it sits in the same tenant as the damage. Whoever holds the admin rights to your tenant also holds the safety net.
Retention policies and archiving are not a backup either. They exist so you can show what was there, not to put a mailbox or a SharePoint site back the way it was last Tuesday. Which components to protect with a separate service is set out on our page on Microsoft 365 backup.
The extra 1: offline or immutable
There are two ways to meet this 1, and they ask different things of your organisation:
- Immutable. Storage that locks each copy for a fixed period. It happens automatically, every day, without anyone having to do anything. The condition: the lock is set in a mode that even the administrator cannot lift, and the account that manages the storage is separate from your domain and has its own multi-factor authentication.
- Offline. A copy that is disconnected: a tape or disk that comes out of the system after writing and is kept elsewhere. Out of reach of any attack, but only as good as the discipline of whoever swaps it. A disk left plugged in “for now” is online.
The NCSC’s guidance on ransomware makes the same point: keep at least one backup that cannot be reached from the network it protects. For most offices, immutable off-site storage is the practical choice, because it does not depend on someone taking a disk home on a Friday afternoon. The test is the same in both cases: could someone with your domain admin password delete this copy or shorten its retention? If so, it is not a 1.
Choose that retention period by asking how far back you need to be able to go. Attackers are often inside for some time before they strike, and by then yesterday’s copy may already be infected.
The 0: a restore test you have actually done
A green notification that the backup succeeded tells you data was written. It does not tell you that you will get it back, nor how long that will take. The only way to know is to do it. A workable schedule:
- Monthly: restore a few random files and a mailbox, including from the immutable copy.
- Quarterly: restore one server in full to an isolated environment and check that the application starts and the data is right.
- Annually: the “office is unreachable” scenario. What do you restore in which order, who does it, and how long does it take from start to finish?
For every test, record what you restored, from which copy, how long it took and what went wrong. Set that time against the recovery time your organisation can live with. If it is longer, that is not a failed test but a decision still to be made: invest in faster recovery, or accept that it takes this long.
The mistakes we see most often
- Mistaking sync for backup. OneDrive, Dropbox, a mirrored disk or replication to a second server pass on every change straight away, including a deletion or an encryption. That is availability, not backup.
- The NAS next to the server. Same rack, same power, joined to the same domain and mapped as a network drive. For the rule that is not a second copy, and ransomware is just as happy to encrypt a mapped network drive as the server.
- One admin account for everything. If the backup system logs in as a domain admin, it falls with the domain.
- Files only. The ERP database is copied half-way through a transaction, or the system state of the domain controller is missing. A copy like that exists, but it will not start.
- Alerts nobody reads. A job that has been failing for three weeks is three weeks without a backup. Make one person responsible for reading the report and picking up a failed job the same day.
- Retention that is too short. A week of history does not help if the infection is a month old.
Where our work stops
What we do. We set up the copies along these lines, manage the backup jobs, read the alerts, carry out the restore tests and report the recovery times. For Microsoft 365 we arrange the backup outside your tenant; for servers, the local appliance and the off-site copy.
What we do not do. We are not lawyers and do not decide how long you are legally required to keep data, or when you must delete it; that comes from your accountant, your legal adviser or your data protection officer, and we set the retention to match. We are not an insurer and do not say whether your setup meets the terms of a policy; the insurer judges that. Microsoft 365 remains a Microsoft service: we do not fix outages in the platform, we make sure your data does not depend on it. And anything outside the agreed scope, such as files that only exist on a laptop, we cannot restore.
Questions we get about this
The five that come up most often.
What is the 3-2-1 backup rule?
A rule of thumb for backups: keep three copies of your data, on two different types of storage, with one of them in another location. The original you work with counts as one of the three. The aim is that no single event, such as a hardware failure, a fire or a mistake, hits every copy at once.
What does 3-2-1-1-0 mean?
It is the 3-2-1 rule with two additions: one copy that is offline or immutable, so that ransomware or an administrator cannot erase it, and zero errors in the last restore test. That zero refers to a restore that was actually carried out, not to a backup job that reported success.
Does Microsoft 365 need a separate backup?
Yes, if you want to be able to recover from an attack, an administrator’s mistake or losing access to your tenant. Microsoft protects the service against failures on its side, but the data and who can access it are your responsibility. The recycle bin and version history have a limited window and sit in the same tenant as the damage.
How often should you test a backup restore?
A practical schedule is a few files and a mailbox every month, a full server in an isolated environment every quarter, and once a year the scenario in which the whole office is unreachable. Record how long each one took, because that time is the figure that matters.
Do you decide how long we should keep backups?
No. How long data must or may be kept by law is a question for your accountant, your legal adviser or your data protection officer. We set the retention period to whatever you decide and explain what that means technically for storage and recovery.
Further reading
Where this rule connects.
Shall we hold your copies up against the rule?
Tell us what is being copied today, where to and with which account. We will tell you which of your copies genuinely count towards the 3-2-1 rule and which do not.
Practical IT knowledge in your inbox
New guides on management, security and the workplace, written by the people doing the work. No sales talk, and you can unsubscribe in one click.