Backup and recovery: a backup only counts if you can prove you can restore from it

How to set up backups that survive ransomware, set recovery targets per system and prove with timed tests that you can restore.

In this topic: IT and data 18

Backup software reports what it copied, not what you can get back. A green tick says that data went somewhere last night, not whether the copy is complete, readable and beyond an attacker's reach, or how long it would take to run the business from it. Backup and recovery only work as a pair, and recovery is the half that is easiest to put off.

RansomwareRansomwareMalicious software that encrypts or steals data so that criminals can demand payment for its release. gangs know that: the UK NCSC notes that attackers have destroyed copied files or disrupted recovery processes (source: UK NCSC, September 2021). A backup only counts once you have restored from it, against the clock, with someone from the business checking the result.

Synchronised folder Office PCCloudLaptop sync Encrypted once, encrypted everywhere
  • A deletion or encryption reaches every copy
  • No way back to an earlier moment
Backup with versions Working files restore MonTueWedThu Back to the last clean version
  • Separate copies of earlier moments
  • Restore the last clean version

Synchronisation keeps every copy identical, and that is exactly the problem: the UK NCSC warns that cloud services often synchronise immediately after files have been replaced with encrypted copies (source: UK NCSC, September 2021).

In short

A backup is a separate copy of an earlier moment; a synchronised folder is not. Keep at least one copy that no admin account in your network can change or delete, set per system how much data you may lose and how long it may be down, and prove with timed restore tests that you meet both. Management sets the limits; IT makes them work.

  • Copies
  • Recovery targets
  • Ransomware
  • Restore tests
01 The 3-2-1 rule

Three copies, two media, one off-site, and one out of reach

The classic rule still holds. It needs one addition now that attackers go after the backups as well.

The 3-2-1 rule asks for three copies of important files, on two different types of storage media, with one copy stored off-site, away from your business location (source: CISA, November 2026). The original counts as one of the three (source: Canadian Centre for Cyber Security, June 2024). The point is that no single event, a fire, a failed disk or a deleted folder, hits every copy at once.

A stolen admin account is a different kind of event: every copy it can reach is one risk. The UK NCSC advises an offline backup kept separate from your network and systems, or in a cloud service designed for this, and warns against relying on several copies in a single cloud service (source: UK NCSC, September 2021). That is the plus one: a copy that is offline or immutableImmutable storageStorage that offers no way to change or delete a backup before an agreed date, not even for an administrator..

Your network and admin accounts Original copy 1 Local backup copy 2, other media attacker ! same login? blocked Off-site copy copy 3, other place Out of reach offline or immutable
Why the plus one matters. A stolen admin account reaches every copy it can log in to. The out-of-reach copy is offline or immutable, under separate accounts.

For an accounting firm with its files in SharePoint, that could mean the original in Microsoft's cloud, a daily immutable backup at a separate provider under its own accounts, and a monthly encrypted copy on a disk kept elsewhere.

02 Recovery targets

RPO and RTO in plain words

Two limits decide what your backups must be able to do. Management sets them per system, before anything is bought.

The recovery point objective (RPO) is the point in time to which data must be recovered after an outage (source: NIST, May 2010). In plain words: how much work you can afford to lose; with one backup a night, up to a full day. The recovery time objective (RTO) is how long a system may stay in recovery before the business is hurt (source: NIST, May 2010): how long you can do without it. It should be shorter than the maximum tolerable downtime, the outage the owner of the process is willing to accept (source: NIST, May 2010).

Last backupIncidentWorking again RPO work you lose RTO time to recover e.g. one hour e.g. half a day
RPO and RTO around an incident. Grey dots are earlier backups: the gap between backups sets the work you can lose. Example values for order processing.

Take a wholesaler. Order processing may lose at most an hour of orders and must run again within half a working day, which calls for hourly backups and a practised restore. The archive can lose a week and wait a fortnight.

Shorter limits cost more, so weigh them against what a day without each system costs in lost sales, overtime and penalties. That is a management decision, not a setting for the IT partner.

03 Scope

What to back up: servers, laptops, cloud apps and settings

A backup is only as complete as its list. Data on laptops and in online services is what most often goes missing.

Microsoft is explicit about its own cloud: for every deployment type you own your data and identities and are responsible for protecting them (source: Microsoft, August 2026). The services do contain safety nets. Deleted items stay in the SharePoint recycle bin for 93 days, and a document library can be rolled back to any moment in the last 30 days, also after ransomware (source: Microsoft, September 2025). Useful, but these nets sit under the same admin accounts as the data.

For a real backup of SharePoint, OneDrive and Exchange Online, Microsoft sells Microsoft 365 Backup separately, billed on consumption; its storage is append-only, so backups cannot be overwritten, and the data stays within Microsoft's boundary (source: Microsoft, November 2026). That keeps every copy with one supplier, though. The UK NCSC advises keeping an independent copy of data from online services in another safe place or service (source: UK NCSC, January 2024).

What to includeCheck each line against your own set-up.
WhatCommon gapWhat to add
Servers and databasesFiles are copied, but not the full system or a consistent databaseImage backups of whole systems and database-aware backups
LaptopsWork saved locally never reaches the serverCentral folders that are backed up, or a backup of local data
Online services (SaaSSaaS (software as a service)An application you use online on subscription, such as an online accounting package, a CRM or a cloud office suite.)Only the provider's recycle bins and version historyA separate backup or regular export under other accounts
ConfigurationsFirewall rules and network and cloud settings exist in one place onlyAn export after every change; CISA lists configuration files and software settings among the data to back up (source: CISA, November 2026)
Recovery documentsThe restore instructions sit on the server that has failedRunbook, contact list and key locations kept offline, on paper too
04 Ransomware

Keeping backups out of an attacker's reach

A working backup makes a ransom demand pointless, which is why attackers go after it. Four measures keep at least one copy beyond their reach.

  • Immutable storage: the backup store offers no way to alter or delete backups, or delays deletion for an agreed period and raises an alert that reliably reaches the owner (source: UK NCSC, November 2024).
  • Separate credentials: deleting backups should require compromising a second, separate system (source: UK NCSC, November 2024). Manage backups from dedicated admin workstations, with the MFAMFA (multi-factor authentication)Signing in with a second step besides the password, such as a code or an approval in an app on a phone. method on a device other than the one used to manage them (source: UK NCSC, September 2021).
  • An offline copy: devices that hold backups are not permanently connected to your network, and before a recovery they are connected only to known clean devices (source: UK NCSC, September 2021).
  • Alerts: significant changes and privileged actions in the backup system trigger an alert (source: UK NCSC, November 2024).
Ask what "immutable" means in the contract

The word is used loosely. Microsoft 365 Backup, for example, protects backups against being overwritten but does not block their deletion (source: Microsoft, November 2026). Ask who can delete a backup, after how long, and who is told when someone tries.

Encryption and keys

Encrypt backups, certainly the copies that leave the building; CISA lists encryption and offline copies among the protections for backups (source: CISA, November 2026). Then protect the key as carefully as the data. The UK NCSC points out that an attacker need not delete the data if they can delete or change the encryption key, and suggests a backup of the key outside the system, such as a master key printed on paper and kept in a safe (source: UK NCSC, November 2024). Make sure someone besides the IT partner knows where it is.

05 Proof

Restore tests: how often, and how far back

A backup that has never been restored is a promise. A timed test turns it into evidence you can show a board, an auditor or an insurer.

CISA advises testing that your team can restore data both fully and partially, in scheduled recovery tests (source: CISA, November 2026). NIST's contingency planning guide separates a tabletop exercise, a discussion that deploys no equipment, from a functional exercise in which staff carry out their roles; for moderate-impact systems the exercise includes recovery from backup media (source: NIST, May 2010). Time every test against the RTO, and stop the clock when a user can work again, not when the software reports success.

A testing rhythmA suggestion to adapt to your own limits. The business owner of each system signs off the result.
TestWhat it provesHow often
Single files and one mailboxEveryday restores work, and people know how to ask for oneMonthly
A complete application or databaseIt comes back consistent, with its links workingQuarterly, rotating systems
The most critical system, from the out-of-reach copyRecovery after an attack fits within the RTOAt least once a year
Tabletop exercise of the runbookPeople know their roles and the order of recoveryYearly and after major changes

How far back you can go

CISA advises being able to roll back data at least seven days (source: CISA, November 2026). Keep backups for a set period rather than a set number of copies, so that a burst of corrupted backups cannot push out the last clean one (source: UK NCSC, November 2024). If an intrusion goes unnoticed for longer than your retention, every copy already contains it. Two outer limits apply: tax, labour and other record-keeping rules can require a fixed retention period, while privacy rules such as the GDPR require personal data to be kept no longer than necessary (source: European Commission, November 2026). Check both where you operate.

06 Continuity and ownership

Backup, disaster recovery and business continuity: who owns what

The three terms are often used as if they mean the same thing. They answer different questions, and each needs an owner.

Backup

Copies of data from earlier moments. It answers one question: what can we get back?

Disaster recovery

Getting systems running again after a major disruption, if needed elsewhere through failoverFailoverSwitching to a standby system, often at another location, when the main system fails.. NIST describes the plan as recovering systems at an alternate facility (source: NIST, May 2010).

Business continuity

Keeping the business going while systems are down: workarounds, priorities and communication. NIST describes it as sustaining business processes during and after a disruption (source: NIST, May 2010).

Management owns the limits: which systems matter most, the RPO and RTO of each, and the risk that remains. IT, in-house or outsourced, designs and runs the backups and reports on them. Each process owner checks restored data and keeps a workaround for the days without the system. Outsourcing moves the work, not the responsibility: ask your IT partner whether the backup covers online services, whether a copy sits outside the partner's own accounts and whether agreed recovery times also apply after an attack. For the first hours of an attack, see ransomware response; more guides are in IT and data.

Quick scan

Could you prove that you can restore?

Eight statements from this guide. Answer them for your organisation and see where to start.

  1. 1We keep three copies of important data, on two types of storage, with one off-site.
  2. 2At least one copy is offline or immutable, out of reach of our normal admin accounts.
  3. 3Management has set, per key system, how much data we may lose and how long it may be down.
  4. 4Our email and files in the cloud have a backup outside the service's own recycle bins.
  5. 5Backup encryption keys are stored separately, and someone besides the IT partner knows where.
  6. 6In the last twelve months we restored a complete system from backup and timed it.
  7. 7Our backups go back far enough to reach a clean copy from before an intrusion.
  8. 8A written plan says how the business keeps working while systems are restored.
Read next

Related guides

Toolkit

Backup register, restore test log and recovery runbook

Below you will find a backup register worksheet with example rows for a wholesaler, a restore test log to complete after every test, and a recovery runbook outline that sets out who does what, and in which order, after a major incident.

Questions

Frequently asked questions

How often should we make backups?
As often as the RPO of each system requires. If a system may lose a day of work, a nightly backup is enough; an order system that may lose no more than an hour needs hourly backups or continuous replication with versions.
Is a mirrored disk or a NAS a backup?
Not on its own. Mirroring, one form of RAID, protects against a failing disk, but a deleted or encrypted file disappears from both disks at once. A NAS, a storage box on the network, can be a backup destination if another copy sits outside the network.
Do we need our own backup of an online CRM or accounting package?
Check what the provider keeps, for how long, and whether you can restore a single record or only everything. Make sure you can export your data in a usable format, and keep regular exports or a separate backup under other accounts.
Who should have access to the backups?
As few people as possible, with accounts used for nothing else and protected with a second sign-in step. Nobody should be able to delete every copy alone, and at least one person outside the IT partner should know how to reach the out-of-reach copy and the encryption key.