Our Thinking / Field Notes / The Guide

STEP-BY-STEP GUIDANCE / BACKUP

How to Write a Disaster Recovery Runbook for a Small Business

What a small-business disaster recovery runbook should record: what matters most, recovery order, who does what, where recovery copies sit, and how to test it, grounded in current NCSC guidance.

Start here

For each service your business cannot work without, write down how much downtime and lost work you can tolerate, what it depends on, where its recovery copies live, who restores it, the exact steps in order and how you will know it worked. Add a stop-and-check point so nobody restores over an attack, keep a copy you can open when everything is down, then prove the whole thing by restoring your most important service to a separate location and timing it.

What a runbook is, and what to have ready

A disaster recovery runbook is the document someone opens when systems are down and nobody is thinking clearly. Its job is to let a capable person who did not build your setup get the important things working again, in the right order, without guessing. That is narrower than a full business continuity plan and different from a cyber incident response plan, although the three should point at each other.

Writing the runbook changes nothing on your systems. The only part that touches real infrastructure is the restore test near the end, and that is designed to run somewhere separate. For a business of a few dozen people, aim for a short record per essential service rather than a binder nobody will open.

  • Someone with the authority to say what matters most. Recovery order is a business decision, not only an IT one, so the owner or a director needs to be in the room for the first section.
  • Someone who knows how the systems are really set up: your in-house IT lead or your IT provider. If a provider runs your backups, ask them for the restore steps and retention settings rather than assuming.
  • A list of everything you depend on, including cloud services, your domain name and internet line, not only the server in the comms cupboard.
  • A place to keep the finished document that does not depend on your own network. That is covered further down.

Decide what the business cannot do without

Start with the business, not the technology. The NCSC is clear that business needs, rather than IT considerations alone, should drive recovery priority, and that the systems those functions depend on need recovering first. This section gives you the order and the targets that everything else hangs from.

  1. List what the business does that earns money or keeps it legal: taking orders, invoicing, paying staff, answering customers, delivering, keeping required records. Plain business language is fine at this stage.
  2. Under each activity, name every system it leans on. Include the unglamorous ones: staff sign-in, the internet connection, your domain name and DNS, email, the phone system and any supplier-hosted software. When one of these fails, several services fail with it.
  3. Ask the owner two questions about each service. How long could we cope without it? That is the recovery time objective (RTO). How much recent work could we afford to redo? That is the recovery point objective (RPO). Answers such as 'half a working day' and 'back to close of the previous working day' are perfectly good.
  4. Compare those answers with what your backups can actually deliver. If the owner wants no more than an hour of lost work but backups run nightly, the honest RPO is up to a day. Record the gap rather than the wish. A runbook that quietly promises what the backups cannot deliver is worse than no runbook, because people plan around it.
  5. Put the services in recovery order, dependencies first. Sign-in and networking usually come before file storage, and file storage before the applications that read from it. If a database or data store underpins an application, it comes back before that application.
  6. Beside each service, note a manual workaround for the gap, such as paper job sheets or a spreadsheet kept on a laptop, and who may authorise using it. NCSC guidance recommends identifying workarounds in advance, with the caveat that they are a short-term measure.

Give each service a one-page recovery record

Now write one record per service, starting with the highest priority. This is the reference half of the runbook: facts that stay true between disasters, so a stranger knows what they are dealing with before they touch anything. The step-by-step half comes next.

Here is how a record might look for an imaginary ten-person print firm. The names and figures are invented, and the last line is deliberately blank, because a record only earns a test date after a real test.

  • Service name, business owner and technical owner, each with a named deputy.
  • The agreed RTO and RPO, plus any gap between those and what the backups deliver.
  • Dependencies: what must already be working before this service can be restored.
  • Recovery copies: what backs it up, how often, where each copy lives, how long copies are kept and which copy is kept disconnected or otherwise out of reach of your normal network.
  • Access needed to restore: which account, and where its credentials or recovery codes are held and who can retrieve them. Record the location, not the secret.
  • Where to restore to: spare hardware, a new virtual machine or a fresh cloud location, so the recovery does not depend on the damaged system.
  • Anything needed to rebuild: licence keys, certificates, supplier support numbers and contract references.
  • Last restore test: date, who ran it, how long it took and what was found.
text
SERVICE: Shared job files and artwork (example only)
Business owner: Operations Director (deputy: Office Manager)
Technical owner: IT support provider's named engineer (deputy: second engineer on the support contract)
Target time to recover (RTO): back in use within 1 working day
Target data loss (RPO): up to 1 working day of changes
Gap: none against nightly backups. Owner would prefer 4 hours, which is not currently achievable.
Depends on: internet line, staff sign-in service, file server (virtual machine)
Recovery copies: nightly backup to cloud storage (30 days kept); weekly copy on a disconnected disk in the office safe
Access to restore: backup console administrator account; credentials held in the company safe; retrievable by the Operations Director or Office Manager
Restore to: spare host in the comms room, never over the live server
Needed to rebuild: backup software licence key; supplier support number
Proof it worked: Finance opens last Friday's invoice batch; Production opens a named test artwork file; newest file dates match the RPO
Recovery steps: see the recovery procedure for this service
LAST RESTORE TEST: (date, tester, time taken, result)

Write the recovery steps for someone who did not build it

This is the page that gets used under pressure, so write it for a competent colleague who has never done this restore. Use the real labels from your own backup product's restore instructions as you go. They differ between products and versions, and a step that says 'start the restore' without saying where is not much help at the worst possible moment.

  1. Open with the starting conditions: what has to be true before anyone restores anything. Include the stop-and-check point from the next section, so nobody starts rebuilding a system that is still under attack.
  2. List the access needed and where it lives, by location rather than by value. Name who can retrieve it and what to do if that person is unreachable.
  3. Write the restore steps in the order they happen, one action per line. For each, say where it is done (which console, which machine), what it changes and where the restored data goes. State plainly that the restore target must not overwrite live data unless overwriting is the deliberate intent.
  4. After each major stage add a check the person can see for themselves: a file count, the newest file date, a named person opening a named document, a service answering. A step without a check is a hope.
  5. Record what happens afterwards: reconnecting users, any sharing or DNS changes, how to return to the normal setup if you recovered somewhere temporary, and how to merge anything done by hand in the meantime into the restored data.
  6. Leave a blank beside each stage for the time it takes. You will fill these in from the test, not from optimism.

Say who does what and how to reach them

Recovery stalls surprisingly often on questions of who is allowed to decide, rather than on technology. Settle that on a quiet day. NCSC guidance suggests giving critical tasks joint or shared responsibility so there is cover when the usual person is away, and documenting the points at which a decision passes from one person to another. Check the contact details every couple of months, or they go stale faster than you would expect.

  • Roles: someone who decides to invoke the runbook, someone who does the restoring, and someone who talks to staff and customers. Name a deputy for each.
  • Trigger points: when the decision passes between people and when the owner or a director must be told.
  • Contacts: your IT provider, internet and phone suppliers, key software suppliers and your cyber insurer, with the policy number and any incident line. Note what each contract covers and when to call them. If you have an incident response company lined up, include it.
  • Out-of-band communication, meaning a route that does not rely on your own systems: staff mobile numbers held offline, plus a messaging group set up in advance with a service you do not run yourselves. If email and chat are down, or cannot be trusted, this is how you reach people.
  • Obligations: where an incident might need reporting to a regulator, insurer or customer, record who decides and where the rules sit. This runbook is not the place for legal analysis, and you should take proper advice on what applies to you.

Add a stop-and-check point for cyber attacks

Not every disaster is a flood or a failed disk. If the cause might be ransomware or another attack, restoring straight away can put the problem back. The NCSC says that where possible investigation should happen before systems are restored, and that rushing recovery before the incident is understood can significantly increase risk. It also treats the state of your backups as something to check, not assume: when the last ones were taken, whether they are available and whether they might be compromised.

So the runbook needs a gate near the top. Something like: if the cause could be malicious, do not restore until the named person or your incident response provider confirms it is safe to start, which backup copy to use and which clean location to restore into. Your restored systems should be checked before you reconnect them to the main network.

This is also where the recovery copies in each record matter. NCSC guidance suggests keeping at least one backup disconnected so a single incident cannot reach every copy, making backups immutable (unable to be changed or deleted once written) where possible, and covering configuration and identity systems as well as data. The runbook does not need to design that. It needs to say which copy is the disconnected one, where it is and who can get it.

Keep a copy you can open when everything is down

The classic failure is a recovery document stored on the very file server it describes. NCSC guidance says recovery documentation should be held securely and separately from production systems, and Microsoft's disaster recovery guidance adds that offline or printed copies are worth having for worst-case scenarios.

  1. Keep the master copy somewhere that does not rely on the systems it describes, rather than only on your file server or the email account it tells you how to restore.
  2. Print the current version and keep copies where the named people can reach them, such as the office safe and a deputy's home. Keep an offline digital copy on removable media that stays disconnected.
  3. Leave passwords out of the document. Record where they are held and who can retrieve them. Printed runbooks have a habit of ending up on desks.
  4. Put an owner, a version number and a date at the top, and replace the old printed copies whenever you issue a new version.
  5. Check it: ask someone other than the author to find and open a copy with the office network and email unavailable. If they struggle, fix the storage arrangement before anything else.

Test the runbook and check it worked

The NCSC points out that many plans fail in practice, and recommends testing regularly. Two kinds of test earn their keep. A tabletop exercise is a talk-through of a scenario with the people named in the runbook, which tests decisions and roles. A technical restore test proves the steps and the timings. The NCSC's free Exercise in a Box tool can supply scenarios for the first kind.

Use the second test to prove the highest-priority service first. The restore can use real business data, so run it only somewhere access is restricted, and delete it afterwards.

  1. Run a tabletop exercise. Gather the named people and walk through a plausible morning, for example email and file storage are unavailable at nine o'clock on a Monday. Follow the runbook as written and note every place someone says 'well, I'd probably...'.
  2. Choose a separate restore target for the technical test: a spare machine, a test virtual machine or another folder or site that is not live. Restoring to a separate location means nothing live changes, so there is nothing to roll back. If you are not certain where the restore will land, stop and check before starting.
  3. Ask someone who did not write the steps to follow them, with the author watching but staying quiet. Record every question they ask. Each one marks a gap.
  4. Time each stage and the whole restore. How long it takes depends on data volume, hardware and connection speed, and large restores can run for hours, so let it finish and do not estimate. That measured figure is the most useful number in the document. Compare it with the RTO, and compare the age of the newest restored data with the RPO.
  5. Run the proof checks from the recovery record: the named person opens the named document, the service responds, the file dates make sense. Success means a person who did not build the system restored your top service to a separate location using only the runbook, inside the target time, with the checks passing.
  6. Update the runbook, then fill in the last-tested line with the date, who ran it, the time taken and what you found. Delete the test restore.
  7. Set the next review. Check the contact details every couple of months, and review the full runbook at least twice a year and after any major change to your systems, suppliers or staff roles.

If the first test goes badly

That is the point of testing, and a bad first run is the most common result. The useful question is what to do with the findings.

  • The restore takes far longer than the RTO: this is a business decision, not an embarrassment. Either the owner accepts the longer time, or the backup approach needs to change. Record the real figure and do not quietly rewrite the target to match.
  • Steps depended on someone's memory: write them down, then have a different person repeat the test.
  • The restore worked but the application would not start: you missed a dependency, often sign-in, a licence, a certificate or a DNS entry. Add it to the record and to the recovery order.
  • Nobody could get the credentials needed: fix the access arrangement now, with a named deputy who can retrieve them, rather than during an outage.
  • The runbook itself could not be found: revisit where it is stored and who holds copies.
Technical terms explained
immutable
Stored so that the data cannot be altered or deleted for a set period once written, which protects backups from tampering or ransomware.
out-of-band
Communicating or accessing something by a route that does not depend on the systems that have failed, such as personal phones when company email is down.
RPO
Recovery point objective: the most recent work, measured in time, that the business can afford to lose and redo after a recovery.
RTO
Recovery time objective: the longest the business can accept a service being unavailable before the harm becomes unacceptable.
runbook
A written set of instructions that lets someone carry out a task, such as recovering a system, in the right order without relying on memory or the original builder.
tabletop exercise
A discussion-based rehearsal where the people named in a plan talk through a scenario to test roles and decisions without touching any systems.

Sources and further reading

KEEP THE CONVERSATION GOING

Want a second pair of eyes on your recovery plan?

If you would like help working out what to recover first, or a restore test planned against your real systems, tell us what you rely on and what hurts most when it stops.

Talk to us