The backup software’s dashboard is green, with no failed jobs in the last thirty days. But when the auditor picks a file and says, “Bring this back now”, the team starts looking for where the encryption key is kept for the first time. ISO 27001, TISAX and IATF 16949 auditors no longer settle for “Do you take backups?”; they want evidence that backups are taken according to policy and that restores are tested regularly. A green report doesn’t prove the data can come back in usable form. Below is a step-by-step guide to building backup evidence you can present in an audit. The technical design of restore testing is explained in the article 3-2-1-1-0 backup and restore testing.
The evidence chain: policy → job → test → action
The auditor looks not at a single report but at linked links in a chain. If one link is missing, the chain breaks.
| Ring | Evidence | The auditor’s question |
|---|---|---|
| Policy | Frequency, retention period, location and owner per system | What do you commit to backing up, and how often? |
| Scope | Inventory–backup reconciliation | Are all critical systems backed up? |
| Job monitoring | Daily/weekly job reports, closure of failed jobs | Are failed backups noticed and fixed? |
| Restore test | Dated test records | Could you really bring the data back, and how long did it take? |
| Action | Records of fixes for problems found in testing | What did you learn from the test results? |
The weakest link is usually “Scope”. A new virtual machine added to the virtualization cluster six months ago never shows up in the report if it wasn’t added to the backup job. The report stays green. The system has no backup. That’s why, once a month, compare the hypervisor inventory (vCenter, Hyper-V, Proxmox) with the list of machines in backup jobs and record the difference. Many backup products offer dynamic scope based on tags or folders; if you use it, keep the rule itself as evidence too.
In the “job monitoring” link, the auditor asks what happened to each failed job. If the alert email lands in a distribution list but nobody opens a ticket, there’s monitoring but no control. The ticket number for the failed job, its root cause and the date it ran successfully again should be kept together.
A good restore test record
- Date and the person who tested.
- System and data: E.g. “ERP database, manufacturing company code”.
- Test type: At file, virtual machine, database or application level.
- Backup used: Date and copy (local, off-site, immutable).
- Duration: Start–end; comparison with the RTO target.
- Data currency: Time of the last transaction recovered; comparison with the RPO target.
- Verification: The application owner’s confirmation that “the data is usable” (e.g. a sample order was opened, a report was produced).
- Result and action.
The “Verification” row is the field most often left blank. A virtual machine starting up doesn’t mean the application works. An ERP may have come up on screen with the database service not started or unable to reach the license server. That’s why the test should be closed by the application owner performing a real transaction: opening an order, running a report, a user signing in.
Take care with the “Backup used” field too: if every test is done only from the nearest, local copy, the off-site and immutable copies are never tested. In a ransomware scenario, the copy you’ll restore from probably isn’t the local one.
Test types and what they prove
- File level: Shows the backup can be read and is indexed. Doesn’t prove application consistency.
- Virtual machine level: Shows the operating system boots. Must be done on an isolated network; otherwise it clashes with the production machine of the same name and IP.
- Database level: Tests point-in-time recovery and the integrity of the log chain. This is the real evidence for your RPO claim.
- Application / scenario level: Shows that several servers (database, application, license, identity) come up together. This is the real evidence for your RTO claim.
A problem you may run into in virtual machine tests is the absence of a domain controller on the isolated network. The application server starts, but services don’t because it can’t authenticate. In scenario tests, a copy of a domain controller needs to be brought into the same isolated segment.
How often should you test?
The standards don’t set a specific frequency; it’s justified by risk in the policy. Common practice: every three months for critical systems, once a year for others, plus after major changes (version upgrades, storage changes). Tests should be tied to a calendar and can be planned on a rotating basis so each test covers a different system from the previous one.
The moment you write the calendar into the policy, that frequency becomes a commitment. The auditor looks for a record for every row in the calendar. Choose a realistic frequency and stick to it.
Additional evidence for the ransomware era
Attackers now target backups first. That’s why audits also raise these questions:
- Is at least one backup copy immutable or offline?
- Is the backup system managed with credentials independent of the domain admin accounts?
- Is MFA required to access the backup console?
- Is recovery from backup possible if the identity infrastructure (AD) is lost?
The common 3-2-1-1-0 approach (three copies, two different media, one off-site, one immutable/offline, zero errors on verification) provides a practical framework for these questions.
The second point is one that may be left open. If the backup server is a domain member and the console is accessed with the “Domain Admins” group, an attacker who takes over the domain deletes the backups too. Keeping the backup server outside the domain or using a separate management identity breaks this link. As evidence, keep the console’s user list and a screenshot of the MFA setting.
A note on the fourth point too: if there’s an AD backup but the restore steps aren’t written down, using that backup in a crisis takes hours. Restoring a domain controller requires a separate procedure and a separate test record.
The same evidence across three frameworks
| Framework | Related requirement (summary) |
|---|---|
| ISO/IEC 27001:2022 | Annex A controls on information backup and ICT readiness for business continuity |
| TISAX (VDA ISA) | Backup and recovery, and business continuity controls |
| IATF 16949 | Record retention and effectiveness testing of 6.1.2.3 contingency plans |
Keeping evidence in a control-based evidence library makes the same test record easy to find in all three audits.
Pre-audit checklist
- Does the backup policy include frequency, retention and location per system?
- Is the inventory–backup reconciliation for the last three months recorded?
- Is there a closure record for each failed job?
- Does every restore test in the calendar have a dated record and application owner approval?
- Have measured times been compared with RTO/RPO targets, and have actions been opened for deviations?
- Has the immutable or offline copy been used in a test at least once?
Frequently asked questions
Should Microsoft 365 data be backed up?
Microsoft’s retention and recycle bin features may not cover every scenario. The decision should be made by assessing legal hold, accidental deletion and ransomware scenarios.
We don’t have a test environment; how do we test restores?
Options include a temporary virtual machine in an isolated network segment, restoring files to an alternative location, or a controlled test in a maintenance window.
Does a failed test look bad in the audit?
No; recording a failed test, finding its root cause and fixing it shows the control is working. What looks bad is never testing at all.
To assess your backup policy, test program and evidence, take a look at our Backup and Restore Audit service. For setting up and operating backup infrastructure, you can work with our Eretek System Support team.