Backups: what must you back up, and would restores work?

This quarter, agree with each system's owner how much data it may lose and how long it may be down, and check that your backups cover every system, SaaS service and configuration, with one copy outside each provider's account. Then restore something every month, from single files to whole systems and SaaS apps, and treat any target you haven't met in a test as unmet.

Reviewed
Next review
Version
1.0
Written for
IT teams in small, mid-size and large organizations with on-premises servers, virtualization, Microsoft 365 or Google Workspace, other SaaS and cloud workloads

If you only do five things

  1. Let each system's owner set its targets.

    Ask how much data it may lose (RPO) and how long it may be down (RTO). The business decides; IT makes it happen.

  2. Back up what the provider won't restore.

    Your data and configuration stay yours to protect in SaaS too3. Keep an independent copy in another service4.

  3. Keep a copy outside the platform it protects.

    In 2024 a Google error deleted a customer's VMware Engine private cloud without notice. Backups kept in Cloud Storage, outside the deleted private cloud, helped restore it5.

  4. Check coverage weekly, not just job status.

    Alert on any system without a recent restore point. New servers, sites and SaaS apps are the usual gaps.

  5. Restore something every month.

    Rotate a file, a mailbox, a server and a SaaS app, and have the owner check each. The ransomware sheet covers the full test.

Set scope and targets with the business

  1. What do you run?

    Take the asset inventory and add SaaS from invoices and single sign-on. Include network devices: they store configuration too2.

  2. How much loss and downtime can each system take?

    Agree RPO and RTO with the owner, within the business continuity sheet's process limits, plus the legal and deletion needs the BSI lists2. Base backup frequency on criticality and change rate1.

  3. How far back must you go?

    Set retention from the slowest loss anyone would notice.

  4. Can the method meet the target?

    Check that restore bandwidth fits the RTO, and that online backup contracts state location, SLA and encryption2.

  5. Can you reach the plan in a crisis?

    Write one page per system: what, how, how long, restore order, owner. Keep it where it survives the backup system2.

What people forget to back up

  • SaaS beyond mail and files. CRM, ticketing, wikis, HR, finance. Atlassian's backup add-on keeps copies only 30 days6.
  • Repositories and CI. GitHub's migration archives skip LFS objects, discussions and packages, and GitHub has no supported way to restore them7. Self-managed GitLab backups omit gitlab-secrets.json, which holds the key to 2FA data and CI/CD variables8.
  • Tenant configuration. Microsoft Graph can snapshot Microsoft 365 tenant settings, but deletes each snapshot after 7 days9. Export them.
  • BitLocker keys. Deleting an Entra device deletes them for good10.
  • Network configs and the admin vault. Export DNS zones and device configs after each change, and keep an encrypted vault export offline.

What SaaS providers restore, and for how long

  • Exchange Online. Deleted items are recoverable 14 days by default, 30 at most11. A deleted mailbox without a hold is gone after 30 days12.
  • SharePoint and OneDrive. The recycle bin keeps items 93 days. After that, Microsoft support can roll back only a whole site or subsite, within 14 more days13. A deleted user's OneDrive is kept 30 days by default, then 93 days in the recycle bin14.
  • Microsoft 365 Backup. It covers SharePoint, OneDrive and user and shared mailboxes, not group or room mailboxes16. Its copies stay inside Microsoft 36515.
  • Google Workspace. Admins can restore Drive files for 25 days after the trash is emptied17 and deleted users for 20 days18. Google says Vault isn't designed as a backup tool19.

How far back to keep

  • Keep dailies 35 days and monthlies 12 months. That spans a month-end close, and the 35 days match AWS Backup's default audit control20.
  • Know what versioning keeps. With an expiry limit, SharePoint files nobody edits can lose every version21. Raise Exchange's deleted item retention to 30 days11.
  • Archive records; don't keep backups for years. The BSI treats long-term retention as archiving2. Meet tax and legal retention in a records system.
  • Let personal data age out. A fixed retention period does that. The retention and deletion sheet covers erasure requests and restores.

Prove coverage and restores

  • Find what nothing protects. AWS Backup Audit Manager flags resources outside any backup plan or without a recent recovery point20.
  • Alert on silence. A job that never starts raises no failure, and full storage can stop backups or overwrite old ones unnoticed2.
  • Test small and large restores. ENISA's NIS2 guidance suggests testing full system and single file restores, validated by business users22. In tests, restore SaaS to a new location, not over production.
  • Automate in the cloud. AWS Backup can restore a random recovery point on a schedule, time it and delete it23.

What changed in 2026

  • Microsoft 365 Backup restores single files and keeps longer. Granular file restore has been generally available since September 2026, for the last 14 days. Recovery windows have been configurable since September, and since 31 October reach 10 years for SharePoint and OneDrive16.
  • Microsoft 365 Backup catches new items. Full Workload Backup, in preview since September 2026, adds eligible sites, OneDrives and mailboxes that no policy covers16.
  • AWS air-gapped vaults take more workloads. Since March 2026 they accept Amazon EKS backups, which can be shared for recovery across accounts and AWS Organizations25.

What you can skip

  • Image backups of laptops. If files sync to OneDrive or Google Drive and Intune or Autopilot rebuilds devices, reinstall instead.
  • Whole-VM backups of servers built from code. Back up the data, the code and the secrets.
  • One schedule for everything. NIST bases frequency on criticality and change rate1. Most file shares are fine with daily backups.
  • Decade-long backup retention. It multiplies cost and erasure work. Archive what the law requires.

Mistakes that cost the most

  • Deleting a leaver's account first. The mailbox is gone 30 days later12, a Google account after 20 days18.
  • Counting RAID or replication as a backup. A mirror changes with the original, so overwrites and malware hit both2.
  • Keeping cloud backups in one region. Azure's Cross Region Restore needs a geo-redundant vault, set before the first backup24.
  • Trusting a green dashboard. Jobs can succeed while skipping new systems. Only a restore proves a backup.

What this looks like at your size

Small team, no security staff

  • Raise Exchange deleted item retention to 30 days, and move a leaver's OneDrive files before deleting the account.
  • Keep a printed page per system: owner, what's backed up, where, how to restore.
  • Restore one file and one mailbox item every month and write down how long it took.

Growing, with IT or security staff

  • Spend an hour with each system owner on RPO, RTO and retention, and get their sign-off.
  • Send a weekly report of assets without a recent restore point to a named person.
  • Back up repositories, CI settings and tenant configuration, not just data.

Enterprise or regulated

  • Cloud, data center, managed service and other digital providers under NIS2 need backup plans covering configuration and cloud data, distant copies and retention, plus integrity checks22.
  • Report measured restore times against RPO and RTO per system each quarter.
  • Copy critical SaaS and cloud data to a second provider and restore from it yearly.