What it does

Dataverse backs up every environment that has a database, continuously, using Azure SQL automated backups. A "manual backup" is a labelled restore point rather than a separate copy of anything. Copy environment is the related operation that pushes a source environment's data and customisations over the top of a target.

Key facts

• System and manual backups are retained seven days by default. Production Managed Environments can extend to 14, 21 or 28 days through the admin centre or Set-AdminPowerAppEnvironmentBackupRetentionPeriod. Everything else stays at seven regardless of what the setting says.
• Retention changes apply to future and existing in-window backups, take up to 24 hours to land, and can remove older backups sooner than you'd expect.
• You can't restore directly onto a production environment. Switch it to sandbox, restore, switch it back — and while it's a sandbox you can only pick a restore point from the last seven days.
• Source can be production, sandbox or developer. Target can only be sandbox or developer, and a developer target requires a developer source. Teams environments only self-restore.
• Source and target must be in the same region. Managed Environments only restore to Managed Environments. If the source has CMK applied or Virtual Network support, the target needs the same enterprise policy.
• Restoring needs 1 GB of free capacity. Backups themselves don't count against capacity.
• Audit logs are excluded from both restore and copy by default because they add substantially to the runtime. Copy and restore usually finish within 8 hours; with a lot of data, including audit data, they can take 24.
• Copy comes in two flavours: Everything (data, users, customisations, schema) and Customizations and schemas only, which truncates a long documented list of tables including Account, Contact, Incident, Annotation, Attachment, Audit and PrincipalObjectAccess.
• Only up to the 20 most recent revisions of each canvas app come across in a copy.
• You can't download a backup. There's no offline copy, and moving data out means a migration exercise.
• After copy or restore the target lands in administration mode with background operations disabled.

When to use / skip

Take a manual backup before every solution import into production that you'd struggle to reverse — it costs nothing and it's a labelled point you can name. Beyond that, treat the seven-day window as the real constraint: this is a rollback mechanism, not an archive, and any client who says "backup" while meaning "we keep everything for seven years" needs redirecting to long-term retention and their own export pipeline. Use Everything copy for UAT and upgrade testing where the data matters, and Customizations and schemas only where handing production data to a development team is the thing you're trying to avoid.

Configuration decisions

• Which production environments become Managed Environments purely to get retention past seven days, and whether that's set at environment group level (where it can't be overridden individually) or per environment.
• Whether copy operations include audit logs, trading a much longer runtime against a complete history in the target.
• Everything versus customisations-and-schemas per refresh, driven by who has access to the target and whether the data is personal.
• What the post-copy checklist covers — mailboxes, SharePoint sites, plug-ins, connection references, custom connectors, HTTP-triggered flows — and who owns running it.
• Whether the source environment survives the copy or gets deleted afterwards.

Gotchas

• Anything not in a solution — canvas apps, flows, custom connectors, connections — might not come across in a copy. Validate apps afterwards rather than assuming.
• Solution flows in the target are deleted and replaced, arrive turned off, and need connection references reconnected by the right user. Non-solution flows in the target survive untouched, which is usually not what anyone wanted.
• HTTP-request-triggered flows get a new URL in the target. Anything calling them breaks silently.
• Canvas app IDs change after a restore, and apps shared with Everyone come back not shared with Everyone. Sharing with a security group survives.
• A failed copy or restore leaves the target disabled. Wait 30 minutes and retry with the same environment as target — that's the only way back other than reset or delete.
• Restoring an environment onto itself doesn't delete audit logs generated after the restore point, so the audit trail runs ahead of the data.

Consultant notes

• Say the words "seven days" out loud in the DR conversation. Clients routinely assume the platform is their archive, and the gap between that assumption and reality is the single biggest surprise in Dataverse governance.
• The mailbox point is worth its own line in the runbook: a mailbox can't sync with two environments, so an Everything copy that keeps server-side sync on will start sending real email from your test environment.
• Budget a full working day for the post-copy tidy-up on any environment with integrations. It's never the copy that takes the time.
• If a client needs to recover from a bulk delete without losing everything since, restore into a new empty environment and lift the rows across. Don't restore over the live org.

---

Worth checking again if backup retention options change, or if restore-to-production ever becomes possible without the sandbox switch.