Regain control before the engagement ends
Changing IT providers is more than collecting passwords. You need to administer services, renew licences and restore data without depending on the outgoing provider. At fictional Boreal Workshop, the backup portal and DNS are accessible, but recovery contacts still point to the provider. Resolve those dependencies before revoking access. If you suspect unauthorized access or the departure is disputed, work with incident response: urgent revocation may take precedence over the normal sequence.
1. Give each item an owner and an acceptance test
Keep a register with service, internal owner, provider, tenant or account ID, renewal date, dependencies, transfer method, accountable person and acceptance evidence. Separate ownership, administration, billing and data access: having an admin account does not prove you own the subscription. Have contract owners confirm export rights, formats, timelines, costs, retention and transition support. A service shared among customers may require export or migration rather than a whole-portal transfer.
2. Recover the accounts that recover everything else
Prioritize the registrar, DNS, recovery contacts, Microsoft 365/Entra tenant, hosting, backup console, firewall, virtualization and secret vaults. Obtain service identifiers, roles, MFA procedures and renewal dependencies without copying secrets into the register. Create named accounts and test emergency access under internal control. Move required secrets through an approved vault; prefer issuing new credentials. For the domain, confirm registrant, DNS zone access and renewal. A registrar transfer and a nameserver change are distinct operations; check the registry rules for the domain.
3. Microsoft 365: separate partner access from licensing
Review partner relationships in the Microsoft 365 admin centre, including GDAP and any legacy delegation still present. Removing GDAP roles removes access through that relationship; it neither transfers subscriptions nor removes guest accounts, direct roles or third-party apps. Have the incoming and outgoing partners separately confirm the commercial process, transferable products and deadlines. Verify prerequisites for the features in use, such as eligible role management in PIM and log retention, against tenant licensing.
4. Inventory without making changes
In an already authenticated Microsoft Graph PowerShell session, confirm the tenant first. Use Microsoft.Graph.Identity.Governance and delegated RoleManagement.Read.Directory consent with an Entra role supported for that read. The output shows returned directory role assignments, not every indirect membership or eligible role. Check groups, PIM, partner delegation and Azure RBAC separately. DNS queries record current public responses for the fictional domain; they neither export the zone nor prove ownership. Keep output in restricted internal storage.
# Tenant et portée / Tenant and scope
Get-MgContext | Select-Object TenantId,Account,Scopes
Get-MgRoleManagementDirectoryRoleAssignment -All |
Select-Object PrincipalId,RoleDefinitionId,DirectoryScopeId
# Remplacer par un domaine autorisé / Replace with an authorized domain
Resolve-DnsName atelier-boreal.example -Type NS -DnsOnly
Resolve-DnsName atelier-boreal.example -Type MX -DnsOnly5. Ask for documentation someone can operate from
Request network and VLAN diagrams, exportable device configurations, asset inventory with versions and support dates, runbooks, service account dependencies, certificates and expiry dates, alerts and relevant ticket history. Include code repositories, pipelines, RMM/EDR agents, automation, OAuth integrations and billing accounts. Record each export’s format, compatible version and restore procedure. Have the incoming provider follow one runbook without coaching; an unused PDF does not validate handover.
6. Recover backups you can restore
Identify covered workloads, exclusions, retention, copy locations, decryption keys, licensing and prerequisites for importing a repository or restoring outside the provider’s portal. Arrange an internally controlled copy where the contract and product allow it. Run an isolated pilot: open a reference file and, for a critical service, boot a VM and verify the application with the business owner. Record the point used, elapsed time and measured data loss. An archive you cannot decrypt or a green backup job does not meet the exit criterion.
7. Cut over, then revoke in batches
Changes start here: agree on a window, approver and stop criteria. Pilot a noncritical service first and confirm internal and incoming accounts can administer, monitor and recover it. For integrations, issue a replacement identity or key, update consumers, validate and then retire the old one. After acceptance, remove GDAP delegation, roles and VPN/RMM access, disable outgoing accounts per policy and revoke sessions where needed. Separately check app tokens, API keys, certificates, local accounts and service-native sessions. Some issued access tokens may persist until expiry and applications may control their own sessions. Prove denial through logs and authorized tests without reusing provider credentials.
8. Plan rollback without surrendering control
Before cutover, keep reference configurations, tested internal emergency access and an owner who can restore routing or configuration. If monitoring or backup fails, pause the next batch, restore the last safe configuration and retry with a new time-limited identity if necessary. Do not re-enable compromised accounts or old secrets as rollback. Obtain business and technical acceptance, monitor errors and residual access, then confirm retention or deletion of copies under applicable commitments. An infrastructure and backup assessment can complement this handover checklist.