A practical business guide

How to hand over and remove integration access

Plan the cutover, verify the replacement, remove old provider access, and account for work already in flight.

Download the worksheet ↧

Decide what must keep running

For each connection, decide whether the workflow stays, moves to a new operator or stops. If it stays, prove that the new owner can operate it and that one controlled request completes before removing the old access. If it stops, hold new work, account for jobs that already started and then revoke access at the provider. Record how you verified the result.

Start with the workflows that send messages, move records or run on a schedule. A person leaving the project may own a login, a scheduled job, a stored connection or a recovery mailbox. Those dependencies can survive a change to the project member list. Use the limited-access guide to choose the permissions for a replacement. Use the operator handoff guide for the broader runbook.

Trace each workflow to its owners and credentials

Open the workflow runner and the account that grants access. Record the trigger, destinations, owner, backup operator and the name of each connection. Record references to secrets in the approved vault. Keep the secret values out of the worksheet and evidence. Check for other workflows that use the same identity or connection before you change it.

What to inspect before a handoff
DependencyRecord and test
Owner and recoveryWho controls the runner, provider account, alerts and recovery route? Have the incoming operator open the approved settings and acknowledge a test alert.
Triggers and queued workList schedules, incoming events, retry jobs and pending approvals. Confirm how to hold new work and where to reconcile items already in flight.
Access and stored dataIdentify which provider grants access, which workflows share it and where exported records and logs live. List who owns each retention or deletion decision.

Check the provider control for the exact connection

Google Account Help describes removing a third-party connection and explains that data already shared may still be held by that app. Google's OAuth documentation describes grant-wide effects for clients in the same project and warns that revocation can take time to take full effect. Check the scope and shared dependencies before acting.

For GitHub Apps, GitHub documents installed app access separately from apps authorized to act for a user. Record whether the connection depends on an installation, a user authorization or both. Removing a connector from your workflow editor does not prove that either provider permission was removed.

Switch the runner while work is held

  1. Agree on an operator, a change window and the fallback. Decide which work can wait and who handles work that cannot wait.
  2. Hold new work in one known place. Record pending items, completed effects and requests awaiting retry. Resolve ambiguous results at the destination before replaying.
  3. For a workflow that stays, connect the approved replacement identity. With the old trigger still off, run a harmless request through the new path. Confirm the expected result at the destination and the alert at the incoming operator.
  4. Release one trigger and reconcile the held work with its original request identifiers. If the test fails, keep the work held or use the agreed manual route. Escalate before resuming an old connection.
  5. When the owner accepts the handoff, revoke the obsolete provider access and remove unused stored connections or secret copies under the applicable retention policy. For a workflow that stops, leave its trigger off and confirm that pending work is closed or transferred.

If access has to stop immediately for a suspected compromise or an urgent policy requirement, follow the incident or access-removal process first. Hold automated work or use the agreed manual route. The planned sequence above assumes an approved handoff window.

Fictional example: a workshop parts request

This example is fictional. Blue Mesa Workshop is moving a request queue from a contractor's connector to a workshop-owned connection. A manual intake sheet holds requests during the switch. The destination creates internal draft tasks. The old connection can also serve another approved workflow, so the owner checks that dependency and transfers it before revocation.

What the incoming operator should see
Held requestDestination evidenceExpected decision
SYN-41One draft exists with SYN-41. The old runner shows a timeout.Record the existing result and do not replay it.
SYN-42Confirmed no matching draft. The request passed validation.Release once through the new runner with SYN-42.
SYN-43Destination cannot be checked. The old runner outcome is unknown.Hold and assign an owner. Inspect before replaying.

At reopening, the operator should see two confirmed draft results after SYN-42 succeeds and one held request. Confirm a test with the new identity succeeds. After revocation takes effect, use a harmless, approved access check for the old identity. An explicit access refusal is useful evidence. A connection timeout leaves the old access state unknown. The duplicate-action guide explains why a timeout can hide a completed result.

Close only after the new and old paths are checked

Keep the provider change record and a redacted result from each verification. Check the next expected scheduled run or confirm that the stopped workflow remains inactive through its next scheduled interval. Make sure the backup operator can find the runbook, alerts and safe credential references. If any item is unresolved, put it on the handoff record with an owner and a date.

INTEGRATION HANDOFF AND REMOVAL CHECKLIST
Workflow, owner, backup and change window:
Keep, move or retire:
Trigger, destinations and recovery route:
Provider grant / installation / credential reference:
Other workflows sharing that access:
Held work and destination reconciliation:
Replacement test, expected and observed result:
Cutover approval and trigger state:
Old provider access removed by / when:
Harmless old-access check / result / recheck time:
Next scheduled result or expected stopped interval:
Exports and logs / retention owner:
Unresolved items / fallback / owner and next date:

If a workflow needs a new operator or an agreed stop procedure, use the support and handoff page or book a workflow fit call. Bring a redacted dependency list and one sample request.

Put the plan to work

Start with one useful change.

Bring the process, the people involved, and a redacted example. A free 30-minute fit call can establish the next step.