A revoked connection is an authorization incident with customer work attached. Reconnecting the account fixes access; it does not tell you which appointments, notes, or messages were completed before the connection failed.
For an agency running voice workflows, recovery should begin by stopping new remote writes for the affected location. Keep taking messages only if you can explain that booking is awaiting confirmation. Preserve the caller's request in a separate durable intake queue with a timestamp and an incident identifier.
Separate revoked access from temporary failure
Record the failing operation, location identifier, response category, and last successful action. Never put credentials into the incident log. A timeout should follow a different recovery path from an explicit access failure: the timeout may hide a successful write, while revoked access requires an authorized account owner to reconnect.
HighLevel documents both private integration tokens and OAuth authorization. Identify which connection the client actually uses before changing credentials. A token intended for another location is not a recovery strategy. See the official authorization overview.
Reconnect without losing the action ledger
Keep old action records after reconnecting. Each intended operation needs a local identifier, destination location, desired effect, and status such as pending, confirmed, uncertain, or abandoned. Save the remote record identifier whenever available.
In an illustrative incident, three calls arrive during a ten-minute outage. One requested appointment failed explicitly, one timed out after submission, and one was captured after writes were suspended. The first and third can enter a controlled retry queue. The second requires a remote lookup or staff investigation before another appointment is created.
| Recovery test | Expected evidence |
|---|---|
| Reconnect the intended client | Location identity matches the incident |
| Read one known record | Authorized access is restored |
| Submit one approved test action | Remote identifier is recorded |
| Replay an uncertain action | Existing result is reconciled first |
| Leave another client disconnected | No shared credential fallback occurs |
Resume with an owner and a cutoff
Have a person approve the outstanding queue, especially requests whose original time has passed. A customer who wanted a morning visit should not silently receive a booking tomorrow. Track the age of the oldest pending request and assign each unresolved item to an operator.
The tradeoff is slower recovery than blindly replaying everything. The benefit is an explainable outcome for every caller. Retaining full transcripts indefinitely is unnecessary for this purpose; retain the minimum action evidence needed under the client's policy.
Questions about connection recovery
Should the agent keep promising appointments?
Only if the scheduling system confirms them. During suspension, explain that the office must confirm the request and give the agreed follow-up route.
Can all failed requests be replayed automatically?
Explicitly rejected writes may be eligible. Requests with unknown outcomes need reconciliation, and expired customer preferences need review.
What should the client receive after recovery?
Provide the incident window, affected request count, resolved items, and named owners for anything outstanding. Use the voice integration guide to review the normal action path before closing the incident.