In this article
An account stops producing its weekly reports. Product usage falls, the customer receives a re-engagement email, and nobody notices that its data connection expired. The customer may still want the product; the product has stopped doing the job.
A disconnected integration deserves its own recovery path. Detect the interruption, reach someone able to fix it and reconcile the work that did not complete. Until those steps are resolved, a declining activity chart is an incomplete diagnosis of the customer’s intent.
Establish what was working before the interruption
A recovery record should begin with the last confirmed useful result. For a reporting integration, that might be a report generated from current source data and delivered to its intended destination. Record the source, destination, connection owner and expected cadence alongside it.
This gives the team a boundary for investigation. A connection that never completed its first successful sync belongs in setup support. A previously reliable connection that stopped after a permission change is a continuity problem. Those customers need different instructions.
Keep several timestamps separate: last successful authentication, last source data received, last completed processing job and last output delivered. Their differences help locate the break. If authentication succeeds but the data is stale, another “reconnect your account” message may simply send the customer in circles.
Describe the failure in terms the customer can act on
An internal error code is useful for support. The customer also needs to know which work is affected, whether anything has been lost and who can take the next step.
Zapier’s connection guidance documents expired connections and a reconnect action available for connections the user owns. A connection owner matters because the person looking at a failed report may not be able to renew the authorization.
Design the notice around that handoff. It might say that a named data source needs reconnection, identify the responsible administrator and allow the reader to send that administrator a request. Preserve the failure context in the request so support does not have to reconstruct it through screenshots.
Avoid claiming a precise cause before you know it. A revoked permission, changed account, unavailable service and malformed record may produce similar symptoms. Explain what is established and what the next action will test.
Track recovery as a sequence
The following state map is an editorial design tool. It describes a fictional reporting integration, not the behavior of every integration platform.
| State | What the product knows | Next useful action |
|---|---|---|
| Interrupted | Expected work did not complete | Identify affected jobs and owner |
| Owner notified | A request reached the responsible person | Preserve context and show current status |
| Reauthorized | The connection can authenticate again | Test required access with the actual workflow |
| Current work restored | A new job completes successfully | Decide what to do with missed work |
| Backlog reconciled | Missed jobs are replayed, skipped or otherwise resolved | Confirm the next expected result |
The transition from reauthorized to restored is especially easy to skip. A valid token cannot tell you whether the destination still exists, the account can read the right records or a changed field mapping now produces a usable report.
Do not make a customer redo every setup step merely to prove the connection works. Offer the smallest meaningful check, then show its result. Where the check writes data or sends a message, explain that effect before running it.
Decide which missed work is still worth doing
Replaying everything is not always the right recovery. A missed daily summary may be useful for an archive but pointless as a late notification. A missed invoice or customer record may need reconciliation. The product team has to make that distinction before building a single “retry all” button.
For each missed job, retain its intended action, original time, destination and outcome if any step succeeded. A replay policy can then answer whether to use the original inputs or current data, whether the result is still timely and how to avoid duplicating a completed side effect.
Zapier’s replay documentation distinguishes retrying errored steps from rerunning an entire workflow. Its example of an email succeeding before a spreadsheet step fails shows why the boundary matters. The documented replay attempts the failed spreadsheet step without sending the successful email again.
That behavior is specific to the described replay mode. Check the current rules of the platform and workflow you actually use. A retry is an attempt; the same documentation warns that replay success is not guaranteed.
Make a reconnection example concrete
Imagine an invented customer whose CRM authorization expires before its Monday sales report. The source sync fails, but an older cached report remains visible. The customer’s manager opens the report twice and assumes it is current.
A login-based engagement measure looks reassuring. The product problem is the stale output. The recovery notice should show the report’s source-data time, explain that Monday’s refresh failed and identify the connection owner. After the owner reconnects, the product should confirm that it can read the required CRM records before marking the issue resolved.
Next, decide what happens to the missed Monday report. In this example, the customer chooses a fresh replacement labeled with the period it covers. The system should not quietly present today’s data as the original Monday snapshot. If a historical snapshot cannot be reconstructed, say so.
Finally, check the next scheduled report. A single successful manual run establishes that the task can work; it does not establish that the recurring schedule, destination and notifications are all operating correctly. Keep that follow-up visible to the customer and the team responsible for recovery.
Keep the recovery message out of the growth campaign pile
A product notice about a broken customer workflow has a different purpose from a message encouraging more use. Give it an appropriate destination and escalation rule. A weekly promotional digest is a poor place to hide the only notice that an essential integration stopped.
At the same time, avoid sending identical failure notices for every retry. Group them around the unresolved incident, update the status when something changes and stop the notices when the relevant recovery condition is met. If ownership changes during recovery, preserve the incident history rather than starting an unrelated support conversation.
Measure the process with task-specific outcomes: time to owner acknowledgement, time to a successful current job, unresolved missed work and recurrence of the same failure. Those observations can reveal a product defect without pretending that every recovered integration prevented churn.
When an account’s activity falls, start from the customer outcome and check whether the product still enables it. Restoring a broken workflow may be the most useful retention work available, even when it produces no new feature announcement.
Sources
Something we should clarify?
Suggest a correction


