Automation Stopped Working? What to Check First

If your automation has stopped working, first check whether it is still sending messages or changing records. Contain incorrect actions, keep the run history and trace one affected item through the connected apps. Check what already completed before you retry: a failed run can still have carried out earlier steps.

Need someone to investigate? Send an automation repair request with the platform, symptoms and impact.

1. Contain incorrect actions

If the workflow is creating duplicates, sending the wrong replies or changing the wrong records, ask the account owner to pause the affected workflow or action. Record what you paused and when. Put a named person in charge of handling new enquiries or tasks manually while it is stopped.

Check separately for runs already in progress, scheduled retries and messages queued in the destination app. Turning off a trigger may leave those items active. Use the platform's controls to inspect or cancel the affected work where available.

For an AI agent taking incorrect actions, restrict the affected tool permission or place the action behind human approval. Keep the input, instructions and proposed action for review. Changing the prompt alone does not establish that a workflow is ready to resume.

2. Keep the details needed to diagnose it

Before editing or deleting anything, save the relevant configuration or version where your platform allows it. Keep incident notes in a location your team controls:

• Workflow name, platform and connected apps.
• Time of the last correct result and first incorrect result, including the time zone.
• Affected run IDs and the exact error message.
• One expected result compared with what happened.
• Recent changes to accounts, fields, filters, code, prompts or billing limits.
• What has been paused and whether anything is still queued.

Logs and exports can contain personal data, tokens and credentials. Keep originals within your controlled account. Remove private details from screenshots or error text you send in an enquiry.

3. Start with the symptom

Microsoft's Power Automate troubleshooting guide distinguishes trigger, action and logic failures. That distinction is also useful when investigating other workflow tools: first establish whether the work started, failed partway through or completed with the wrong result.

4. Check connections and recent changes

Use the error to narrow the investigation. An expired connection needs a different response from a renamed field or an exhausted usage allowance. Check the provider's status page as well as your own run history.

Look at the account the workflow runs under. Can it still access the right mailbox, CRM pipeline, folder or spreadsheet? Did someone leave the business, change permissions or reconnect a different account? Ask the account owner to restore only the access the workflow needs.

Next, compare the last successful input with a failing one. A blank value, renamed field or changed record ID can alter the outcome even when the connection still works.

5. Check duplicates before replaying failed runs

A timeout does not prove the destination did nothing. The app may have created the record while the automation was waiting for a response. Zapier identifies replaying timed-out actions as a possible source of duplicate data.

Before retrying an affected item:

1. Search the destination using a stable source ID, such as the enquiry or order reference.
2. Record which actions completed, including messages already sent.
3. Check where the chosen replay option starts and which workflow version it uses.
4. Test one controlled item before considering a larger replay.

Platform behaviour differs. Read Zapier's replay guidance or Make's incomplete execution guidance for the action you intend to take. Make notes that retried incomplete executions use the configuration from when the error occurred; editing the current scenario is not enough to assume an old execution will use the change.

6. Reconcile missing leads and affected records

Use the source system as the starting list: form submissions, orders, bookings or inbound messages. Compare it with the destination over the affected time period. Separate items into completed, missing, duplicated and uncertain.

For missing leads, check more than whether a CRM record exists. Confirm the owner, pipeline stage and expected follow-up. A lead in the wrong queue still needs attention.

For duplicates, keep the record history and agree which record to retain before merging or deleting. Record any manual follow-ups so a later replay does not contact the same person again.

7. Check the repair before restarting

Agree what a correct result looks like, then test with controlled data and recipients. A useful acceptance check covers:

• One new input reaches the right destination with the right fields.
• Repeating that input does not create an unintended second action.
• Missing or invalid data goes to review instead of producing an incorrect update.
• An unavailable connection produces a visible alert for the named owner.
• AI actions outside the approved rules stop for review.
• The backlog is reconciled with work already completed manually.

Keep the change record and a way to pause the workflow. Resume in a controlled sequence, then inspect the first live results in the destination systems.

Get help repairing your automation

Our automation repair service covers failed workflows, duplicate actions, missing leads, broken integrations and incorrect AI actions, including workflows someone else or an AI tool helped you build.

Send your repair request. Name the tools, when the issue started and what is affected.