A workflow watching record updates can write to that record and create the next update itself. The run history looks repetitive because each write makes a new return event. Microsoft and Zapier document this failure mode, including feedback between two workflows.
Short answer
Pause the affected workflow, especially if it sends messages, creates records or consumes tasks. Follow one record ID through the history: initial event → workflow write → next trigger event. If the write caused the next run, narrow the trigger to a meaningful change or use a processing state that excludes the workflow’s own write. Test with a dummy record before enabling it again. A delay alone moves the next run; it does not remove the return path.
Identify the kind of repeat
| History pattern | Compare | Next decision |
|---|---|---|
| The same event is delivered twice | Event ID and delivery attempts | Use the duplicate webhook guide to control the business effect |
| A new update follows each run | Record ID, write time and next trigger time | Find the field or system your action changed |
| Two workflows alternate | A writes to B; B writes back to A | Assign ownership of the change and exclude the return |
| Several items run inside one execution | Input items and internal iteration | Review deliberate looping, not the external trigger |
This is Pulse’s diagnostic model. Do not permanently block every later update to the record: a colleague may make a valid edit tomorrow.
Trace the return event
Person edits Record 42 → trigger A → A updates Record 42 → new update event → trigger A
In an authorized account, record the record ID, run times, changed field and writing account. The example is hypothetical. Keep customer data and tokens out of public troubleshooting tools. For two workflows, draw both arrows and locate the first write that matches the other workflow’s entry condition.
Where supported, Power Automate offers trigger conditions and column filtering. A condition later in the flow may stop subsequent actions while a run has already started; check the actual connector and plan. In Zapier, compare the “new or updated” trigger with the update action’s target. Connector behaviour differs across products and regions; US, Canadian and UK teams should verify the features shown in their own account.
Prove the stop condition
- Use a dummy record, with real recipients and external writes disabled where possible.
- Make a legitimate edit that should cause one business action.
- Check that the automated write causes no second business action. A return event may still exist if it is safely filtered or stopped.
- Make another valid human edit to the same record and confirm it still works.
- Restore the workflow in stages and watch run history and task use. Pause again if feedback returns.
A simple “processed” flag can suppress valid future edits unless it is tied to a meaningful state transition. Filtering by changed fields, status or writer depends on what the connector exposes. For a cost estimate, see the automation cost model. The appropriate next step is a controlled workflow fix, not a paid upgrade.