Operations, Workflows, Automation Health. This surface exists because every layer of the automation chain treats "no error thrown" as "succeeded", which let problems live undetected for days. It makes the chain loud. It refreshes every 30 seconds.
Come here first whenever something did not happen that should have.
The Chain
Understanding the sequence is most of the diagnosis:
Something happens → an event is recorded → a trigger matches it → the workflow executes → a notification is queued → the notification is sent
A failure at any point looks identical from the outside: nothing happened. The four panels tell you which link broke.
The Four Panels
Top-line strip - the summary tiles. Start here for the shape of the problem.
Event stream - the last 30 events recorded. Each row shows what happened and when. Crucially, an event carries an orphan flag when no workflow trigger matches it.
Workflow executions - the last 30 runs, with duration and any error.
Notification outbox - pending and recently sent, side by side. A growing pending column means notifications are being created but not delivered.
Above the panels is a diagnostic hints banner with plain-language messages about common failure modes. Read it before reading the panels - it often names the problem outright.
Diagnosing "My Automation Did Not Run"
Work down the chain in order. Stop at the first thing that is wrong.
- Is the event in the event stream? If not, the thing you think happened did not emit an event. Check you performed the action that emits it - moving a deal by dragging it and changing it in the record can emit different events.
- Is the event flagged orphan? That means the event fired but nothing is listening. Your trigger does not match the event key. Open the workflow and check the trigger is the event you actually see in the stream.
- Is there an execution? If the event matched but no execution appears, the workflow is probably inactive. Saving and activating are separate steps.
- Did the execution error? The executions panel shows the error. Most are prerequisites - see Automation Actions and What They Need.
- Did the execution succeed but nothing arrived? Look at the notification outbox. If it is pending and staying pending, delivery is stuck. If it never appeared, the action did not queue anything - check the action's recipient resolved to an actual person.
Tip
The orphan flag is the single most useful thing on this page. It distinguishes "the platform never noticed" from "the platform noticed and nothing was listening", and those have completely different fixes.
Reading Execution Duration
Duration is on every execution. A workflow that normally takes under a second and suddenly takes thirty is usually waiting on an external call - a webhook to a slow endpoint, or a provider under load. Consistently slow executions are worth investigating even when they succeed, because they will eventually time out.
When The Outbox Is Backing Up
A growing pending count with a static sent count means delivery is stuck rather than slow. Notifications are being created correctly - the problem is downstream. This is a support matter rather than something to fix in the builder; note the time it started and get in touch.
What This Page Does Not Tell You
It shows the last 30 of each. For a failure last week you will need support rather than this page. It also does not show what a workflow should have done - only what it did. Comparing intent against behaviour is still your job, which is why the dry run in the builder matters.
Common Questions
Everything looks green but my colleague did not get the email. The chain succeeded up to sending. Check their notification preferences and whether the rule includes email at all. See Notifications in the Settings section.
I see executions I did not expect. Your trigger is too broad. A "deal changed" trigger fires on every edit. Narrow it to the specific event.
The event stream is empty. Nothing has happened recently, or the page needs its refresh. It only holds the most recent 30.
Can I replay a failed execution? Not from this page. Fix the cause and let the next event trigger it, or trigger it manually by performing the action again.
Did this answer your question?
No, ask a person