When automations fail silently: alerts, retries and error paths
Short answer: Automations rarely fail loudly. A login expires, a form changes or an app updates, and leads stop flowing into your CRM while everyone assumes it's working. The fix is to plan for failure: send error notifications to someone who will act, retry temporary failures automatically, add error paths for predictable problems, check volumes daily so silent gaps show up, and give every automation a named owner.
Small UK businesses increasingly run on automations: web form to CRM, order to spreadsheet, booking to calendar, invoice to accounting. When they work, nobody thinks about them. When they break, the damage often goes unnoticed until a customer complains. Here's how to recognise the usual failure patterns and what to put in place for each.
Symptom 1: "We never got that enquiry"
A customer says they filled in the form; it never reached the team.
Likely causes:
- The automation errored and notifications went to an inbox nobody reads, often the email address of whoever set up the account.
- Notifications were switched off or batched.
Fix: check where error notifications go and how often. Zapier, for example, sends error notifications to the account email by default and lets you choose a frequency: immediately (the default), immediately plus an hourly summary, an hourly summary only, or never, which Zapier itself doesn't recommend. You can also label workflows by priority. Route alerts for important workflows to a shared address or a team channel someone watches.
Symptom 2: Failures during a short outage are lost forever
An app was down for twenty minutes, and every order in that window is missing from the spreadsheet.
Likely cause: no automatic retry for temporary failures.
Fix: turn on retries where your platform supports them. Zapier's troubleshooting guide describes a "Scheduled" status for runs that errored while Autoreplay is enabled, meaning Zapier will retry them automatically, and a "Held" status for runs you can replay once the issue is fixed. Check what your platform offers, and make sure retries can't create duplicates (see Symptom 5).
Symptom 3: The automation stops entirely and nobody notices
No errors are reported because nothing runs at all: the trigger stopped firing, the workflow was switched off, or a connected account was disconnected.
Likely cause: monitoring relies only on error alerts, and "nothing happening" isn't an error.
Fix: add a volume check. Once a day, compare what should have happened with what did:
- form submissions on the website versus new CRM records;
- orders in the shop versus rows in the sheet;
- bookings in the calendar versus confirmation emails sent.
A simple scheduled automation that counts yesterday's records and alerts if the number is zero, or far below normal, catches silent failures that error notifications miss.
Symptom 4: One bad record breaks the whole run
A single order with a missing field or an unusual character stops processing for everything after it.
Likely cause: no error handling for predictable problems.
Fix: build error paths. In n8n, for example, you can set an error workflow in a workflow's settings; it runs whenever an execution fails and starts with an Error Trigger node, so you can send an email or chat alert with details of what failed. The same error workflow can serve several workflows. n8n also provides a Stop and Error node to deliberately fail a run under conditions you choose, such as a missing email address, so it's routed to the error workflow instead of continuing with bad data. Zapier's troubleshooting guide likewise describes a "Handled error" status, where a custom error handler ran an alternative path.
Good error paths:
- send the problematic record to a "needs attention" list;
- notify a person with enough detail to fix it;
- let the rest of the data continue.
Symptom 5: Duplicates after retries
After an outage, customers receive two confirmation emails, or the CRM has two copies of each lead.
Likely cause: retries without checks for whether the action already happened.
Fix: before creating a record, search for an existing one by a unique value (order number, email plus date, form submission ID). Make actions safe to repeat.
Symptom 6: Nobody knows what the automation does
The person who built it left. When it breaks, nobody knows what it was supposed to do.
Likely cause: no documentation and no owner.
Fix: for every automation, keep a short note:
- what it does, in one sentence;
- what triggers it and where data goes;
- who owns it, and who is the backup;
- where alerts go;
- what to do if it fails (manual workaround).
Review the list quarterly and remove automations nobody uses.
Symptom 7: Errors you can't diagnose
You get an alert, but it only says "failed".
Likely cause: not enough information captured.
Fix: include the workflow name, the failing step, the error message and a link to the run in every alert. n8n's documentation lists execution details, including an execution URL when the execution is saved, among the data available to an error workflow; Zapier keeps a run history you can review.
A minimum standard for every important automation
- Alerts go to a watched address or channel, immediately.
- Retries enabled for temporary failures, with duplicate checks.
- Error paths for predictable bad data.
- A daily volume check against the source.
- A named owner and a one-paragraph description.
- A manual fallback written down.
FAQ
How many alerts is too many?
If people start ignoring them, too many. Send immediate alerts only for critical workflows and use summaries for the rest.
Should we avoid automations because they fail?
No. Manual processes fail too, just less visibly. Plan for failure and automations become more reliable than retyping.
Which platform handles errors best?
Most mainstream platforms offer notifications, retries and error paths. How you configure them matters more than which one you use.
How often should we review automations?
Check alerts daily, volumes weekly at first, and the whole list quarterly.
If your business runs on automations nobody is watching, send us a list of the important ones and we'll suggest alerts and checks for each.