Contents
At a glance
- Category
- Automation mechanics
- Five failure classes
- Transient, authentication, data, logic, downstream rejection
- Transient failures
- Retry with exponential backoff and jitter
- Authentication failures
- Do not retry - alert a human, the connection needs reauthorising
- Data failures
- Route the item to a review queue rather than failing the whole run
- Worst outcome
- Silent success - a run that reported fine and did nothing useful
Why error handling is most of the work
The happy path of an automation is the easy part and usually the smaller part. Error handling is everything else: what happens when the CRM is briefly down, when a required field is empty, when the API returns a shape you did not expect, when the same event arrives twice, when a customer's phone number has a space in it.
The gap between a demo and a production automation is almost entirely this. A demo runs once with clean data. A production workflow runs four hundred times a day against data entered by people in a hurry, and every one of those cases will occur.
The failure classes to plan for
Five, each needing a different response. Transient - a timeout, a 5xx, a rate limit: retry with backoff. Authentication - a 401 or 403: retrying is pointless, alert a human immediately because the connection needs reauthorising. Data - a missing or malformed field: route the item to a review queue rather than failing the whole run.
Logic - a record in a state you did not anticipate: log it with the full context and stop, because guessing is how records get corrupted. Downstream - the target system rejected the write: capture its error message verbatim, because it usually names the exact field that was wrong.
Building it in a workflow
Four components. A validation step early that checks required fields are present and correctly shaped, so bad data is caught before it reaches an action. An error route on steps that can fail, which every serious platform supports - n8n has error workflows, Make has error handlers, Zapier has error paths.
A failure record: write the failed item, the step, the error message and a timestamp somewhere durable, which is the cheap version of a dead-letter queue. And an alert, sent to a person, when failures exceed a threshold or when any authentication error occurs. Without the last one the first three are archaeology rather than operations.
The silent failure problem
The worst outcome is not an error - it is a run that reports success and did nothing useful. A mapping produced an empty string and the API accepted it. A filter excluded everything and the workflow completed with zero items. A model returned a valid but wrong classification. None of these raise an exception anywhere.
Guard against them with positive assertions rather than absence of errors: check that the expected number of items came through, that critical fields are non-empty before sending, that the downstream write returned the ID it should have. And alert on suspicious silence - a workflow that normally runs forty times a day and ran twice is a problem that no error log will show you.
What you should know
Decide the failure policy per workflow before building, and write it down: which failures retry, which route to a human, who gets alerted, and what the customer experiences meanwhile. For anything customer-facing, that last question is the important one - a quote follow-up that fails silently costs a job, so the fallback should be a human notification rather than nothing.
Log enough to diagnose without reproducing: the input, the step, the raw error, and a run ID that also appears on the related record. That is the same discipline as observability in an agent, and it is what turns a 6am incident into a five-minute fix.
Error handling FAQ
What is error handling in workflow automation?
Everything the workflow does when a step fails: retrying transient problems, routing bad data to a human, recording the failure, and alerting someone. In a production automation it is typically more of the build than the path where everything works.
What should I do when a step fails?
It depends on why. Timeouts, 429s and 5xx errors should retry with backoff. Authentication failures should alert immediately, since retrying will not fix them. Bad data should route to a review queue. Unexpected record states should stop and log rather than guess.
How do I catch failures I did not anticipate?
With positive assertions rather than absence of errors: check the expected number of items arrived, that critical fields are non-empty before sending, and that writes returned an ID. Also alert on suspicious silence - a workflow that usually runs forty times and ran twice.
What is a silent failure?
A run that reports success while achieving nothing useful - an empty value the API accepted, a filter that excluded everything, a model that returned a plausible wrong answer. Nothing errors, so nothing alerts, and you find out when a customer asks why nobody called.
Where should failed items go?
Somewhere durable and visible: a table recording the item, the step, the raw error message and a timestamp, with an alert when it is not empty. That is the cheap equivalent of a dead-letter queue and it turns invisible failures into a list someone can work through.
How much error handling does a simple workflow need?
At minimum: validation of required fields before any action, an error route that records failures, and an alert on repeated failure or any authentication error. That is perhaps an hour of work and it is the difference between an automation you trust and one you check manually.
Primary source
n8n - error handling documentation
Checked August 2026. Anything priced or rate-limited moves, so verify against the source before you budget.
Want this built rather than explained?
Knowing what error handling means is one thing; having it running in your business is another. We scope one workflow at a time, quote a fixed price in writing after a free consult, build it, and hand it over with documentation you own.
