Automation Failure Checklist

Before you trust anything you've built, or anything your team has built, ask one question: who finds out when this breaks?

The 3-part checklist

Automation: ___

1 · An owner. One named person who's responsible for it. Owner: ___

2 · A success signal. Something that proves it ran today. A report that arrives. A row that updates. A message in a channel. Signal: ___ Where the owner sees it: ___

3 · A failure alert. If it doesn't run, a message goes to that owner. Alert goes to: ___ Tested on: ___

If the answer to "who finds out" is nobody, it isn't finished yet.

Ask your tool to add the alert

You don't need to code this. Ask the tool you built it with, or its built-in assistant:

This automation [what it does] must tell [owner] when it fails. Add a step that sends [owner] an email if any step fails, with the name of the step that failed and a link to the run. Show me where to test it.

What to look for in common tools

Settings change often, so check your tool's own help pages before you rely on these.

Zapier. Error emails go to your Zapier account email by default. Check your notification settings so they aren't set to "Never", and make sure the account email is one the owner actually reads.

Power Automate. Add a "send email" step and set its "Configure run after" to "has failed", so it only runs when an earlier step fails. Include a link to the failed run.

ChatGPT scheduled tasks. A task can pause if it's inactive or needs action. The result arrives by push notification or email, so the arrival is your success signal. If it doesn't arrive, open Scheduled in ChatGPT and check whether the task is paused.

Break it on purpose, once

Pick a safe way to make it fail. Then check that the alert reaches the owner.

Did the alert arrive? Did it say what broke? Turn everything back afterwards. Then write the test date on the checklist.

The weekly 2-minute check

Every Monday, for each automation you own:

Why this matters

An automation that fails quietly is worse than no automation, because you stop checking. Mine failed for 9 weekdays in a row while I was away, and nothing told me.