Set rules and put human checks in place

Decide what the system will do and when it should stop

Automation needs a clear rule for the usual case and instructions for what to do when there’s uncertainty. Write down the trigger, conditions, output, and the circumstances in which the process should stop or request a review. Avoid rules that carry on silently when required data is missing.

Assign an owner for the process. Someone needs to know who monitors the result, who handles alerts, and who can pause the automation. For a low-risk task, checking a regular sample may be enough. For a payment, sending sensitive data, or making a decision with a significant impact, consider requiring human approval before completing the action. A check shouldn’t just be another button that people click automatically; it should be clear what they’re approving.

Plan how to return to the original process in advance. Find out whether you can correct an inaccurate record, cancel a message, or turn off the automation without losing important data. When creating a notification, explain what happened and what action to take. If testing reveals many cases that need manual review, don’t treat that as a failure; you may have found the point beyond which the task shouldn’t be automated. Clear rules help people know when they can rely on routine automation and when they need to take over the decision.

Assign human checks to someone who understands what they’re reviewing and can change the result. If someone is only confirming every alert without time to assess it, the check loses its purpose.

The automation encounters missing data. How should it be designed to respond?

Automation in Practice: Choose Tasks and Check Results

Continue the training

Enter your email address to continue the training.