Compare a successful run with a failed one
Note when it last worked, what changed and which step first differs. Look at actual input and output, with sensitive data redacted. Do not assume the AI step caused an error elsewhere in the process.
Check connections, fields and limits
Review authorisation, permissions, renamed fields and missing values. Look for service errors or a reached limit. Make's error guidance distinguishes issues such as authorisation and rate limits, each needing a different response.
Validate before writing
A result can fit a required format and still be unsuitable. Check required values and use human review where an incorrect result matters. Preserve useful existing data when a step fails or returns an empty value.
Understand replay and failure behaviour
Find out what already completed before retrying. A repeated run can create another task or message. Agree a bounded test, a failure alert and an owner for exceptions. If there are no useful logs, an investigation may need to start by adding visibility.
Official references
Check the current documentation for your account and edition before changing a live process.
Keep passwords, API keys and customer records out of public enquiries. Secure access is agreed separately.
Explore automation repair