Skip to content
Troubleshooting

When the work does not come back right

A practical troubleshooting path for context, integrations, approvals, meeting write-ups, and recurring Workforce AI jobs.

Product Guide explains how the product works. Help Center is for setup steps, account questions, and support.

Do not keep rerunning a bad job and hoping it becomes good. Find which part of the assignment failed.

Check the job definition

Name the start signal, permitted sources, expected output, finish line, and exceptions. If two people would interpret the assignment differently, the AI probably will too.

Check the connection

Confirm the right account is connected, the provider authorization is active, and the relevant file, thread, calendar, or record is inside the granted scope.

Check the facts

Old, missing, or contradictory records are a common cause of a wrong result. Correct the source of truth instead of adding a longer instruction that hides the conflict.

Check the control state

A job may have completed its draft and still be waiting for review. Inspect the approval queue and the configured rule before treating the absence of an external action as a failure.

Bring a useful support packet

Include the time, employee, job, connected tool, expected result, observed result, and a redacted screenshot or record ID. Never send passwords or more customer data than support needs.

For account-specific help, use the Help Center support path. This Product Guide explains behavior; the Help Center handles troubleshooting and account questions.