Edit

Troubleshoot failed quality evaluations

Summary

Quality evaluations can fail when the AI agent can't complete processing. Error details help supervisors and administrators identify the cause, decide whether to retry the evaluation, and choose the next steps.

Quality evaluation failure causes

Quality evaluations can fail for several reasons, including:

  • Missing metadata for the record being evaluated.
  • Network or connectivity issues.
  • Temporary internal service interruptions.
  • Missing user permissions.
  • Invalid evaluation configurations.
  • Disabled flows.
  • An unpublished Quality Evaluation Agent.

Some failures are temporary and can be resolved by retrying the evaluation. Other failures require administrative action before you retry.

Review evaluation errors

When an evaluation fails, the AI agent status is set to Error.

To review the failure reason, add the Error details column to the evaluation grid view. This column displays detailed error information that helps you understand why the evaluation failed.

For bulk evaluation runs, the run history includes a Records failed column. Use this column to identify the failed records in the evaluation run. The evaluation list associated with the run displays relevant error messages for records that encountered issues.

Retry failed evaluations

You can retry failed evaluations from the evaluation grid or from the evaluation list associated with a bulk evaluation run.

Retry failed evaluations only when the error can be resolved through reprocessing, such as temporary service issues or connectivity interruptions.

When you retry an evaluation:

  • The original evaluation configuration is reused.
  • The existing evaluation record is overwritten.
  • The original Created On date remains unchanged.
  • The Modified On and Requested By fields are updated automatically.
  • The evaluation expiration date can be changed.

If a newer version of the evaluation criteria exists, the retry continues to use the originally referenced criteria version. The system doesn't automatically upgrade or replace the criteria version. This behavior helps maintain consistency with the original evaluation context.

Review retry results

After you retry a failed evaluation:

  • If the retry succeeds, the AI agent status changes to Complete.
  • If the retry fails, the AI agent status remains as Error. The error reason is updated with the latest message, and a notification indicates that the retry was unsuccessful.
Error code Error details Recommended action
424000 Agent not Published Ask an administrator to publish the Quality Evaluation Agent in Copilot Studio. Don't retry the evaluation until the agent is published.
429000 Too Many Requests The system is temporarily experiencing high demand. Retry the evaluation later.
500001 Internal server error. Contact Microsoft Support. Contact Microsoft Support. Retry the evaluation if advised.
400001 Cross-Geo Data Movement Disabled Cross-geo data movement isn't enabled. Contact your system administrator to enable the setting before retrying the evaluation.
403000 Authentication Issue - Missing Privileges The user doesn't have the required permissions to run the evaluation. Contact your system administrator. Don't retry until the required access is granted.
404000 Source record missing data The source record can't be found or accessed. The record might be deleted or might not be within the accessible business unit. Contact your system administrator before retrying the evaluation.
400031 Input configuration incorrect. Contact administrator. The evaluation configuration is invalid. Contact your system administrator. Don't retry until the configuration is corrected.
500012, 500024 Internal server error. Contact Microsoft Support. Contact Microsoft Support. Retry the evaluation if advised.

Next steps

  • Monitor evaluation run history to identify failed records.
  • Review error details before retrying an evaluation.
  • Contact your administrator or Microsoft Support when the error message indicates that reprocessing won't resolve the issue.

Related content

  • Last updated on