Can’t get this config to load properly… The bot is failing to trigger the ‘order_status’ intent despite the transcript matching perfectly, whereas the equivalent Zendesk trigger would have fired instantly.
“If the confidence score is below the threshold, the interaction continues to the next node.”
The Architect flow just skips to the fallback without logging why. Is there a way to log the actual confidence score in the debug stream?
This looks like a configuration visibility issue rather than a true failure. When NLU scores drop below the threshold, the system silently routes to the fallback to maintain flow stability. To diagnose the exact score, try these adjustments:
Enable “Debug Mode” in the Architect settings for that specific flow. This forces the system to output raw NLU metadata, including confidence percentages, to the interaction log.
Add a temporary “Set Variable” step before the NLU node. Capture the system.nlu.confidence value into a custom variable. Then, route that variable into a log message node so you can see the exact number in the transcript.
Check if your intent model needs retraining. If the score is consistently hovering just below the threshold (e.g., 0.68 vs 0.70), adding more diverse utterances for the ‘order_status’ intent usually boosts accuracy.
Consider lowering the threshold slightly for this specific node if false negatives are high. You can set node-level overrides in the NLU configuration panel.
Monitoring the actual scores helps determine if it is a data gap or a strict threshold setting.