I’ve spent hours trying to figure out why the bot flow is proceeding to the ServiceNow incident creation stage despite the NLU confidence score falling below our defined threshold. The environment is Genesys Cloud Engage (EU-West-1) running the latest stable release, integrated with ServiceNow via a custom Data Action. The bot is configured to handle digital channel interactions, specifically WhatsApp and Web Chat. The Architect flow utilizes the ai.chat node for intent classification, followed by a conditional gate checking if ai.chat.confidence > 0.75. When the confidence is lower, the flow should route to a human agent via the queue.add node. However, for specific intents related to ‘account suspension’, the flow bypasses this check and triggers the webhook to ServiceNow, resulting in invalid incident records being created. The ServiceNow REST API endpoint is api/now/table/incident, and the payload includes the transcript and metadata. The issue persists across multiple test runs. The webhook response from ServiceNow is a 201 Created, confirming the request was processed, but the data quality is poor because the intent was not clearly identified by the AI. I have verified the NLU model training data and retrained the model twice, but the confidence scores for ‘account suspension’ remain volatile, often landing between 0.60 and 0.70. The Architect flow logic seems sound, as other intents with low confidence correctly route to agents. Is there a known issue with the ai.chat node returning cached or default confidence values for high-priority intents? Or could the webhook execution be overriding the conditional gate logic? I have attached the flow JSON snippet and the ServiceNow incident creation script for reference. The timezone for all logs is Europe/London. Any insights into why the conditional gate is being ignored would be appreciated.