The Genesys Cloud Architect flow is routing interactions to the ‘Human Agent’ queue instead of the ‘Digital Self-Service’ path, even when the user input clearly matches our primary intent. The bot response log shows intent_confidence: 0.42, which falls below our configured threshold of 0.65. In the old Zendesk Answer Bot setup, this same query triggered the correct article with a relevance score of 0.88 because the fuzzy matching algorithm was more lenient with synonym variations.
We are migrating from Zendesk Sunshine Conversations to Genesys Cloud CX. The environment is EU1. We have imported the Zendesk Help Center articles into the Genesys Knowledge Base and mapped the old bot intents to new NLP models. The specific issue arises with queries containing mixed language terms or slang, which Zendesk handled via its native keyword matching but Genesys is treating as low-confidence noise. We tried adjusting the minimum_confidence setting in the bot configuration panel, but lowering it to 0.4 causes false positives on unrelated intents, leading to a confusing user experience.
How do we tune the NLP model or the Architect flow logic to mimic the broader matching behavior of Zendesk without sacrificing accuracy? We need a strategy to handle these edge cases where the intent is clear to a human agent but the AI confidence score remains stubbornly low. Any advice on using fallback intents or adjusting the training data weights would be appreciated. We want to ensure the migration feels seamless for customers who are used to the previous bot’s responsiveness.