From a Salesforce admin perspective, we pass the isPremium flag from SFDC to GC via a Data Action.
The Apex code returns true as a Java boolean, but the Data Action translationMap converts it to the string "true". This type coercion is silent and breaks every boolean comparison downstream in Architect. Always use ToString(isPremium) == "true" as a workaround.
Before sending daily reports via email, verify with your DPO that the wrap-up code summary doesn’t contain PII.
If your wrap-up codes include labels like ‘Customer Complaint - [Customer Name]’, that email is transmitting personal data. Under GDPR Article 32, email is not an adequate safeguard for PII transmission unless it is encrypted end-to-end.
Our Knowledge Workbench articles aren’t surfacing during Agent Assist in real-time.
We imported 500 FAQ articles, tagged them with categories, and enabled the knowledge base in the queue settings. But during live calls, the agent’s knowledge panel stays empty. The transcription is working - I can see the words appearing - but no articles are suggested. Is there a minimum confidence threshold?
The Knowledge Workbench requires articles to be in ‘Published’ state, not ‘Draft’.
I’ve helped many community members with this. After importing articles, they sit in ‘Draft’ by default. You must go to Knowledge > Articles and publish each one individually (or use the bulk publish action). The Agent Assist engine only indexes published articles.
From a Salesforce perspective, we actually bypass the native Knowledge Workbench entirely.
Our knowledge base lives in Salesforce Knowledge. We built a Data Action that queries the SFDC Knowledge REST API in real-time using the transcription keywords. The results are pushed to the agent’s script panel via participant data. This gives us a single source of truth for knowledge across both SFDC and GC.
If you are using dual-channel recording, the knowledge base matching improves dramatically.
The engine only needs to match against the CUSTOMER’s speech, not the agent’s. With mixed recording, the agent’s voice pollutes the keyword extraction, causing irrelevant articles to surface. Dual-channel isolates the customer’s words and produces much more accurate knowledge suggestions.