Reviewing feedbacks from implicit intent detection
Zoom Virtual Agent supports collecting feedback on detected intents. This feature allows:
- Zoom Contact Center AI Expert Assist agents to provide explicit feedback on intent results within AI Expert Assist.
- End users interacting with the Zoom Virtual Agent chatbot to have their feedback inferred through their behavior during interactions.
- Admins to review the feedback and take appropriate actions to refine intent detection accuracy and enhance the overall user experience.
Requirements for reviewing intent feedbacks
- For Zoom Virtual Agent
- Account owner or admin privileges; or relevant role/privilege
- Basic, Pro, Business, Education, or Enterprise account
- Zoom Virtual Agent license
- For Zoom Contact Center AI Expert Assist
- Zoom Contact Center license
- Elite license, or Basic/Essential license with AI Expert Assist add-on
How to provide feedback on detected intents
Mark intents as Correct or Incorrect (For Zoom Contact Center AI Expert Assist agents )
During an active voice or messaging engagement, AI Expert Assist displays an icon next to relevant messages when an intent match meets the confidence threshold set by the admin.
- The Information Retrieval details appear in the AI Expert Assist tab.
- When the agent hovers over the Information Retrieval card, thumbs-up and thumbs-down icons appear for feedback.
- Agents can provide feedback by clicking:
- Thumbs-up (Helpful): Confirms the intent detection was accurate.
- Thumbs-down (Not Helpful): Indicates the intent detection was incorrect.
Infer feedback from end users interactions (For Zoom Virtual Agent)
End users do not provide direct feedback but are evaluated based on their interaction behavior. The system assigns feedback labels based on specific patterns:
No Response
When an end user's query matches an intent but the user does not follow the expected flow, this feedback is considered as No Response. Similarly, if multiple intents are matched but the user selects only one or none, the feedback for the unselected intents is also considered as No Response.
Example: A banking virtual assistant helps a customer with account management.
- End user: "I want to check my savings and transfer some money."
(Matches Intent A: Check Savings Account Balance and Intent B: Transfer Funds) - Bot: "Would you like to check your balance first or proceed with the transfer?" (Displays two buttons: Check Balance and Transfer Funds).
- End user: Clicks Check Balance and completes the balance inquiry but does not proceed with the transfer request.
- The query ("I want to check my savings and transfer some money.") appears in Intent B's Coach page > Matched Queries list with label No Response as the user did not follow through with the second detected intent.
Sub Flow Interrupt
The end user is guided into a sub flow but shifts away by asking a different question instead of completing the expected interaction. The feedback is considered as Sub Flow Interrupt.
Example: A banking virtual assistant assists a customer with a credit card payment inquiry.
- End user: "How can I pay my credit card bill?"
(Matches Intent A: Credit Card Payment Methods) - Bot: "You can pay via online banking, mobile app, or at a branch. How would you like to proceed?"
- End user: Instead of choosing a payment method, the end user asks a new question: "What is the credit limit on my card?"
(Triggers Intent B: Check Credit Limit) - End user proceeds with checking their credit limit.
- Query A ("How can I pay my credit card bill?") appears in Intent A's Coach page > Matched Queries list with label Sub Flow Interrupt, as the end user abandoned the expected response path.
Note: The examples provided illustrate common scenarios, but other variations may exist.
How to view feedback in the intent coach page
- Sign in to the Zoom web portal.
- In the navigation menu, click AI Management then Intent Management.
- Click an intent group.
- Click the Coach tab.
- Click the Matched Queries sub-tab. The list displays queries that match this intent.
- In the Intent matching outcome column, view the feedback corresponding to each query. You can take appropriate actions on the queries based on your assessment.