Appearance
Providing Feedback on AI Decisions
The Training Loop system allows you to provide feedback on AI decisions to help improve performance over time. Your feedback helps identify what works well and what needs improvement.
Overview
Feedback is a simple way to tell the system whether an AI decision was helpful or not. This feedback is used to:
- Track quality: See which decisions are working well
- Identify issues: Find patterns in problematic decisions
- Guide improvements: Use feedback data to improve AI performance
- Compare versions: See how different AI versions perform based on feedback
How to Provide Feedback
In Conversation Views
When viewing a conversation, you'll see feedback controls for AI actions:
- Thumbs Up (👍): Click to indicate the AI decision was helpful
- Thumbs Down (👎): Click to indicate the AI decision was not helpful
Optional Reason Selection
After clicking thumbs up or down, you can optionally select a reason:
- Helpful: The decision was useful and appropriate
- Unhelpful: The decision wasn't useful but wasn't necessarily wrong
- Incorrect: The decision contained incorrect information
- Inappropriate: The decision was inappropriate or violated guidelines
- Other: Any other reason (you can add notes)
Feedback Timing
You can provide feedback:
- Immediately: Right after seeing an AI decision
- Later: After seeing the outcome (customer reply, escalation, etc.)
- Anytime: Review past conversations and provide feedback
What Gets Recorded
When you provide feedback, the system records:
- Feedback value: Thumbs up or thumbs down
- Reason: The reason you selected (if provided)
- Timestamp: When you provided the feedback
- Your user ID: Who provided the feedback (for tracking)
This feedback is linked to the training record for that AI decision.
Feedback Best Practices
When to Give Feedback
Provide feedback when:
- AI did something great: Positive feedback helps identify what works
- AI made a mistake: Negative feedback helps identify issues
- Outcome is clear: After seeing how the customer responded
- Pattern emerges: When you notice recurring issues
What to Consider
When providing feedback, consider:
- Accuracy: Was the information correct?
- Helpfulness: Did it help the customer?
- Appropriateness: Was the tone and content appropriate?
- Outcome: How did the customer respond?
Be Specific
When selecting a reason:
- Helpful: Use when the AI provided useful information or assistance
- Unhelpful: Use when the AI didn't help but wasn't necessarily wrong
- Incorrect: Use when the AI provided wrong information
- Inappropriate: Use when the AI violated guidelines or was inappropriate
- Other: Use for any other issues (add notes if possible)
Viewing Feedback
Your Feedback
You can see:
- Feedback you've given: All feedback you've provided
- Feedback trends: How your feedback changes over time
- Feedback by type: Breakdown by reason
Team Feedback
If you have access:
- All team feedback: See feedback from all team members
- Feedback statistics: Overall feedback trends
- Problem areas: Decisions with negative feedback
Feedback Analytics
Feedback data is used in:
- Cohort comparison: Compare feedback across different AI versions
- Performance metrics: Track quality improvements over time
- Issue identification: Find patterns in negative feedback
- Success tracking: Identify what's working well
Privacy & Anonymization
- No PII in feedback: Feedback doesn't include customer information
- Secure storage: All feedback is stored securely
- Access control: Only authorized users can view feedback
Tips for Effective Feedback
- Be consistent: Use the same criteria for similar situations
- Provide context: Use the reason field to add context when helpful
- Review regularly: Check feedback trends to identify issues early
- Focus on outcomes: Consider how the customer responded
- Balance feedback: Provide both positive and negative feedback
Common Scenarios
Scenario 1: AI Provided Correct Information
Action: Click thumbs up, select "Helpful"
When: The AI answered a question correctly and the customer was satisfied
Scenario 2: AI Made a Mistake
Action: Click thumbs down, select "Incorrect"
When: The AI provided wrong information or misunderstood the question
Scenario 3: AI Was Inappropriate
Action: Click thumbs down, select "Inappropriate"
When: The AI used inappropriate language or violated guidelines
Scenario 4: AI Didn't Help
Action: Click thumbs down, select "Unhelpful"
When: The AI didn't provide useful information but wasn't necessarily wrong
Next Steps
- Learn about Analytics & Cohorts - How feedback is used in analytics
- Review Training Loop Overview - Understand the full system
- Check Training Context - How AI uses your business context