Quality scorecards
Set the criteria, weights and instructions the AI uses to score every conversation.
Scorecards
Assigned by category and interface
- Active
- Active
- Draft
Support v4 criteria
Weights must total 100% · currently 100%
| Criterion | Weight | Scoring instruction | |
|---|---|---|---|
| Greeting and identification | 10% | Agent states name and verifies two identifiers | |
| Discovery | 20% | Asks at least two clarifying questions before proposing a fix | |
| Policy accuracy | 25% | Quotes the current policy without inventing terms | |
| Empathy | 15% | Acknowledges frustration each time it is expressed | |
| Resolution | 20% | States what happens next with a timeline | |
| Closing | 10% | Recaps the case number and thanks the customer |
Auto-fail rules
Any match scores the conversation zero
- Payment card read back aloud
- No recording disclosure on an outbound call
- Rude or dismissive language
- Promise made outside published policy
Who gets reviewed
Sampling rules per interface
- Voice
- 100% AI scored · 5% human review
- Chat
- 100% AI scored · 3% human review
- 100% AI scored · 2% human review
- Escalations
- Always human reviewed