Feedback & Analytics is where ratings, flags, CSAT, and conversation scores roll up. Trends drive the next round of training; exports drive QA and compliance reporting.
In this section
- Capture feedback from three sources: end users, operators, and automated scoring.
- Rate any agent turn with thumbs up/down plus a structured reason.
- Export raw feedback for spreadsheet analysis or a BI pipeline.
- Close the loop by turning negative feedback into documents, instructions, or simulations.
What's inside
- 01
End-user feedback
Thumbs up/down and optional comments collected inside Live Chat and in email footers.
- 02
Operator feedback
Ratings and structured reasons an operator adds while reviewing a thread in Copilot.
- 03
Automated scoring
Conversation-level scores produced by the Scoring Skill — engagement, qualification, or a custom rubric.
- 04
Training loop
Flagged turns queue up as training candidates: add a document, edit an instruction, or write a simulation, then re-run and verify.
At a glance
| Metric | What it tells you | Act on it by |
|---|---|---|
| Thumbs-down rate | Share of agent turns end users rejected. | Reviewing the flagged turns in Admin Mode |
| Reason breakdown | Why answers were rejected — wrong, incomplete, off-tone, or slow. | Choosing between data, instruction, or skill fixes |
| Escalation rate | Share of conversations that needed a human. | Closing the top escalation reasons first |
| Resolution rate | Conversations closed without escalation. | Tracking agent maturity over time |
| CSAT | Post-conversation satisfaction where the survey is enabled. | Reporting to stakeholders |
| Conversation score | Scoring Skill output per thread. | Routing hot leads or at-risk accounts |
Cadence
Review feedback weekly during the first month of an agent's life, then monthly. Every review should end with a concrete change and a simulation that proves it.