Low CSAT Analysis: Filter bad conversations by failure type
A CSAT of 1 or 2 is a signal, not a diagnosis. Was the AI's answer off-topic? Did the customer reject the handoff? Was the refund too small? Or did the situation break down before the conversation even started, a delayed delivery or wrong order that had nothing to do with how the AI responded? Without that distinction, every failed conversation looks the same, and you end up iterating on the wrong thing.
Low CSAT Analysis brings AI-powered failure categorization into the Conversations tab, giving support managers and AI Agent admins a structured way to find, filter, and understand dissatisfied conversations.
Find the right conversations
Filter by Low CSAT to surface only CSAT 1–2 conversations. Stack it with the AI Confidence filter, which now operates at the conversation level rather than the message level, to surface conversations where the AI both underperformed and expressed uncertainty. That combination gives you the shortest path to the conversations that most need attention.
Understand why they failed
Each conversation is automatically tagged with one of five failure categories:
- Inadequate Response: The AI answered the wrong question, or gave a vague non-answer
- Inadequate Compensation: The customer pushed back on a refund or coupon offer
- Escalation Refusal: The customer declined a handoff to a human agent
- Situation Frustration: The customer was upset about something external. The AI did its job. This is not an agent problem.
- Others: Does not fit the above
Click into any flagged conversation and the Insights panel shows the category and a plain-text reason behind it, ordered by impact.
Track it over time
The trend chart updates to reflect whichever filters are active. Watch whether a spike in Inadequate Response failures is a one-week anomaly or a pattern that points to a gap in your agent configuration.
