Skip to main content

Documentation Index

Fetch the complete documentation index at: https://setup.cevro.ai/llms.txt

Use this file to discover all available pages before exploring further.

The Feedback page in Insights answers a different question from Drivers.
  • Drivers = “What did players contact us about?” (product / process problems, classified per ticket)
  • Feedback = “What did players say about us?” (CSAT ratings, post-chat comments, sentiment summaries)
These are different stories. A player can have a real product issue (Drivers) and love your agent (Feedback). Insights keeps them separated. URL: https://app.cevro.ai/insights/feedback
This is the Insights → Feedback page. Not to be confused with the per-message thumbs up/down feedback inside conversation logs — that’s a separate tool for telling Cevro how to improve your AI agent.

​
What You’ll See

​
CSAT distribution

A histogram of CSAT ratings (1★ → 5★) for the selected period. Colored red (1★) → green (5★). Lets you see at a glance whether your scores are bimodal (love-it-or-hate-it) or skewed.

​
Negative sentiment by topic

Top 10 Themes/Needs/Problems ranked by share of negative-sentiment conversations. Only groups with enough volume are included so a single bad sample doesn’t dominate.

​
Top negative feedback

A clustered list of negative comments left by players. Each row shows:
  • A representative comment (the cluster’s most central member)
  • The matching Theme/Need
  • The conversation count for the cluster
  • A +N similar collapsible reveal showing other comments in the same cluster
Clustering is meaning-based, not keyword-based — “no me llegó el retiro” and “withdrawal still pending” land in the same cluster even though they share zero keywords.

​
Top positive feedback

The same clustering applied to praise. Use it for:
  • Internal recognition — surface the agents and brands players love
  • Marketing — pull authentic quotes for testimonials and case studies
  • Reinforcement — feed examples back to your AI training so the agent doubles down on what works

​
Agent CSAT by topic

A per-agent breakdown table with:
ColumnDescription
AgentThe agent the conversation was assigned to
TicketsRated conversation count
Avg CSATAverage rating, colored by band (≥4 green, ≥3.5 amber, otherwise red)
Neg %Share of negative-sentiment conversations
Top issueThe Need/Problem the agent over-indexes on
Useful for spotting “agent X has high CSAT but the Top issue is waiting for response” mismatches — a coaching signal.

​
Filters

The Feedback page shares the same filter set as Drivers:
  • Brand + Date range at the top of the page
  • Advanced filters for Signals, Contact drivers, Status, Agent, Browser, Platform, Language, Channel, Polarity, State, Contact type, CSAT Rating, Scorecard, Page, and Player metadata

​
Where the Data Comes From

Feedback ingests from:
  • Post-chat surveys — the rating + free-text comment players leave at end of chat
  • External CSAT — surveys delivered through your support desk (Zendesk, LiveChat, Intercom) and synced into Cevro
  • Conversation sentiment — derived from message tone when an explicit rating isn’t available

​
Tips

  • Read negative clusters first. They’re where most of the actionable signal lives. Five recurring complaints are a roadmap.
  • Pair Avg CSAT with Neg %. An agent who scores 4.2 but has 25% negative sentiment is probably resolving complaints that should never have happened — investigate.
  • Cross-reference with Drivers. A high-volume Theme in Drivers paired with a high-CSAT cluster in Feedback tells a different story than either alone (service is patching a product issue).
  • Drivers — product/process VoC (the “what did they contact us about” view)
  • Per-message Feedback — thumbs and reviews you leave on AI conversations to improve your agent
  • QA Scoring — systematic quality evaluation feeding into Sentinel
  • Taxonomy — the catalog that maps clusters back to Themes / Needs