Pitchbar lets visitors rate a conversation — good or bad, with an optional comment — when the satisfaction prompt fires. The CSAT panel on /app/analytics turns that raw data into a trend you can act on.

How a rating is captured

The widget shows a thumbs-up / thumbs-down prompt after a conversation reaches a natural pause (visitor closes the panel, or N seconds of silence). Submitting writes three columns on the conversations row:

• satisfaction — good or bad.
• satisfaction_at — when they submitted.
• satisfaction_comment — optional free-text up to 500 chars.

A visitor can only rate a conversation once; resubmissions are ignored.

What the panel shows

• Weekly score — 12 bars, one per week, coloured green / amber / red based on good / total × 100. Hover a bar for the exact percentage and sample size.
• Latest score — most recent non-empty week's score, plus the total rating count across the 12-week window.
• Recent low-rated — up to 10 conversations rated bad in the last 30 days. Each row deep links to the conversation transcript so you can read what went wrong.

Why the score can swing

CSAT is sample-size sensitive. A week with 4 ratings (2 good, 2 bad) shows 50%, but that's not a real trend — you have too few data points. Read the score relative to the :n ratings badge: under ~50 ratings per week, treat swings as noise.

Tenant isolation

All CSAT queries are scoped through your workspace's agents. There is no cross-workspace aggregate or benchmark — your numbers are yours alone.

Privacy

Visitor comments may contain PII (visitor's name, email if they mentioned it). The same GDPR DSR endpoints at /app/dsr erase satisfaction comments alongside the rest of the visitor's record when an erasure is processed.