What’s measured
A 1-5 score on every conversation that gets a customer rating:- 5 — great, fully helped
- 4 — helpful
- 3 — neutral
- 2 — poor
- 1 — bad experience
Headline numbers
The top of the page shows three cards:| Metric | What it tells you |
|---|---|
| Average score | Mean of every rated conversation in the window (out of 5) |
| Response rate | Share of resolved conversations that got rated |
| Total ratings | Count of ratings — small denominators mean noisy averages |
Score distribution
A bar chart from 1 to 5. A healthy support team is heavy on 4s and 5s; a thick lower tail is where to look. If your average is 4.3 but you have a stubborn 8% on 1-2, the average is hiding pain.Trend over time
Daily average across the window. Watch for:- Sudden drops — usually correspond to an outage, a policy change, or a bad release.
- Slow declines — knowledge is going stale or expectations are creeping up. Refresh top knowledge sources and check whether your AI is over-promising.
Breakdowns
Switch the chart to break by:- Channel — which surface drives the worst experience.
- AI vs human — are human replies rated higher than AI? If yes by a lot, look at where the AI replies and whether those topics belong to it.
- Agent (assignee) — useful for coaching individual teammates.
- AI agent — when running multiple agents, see which one rates best.
- Tag — usually the most actionable. Low CSAT on
returnstells you the policy bites; low CSAT onshippingtells you the carrier is the problem.
Comments
Negative ratings often come with a free-text comment. The page lists the most recent 1-2 ratings with comments at the bottom; click through to read the conversation in context. Read every 1 and 2. They’re a free customer-research stream.Why the response rate matters
A 4.7 average from 4% of customers is much weaker evidence than 4.3 from 35%. To increase response rate:- Send the survey on resolution by a human, not on every auto-close.
- Skip the survey for spam, “no action needed”, and duplicates (Resolving).
- Phrase the resolve message warmly — “Glad we could help — how was that?” gets more responses than the survey alone.
Configure CSAT
CSAT is enabled per flow’s Resolve node, or globally via support if you want every resolved conversation to ask. See Resolving conversations for the per-conversation behaviour.Plan availability
CSAT collection is available on Growth and above. The CSAT analytics page is available on every plan, but if no CSAT is configured the page shows an empty state with setup instructions.Related
Analytics overview
Volume, response time, AI deflection.
Resolving conversations
Where CSAT actually fires from.