CX Score
Part of AI Quality Monitoring. New to the feature? Read the Overview first.
CX Score is a single 0–10 score summarising the quality of the conversation experience.
You'll see it in Overview → Health, AI Impact, Louis, Topics, Use Cases, Sentiment → Quality Trends and Performance → Article Performance. The tooltip reads: "AI-computed experience score (effort + sentiment + resolution)".
What goes into it
The CX Score is not a mood reading. It is a weighted composite of three things:
| Weight | Component | The question it answers |
|---|---|---|
| 40% | Resolution | Was the request handled? |
| 30% | Sentiment | What did the user's own words show — politeness, frustration, explicit thanks or complaints? |
| 30% | Effort | How hard did the user have to work? One thing only: how many times did they have to repeat or rephrase because the bot didn't get it? |
Two things follow directly from those weights:
- Resolution is the single biggest driver. An unresolved conversation cannot score well, no matter how polite everyone was. This is why AI Resolution is the metric to fix first.
- A resolved conversation can still score badly. If the user had to ask four times to get there, Effort collapses and drags the score down. That is intentional — "we got there eventually, after exhausting the user" is not a good experience, and this score says so.
The Sentiment component only counts what the user actually wrote. The evaluator is explicitly forbidden from guessing at feelings that were never expressed.
How to read it
| Score | Reading |
|---|---|
| 7 and above | Healthy |
| 5 to 7 | Mixed — worth investigating |
| Below 5 | Something is wrong on this segment |
Use it as a triage tool, not a report card. Its real value is comparative: sort your topics by CX Score and the bottom of the list tells you where your users are having the worst time. That's the list to work through — see How to improve your CX Score.
CX Score vs CSAT: what's the difference?
This is the confusion to kill first. They are two different metrics measuring two different things, and they will disagree — that's not a bug, it's information.
| CX Score | CSAT | |
|---|---|---|
| Who produces it? | The AI evaluator, reading the transcript | Your users, by clicking |
| How? | Composite: 40% resolution, 30% sentiment, 30% effort | A real thumbs up / thumbs down vote |
| Scale | 0–10 | 1–5 |
| Coverage | Every ended conversation | Only conversations where someone voted — a small minority |
| What it captures | Objective quality of the interaction | The user's declared opinion |
| Bias | The evaluator's judgement — imperfect, but applied identically to everyone | Whoever bothers to vote. In practice: the very happy and the very angry |
The CX Score is not "AI-generated CSAT"
They are not versions of the same thing. CSAT is an opinion poll with a low, self-selected response rate. CX Score is a quality audit with full coverage. You need both: CSAT tells you what users say, CX Score tells you what happened.
When they disagree, that disagreement is a lead. High CSAT with a low CX Score usually means the few people who vote are your happy ones. Low CSAT with a decent CX Score often means the bot works and something else — a policy, a price, an outage — is making people angry.
You'll find real CSAT and NPS votes, plus Dissatisfaction Drivers (sorted by Volume × (10 − CX)), on the Sentiment report.
Updated about 3 hours ago