Sentiment analysis
Turn on sentiment analysis to score every customer message from -5 to +5, see each chat's net sentiment, sentiment shift and lowest point, and get recommendations.
Sentiment analysis measures how customers felt during a chat. Robylon scores every customer message for sentiment, then summarises the chat: how positive it was overall, whether the customer got happier or more frustrated, the worst moment, and a few recommendations for what to improve.
Use it to find unhappy customers that CSAT misses, because most customers never fill in a survey.
Turn it on from Settings → AI Intelligence → AI Behavior, under Conversation Insights → Sentiment Analysis. It sits on the same page as the other agent assist features.
Sentiment analysis availability
Availability depends on your plan and usage. If it is not available, or no chats remain for analysis, the toggle says so and new chats are not analysed.
For plan questions, contact your account manager or support@robylon.ai.
Choose which parts to show
When Sentiment Analysis is on, four sub-toggles appear under it. Each one controls whether that part of the analysis is shown:
| Sub-toggle | What it shows |
|---|---|
| Net Sentiment Score (NSS) | How positive the chat was overall. |
| Sentiment Shift Score (SSS) | Whether sentiment rose or fell during the chat. |
| Lowest Sentiment Point | The most negative customer message and its emotion. |
| Actionable Recommendations | Suggested next steps drawn from the chat. |
For how each score is calculated, see what the chat-level scores mean.
How each message is scored
Every message the customer sends gets a whole-number score from -5 to +5. The AI’s and teammates’ replies are read for context but not scored, and button clicks are ignored.
| Score | Emotion | What it looks like |
|---|---|---|
| +5 | Ecstatic | Delight or strong gratitude |
| +4 | Delighted | Very satisfied |
| +3 | Pleased | Happy and content |
| +2 | Satisfied | Mild approval, things are on track |
| +1 | Reassured | Calmer after a concern |
| 0 | Neutral | No strong emotion |
| -1 | Uncertain | Mild doubt or confusion |
| -2 | Concerned | Worry or disappointment |
| -3 | Frustrated | Clear irritation |
| -4 | Angry | Strong negative feeling |
| -5 | Furious | Outrage |
Chat-level scores
| Metric | How it is calculated | How to read it |
|---|---|---|
| Net Sentiment Score (NSS) | Positive messages minus negative messages, divided by all customer messages, times 100. It runs from -100 to +100. | Above 0: more positive than negative messages. Below 0: the customer was mostly unhappy. |
| Sentiment Shift Score (SSS) | The average score of the last 30% of the customer’s messages minus the average of the first 30%. | Positive: the customer ended happier than they started. Negative: the chat made things worse. |
| Lowest sentiment point | The customer’s lowest-scored message, with its text, time and emotion. | The moment to read first when reviewing the chat. |
| Actionable recommendations | Four or five one-line suggestions based on the chat. | Ideas for fixing the agent, a workflow or a policy. |
Each chat is also compared with the average of chats closed the same day, shown as higher or lower.
When insights are available
The day after the chat ends. Robylon analyses chats once a day, covering chats that ended the previous day. Insights are not calculated while a chat is still open.
If sentiment analysis is on but no insights appear the day after chats end, email support@robylon.ai.
Find unhappy customers
Filter conversations by sentiment in Transcripts or the inbox:
- Chat Net Sentiment Score: below 0, below 50, above 0 or above 50.
- Chat Sentiment Shift Score: below -5, below 0, above 0 or above 5.
- Chat Insights Available: only chats that have been analysed.
For example, filter by an NSS below 0 to review the most negative chats of the week, or by an SSS below 0 to find chats where the agent made things worse.
Act on the results
- Review the lowest sentiment point first. It usually shows the exact reply or policy that upset the customer.
- Look for repeats. If many low points follow the same kind of answer, fix that answer with Click to Improve or the agent’s instructions.
- Check handovers. A negative shift right before a handover often means the agent held on too long.
- Track it over time. After a change, compare the share of chats with a negative NSS before and after.