Sentiment analysis
How Chatasist judges the mood of each conversation and what it was about, and how to read the sentiment dashboard.
Every finished conversation is labelled twice: how the customer felt about it, and what they got in touch about. Together those answer the question a support lead actually has each morning — did anyone leave unhappy yesterday, and what about?
Find it under AI & Automation → Sentiment.
The two labels#
Sentiment — how the customer felt
One of unhappy, neutral or satisfied, judged on the conversation as a whole rather than message by message. Somebody who arrives angry and leaves thanking you had a good conversation, and is labelled satisfied.
Topic — what it was about
One of eight: information, order status, delivery, payment, returns, complaint, technical problem, or other. A fixed list, so the same thing can be counted week after week.
When the conversation is too short or too ambiguous to judge, the sentiment is neutral — never a guess. Every label here is counted in a total and read as fact, so a cautious label is worth more than a confident wrong one.
Each label comes with one sentence saying what in the conversation led to it. Hover the sentiment in the chat attributes panel to read it. If a label ever looks wrong, that sentence is how you check.
Reading the dashboard#
The headline is the share of conversations that left customers unhappy, shown against the previous period of the same length. On its own, "12% unhappy" is not something you can act on; next to last month's 4% it is.
The topic breakdown is the part worth opening daily. Each topic is split by how customers felt, so you can see which subject is producing the unhappy conversations rather than just that they exist. Select a topic's unhappy percentage to read those conversations.
Switch between Daily and Monthly. Daily tells you whether yesterday's spike is real; monthly tells you whether the desk is getting better over a quarter. Both show quiet periods as empty rather than skipping them, so the chart never compresses time into a trend that is not there.
The channel breakdown is where a healthy overall figure often turns out to be one channel doing well and another doing badly.
In the inbox#
Conversations that left a customer unhappy carry a label in the conversation list, and there is an Unhappy view beside Unread and Starred. Opening a conversation shows its sentiment and topic in the attributes panel on the right.
This is the practical half of the feature: the dashboard tells you there is a problem, the Unhappy view is the list of people you can still go back to.
When labels appear#
Conversations are labelled after they finish, on an hourly pass — so today's are still filling in, and yesterday's are complete. A conversation that carries on after being labelled is judged again, which matters: a complaint that gets resolved afterwards stops being an unhappy conversation and leaves the list, instead of sitting there forever.
Sentiment analysis is available on the Growth plan and above.