Skip to content
AISARAISAR
All articles

AI conversation review: customer tone, spam and the outcome of an inquiry

AI reads your conversations, detects the customer's tone and flags spam, and for each inquiry suggests whether the issue was solved and whether the customer is satisfied. Here is what is assessed, where you see it, how to read "Not enough data", and how to switch the feature off.

What you need to know

  • The feature is available on the Business plan and above. On simpler plans it is not shown at all.
  • The assessment is made by AI, and AI can be wrong. Treat the marks as a hint about where to look first, not as a verdict on a customer or an employee.
  • The assessments trigger nothing on their own: they do not change a conversation's status, reassign it, send messages or affect automations. They are information for people only.
  • Personal conversations are assessed. Group chats are not.

What AI assesses

IndicatorWhat it meansHow it is shown
Customer toneThe customer's mood in the conversation: positive, neutral, dissatisfied or irritated.In the conversation list only "Customer dissatisfied" and "Customer irritated" are marked, so positive and neutral tones stay quiet. The conversation panel shows any tone.
SpamThe conversation looks like a mass mailing or an unsolicited offer rather than a customer inquiry.A "Spam" mark in the list and the panel. Spam inquiries are not assessed further.
Issue solvedWhether the customer got what they came for by the end of the inquiry.In the conversation panel, under "Latest inquiry".
Customer satisfiedWhether the customer was happy with how they were served. "No" is set only when the customer complains, is disappointed, irritated or reproaches the company. A plain businesslike chat without thanks ("how much is it?", "I'll come tomorrow") is not counted as dissatisfaction.Same place, under "Latest inquiry".
"Customer satisfied" from AI is not the same as a CSAT score. CSAT is the customer's real answer to your question after a conversation closes; here AI draws a conclusion from the text of the chat. They live separately and do not replace one another.

What an inquiry is and when results appear

A single conversation with a customer can run for months and cover a dozen different questions. So AI does not assess the whole conversation at once, it assesses inquiries: an inquiry is a part of the conversation about one question. It ends when the customer is silent for 24 hours or when the conversation is closed. If the customer writes again, a new inquiry begins.

  • AI assesses tone and spam about 30 minutes after the customer's last message. A nightly run picks up anything missed during the day.
  • The outcome of an inquiry ("issue solved", "customer satisfied") appears after 24 hours of customer silence, counted from the last message.
  • While an inquiry is still open, the panel says the outcome will appear later. If an operator closed the conversation, its outcome usually arrives sooner and the mark in the open panel updates by itself.
  • When the feature is first switched on, only active conversations are assessed, meaning those with messages in the last 7 days. Older conversations are not reviewed retroactively.

Where to see it

  • Conversation list. A small tone mark for dissatisfied and irritated customers and a "Spam" mark. Hovering shows the date of the assessment.
  • The conversation panel on the right. An "AI conversation review" section: tone, spam, a "Latest inquiry" block with "issue solved" and "customer satisfied", an expandable history of earlier inquiries and an "AI review from" caption with the date. A "How it works" hint sits next to it.
  • Chat filters. An "AI review" group: customer tone (several can be selected), spam (show only or hide) and latest inquiry (not solved or customer dissatisfied). Filter counters take these conditions into account. There is no language filter.
  • Dashboard. A small "AI conversation review" card for the chosen period: the share of satisfied customers, the number of irritated customers and unsolved inquiries. The irritated and unsolved numbers lead straight to the chat with the matching filter. The card is visible to the owner and manager; operators do not see it.

The dashboard card appears once enough inquiries have ended in the period for the statistics to mean something. Until then it shows "Not much data yet". The numbers carry the note "AI estimate, can be wrong".

What "Not enough data" means

AI returns an answer together with a confidence level. When confidence is low we do not guess or show a doubtful verdict, we write "Not enough data". This is a normal outcome: for example, when an inquiry is only a couple of phrases or the customer answered in a word. The note does not mean the conversation is good or bad, only that it is too early to conclude.

How to switch it off and on

  1. Open Settings → CRM → Chat, the "Rules" tab, the "AI conversation review" section. The owner or an administrator of the company can change the switch.
  2. Flip it. When switching off, a confirmation appears explaining what will change.
  3. Confirm and press "Save". Nothing changes until you save, and "Discard changes" restores the previous state.

What happens when you switch it off:

  • New conversations and messages are no longer assessed.
  • Results already obtained are hidden, not deleted: the marks disappear from the list, filters and dashboard.
  • If you switch it on again, the earlier results come back and new ones appear as customers write.
  • Messages written while the feature was off are never sent for assessment, even after you switch it back on. Only messages that appear after switching on are assessed.

Privacy

  • Before analysis, personal data is removed from the text: phone numbers, card numbers and similar details are replaced with masks.
  • Customer conversations are not used to train models.
  • The results (marks and yes/no answers) are stored in AISAR and visible only to your team.

More on how we handle data is in the Privacy Policy.

Common questions

SituationWhat is going on
Switched on, but there are no marks.Tone and spam appear about 30 minutes after the customer's last message, and the outcome after a day of silence. If there is no assessment yet, give it a little time.
An old conversation has no review.On start only conversations with messages in the last 7 days are assessed.
The panel says "Not enough data".AI is not confident, so no verdict is shown. Treat it as a neutral outcome.
No card on the dashboard.Only the owner and manager see it, and only when the feature is on and the period has enough ended inquiries.
The feature is missing from settings.It is part of the Business plan and above. Plan contents are in the "Subscription" section.
AI got it wrong.It can happen. Rely on the conversation itself and use the review as a hint.

See also

The customer's real answers after a conversation closes are in Quality score. Closing conversations manually and automatically is in Chat basics and Auto-closing conversations. Roles and permissions are in Users and roles.