> ## Documentation Index
> Fetch the complete documentation index at: https://docs.axottle.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Analytics and evaluations

> Review dashboard, queue and time metrics, honest report on operators, customer evaluations after closing and mood analysis.

Support has two analytical screens. The review (main page of the section) answers the question "what's happening right now": active tickets, queue without operator, delays, share of AI responses, the table "Require attention" with the nearest SLA, readiness of AI and the state of channels. The section "Analytics" looks back at the selected window (from 7 to 90 days) and shows how support worked.

## What analytics show

Top row: created, active, resolved, rediscovered, expired by SLA, no operator. Below are four ring indicators: solved percentage, SLA compliance, AI response rate (or rediscovery rate when AI is turned off) and average customer score.

The "Time to Answer and Decision" card gives the percentiles p50 and p90 in human terms: "usually" and "longest." The "Where the Wait Goes" card plots the ticket time to wait for the client, the operator and the internal actions: you can see you lose time with the team or the client. "The quality and sustainability of the decisions" reduces the created and solved with rediscoveries and average evaluation.

When you turn on the AI, you separately show the number of auto-responses followed by reopening, and the metric's signature is honest: it's a proxy signal, not a proof of model error; the module only knows that the ticket hasn't been resolved.

## Report on operators

Table for each operator: processed, reopened, estimation, late SLAs, ticket messages, and when the AI is turned on, still sent and rejected prompts. The naked counter of closed tickets is not intentionally in the report: the number without qualifications encourages closing tickets, not solving problems. If you build a team rating, keep reopenings and ratings in the same line.

## Customer evaluations

After the ticket is closed, the client receives a message on Telegram “Assess the quality of service” with the buttons of stars from 1 to 5. The click saves the rating, the message is replaced by gratitude, the buttons disappear. Only the ticket client can evaluate the ticket and only while the ticket is closed.

Reopening resets the score until the next close: the score is always related to the last decision. The function in Defult Messages is turned off: the Evaluation Request template switch controls both the sending of buttons and the collection of ratings.

The score is visible in the cap of the closed ticket, in the average score on the installation, in the report on the operators and on the team cards.

## The mood.

A separate Mood page (in the Management subgroup) shows a slice of the emotional background: how many dialogues are analyzed, how much attention is required, the state distribution. Six states: calm, grateful, confused, irritable, angry and uncertain. "Uncertain" is the normal response of the model, not a mistake.

The mood score is probabilistic, so each one goes with confidence, and the panel shows it: an "evil" score with a certainty of 0.4 should not be believed as a fact. The risk of leaving is considered a separate signal: the calm client can leave, and the irritated one often stays.

The mood works independently of the AI mode and is able to escalate the ticket to a person: the list of moods for escalation and minimal confidence are set in the settings, by default, the “evil” escalates with a confidence of 0.7. Nothing irreversible in relation to the client mood does not start.

## Knowledge gaps

At the bottom of the analytics, there's a list of things that the knowledge base didn't find an answer for. This is the work line for the AI administrator: you can mark the topic as closed. [Knowledge base](https://docs.axottle.com/en/support/knowledge.md#пробелы-знаний).

> [!NOTE]
>
> Mood analytics, complaint observers, and knowledge gaps also work when dialog-based AI is turned off: assigned role models are sufficient for them.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.