Insight

In the side menu, open Insight. This opens the main Insight page. Insight shows how agents and conversations are performing based on metrics.

Difference From Statistics

Insight and the AI summary are for interpreting quality: they show problematic metrics, agents that require attention, trends, anomalies, metric extraction history, and explanations of changes.

Statistics is for the project's numeric operational picture: credit and token spend, the full cost of AI responses and related tools, messages, dialogs, leads, subscriptions and unsubscriptions, time-based activity, channels, and agents. If the question is "how many", "what made up the spend", "where did costs grow", or "how did the lead base change", use "Statistics". If the question is "why is this metric bad" or "which agent needs attention", use "Insight".

Main elements

the AI summary explains important changes and topics;
the filter panel narrows data by period, agent type, and metric type;
evaluation mode tabs switch between anomalies, trends, normatives, and history;
problem metrics show indicators that need attention;
agents requiring attention help find a problematic configuration;
the metric history shows individual calculation results.

Periods

The period filter offers:

last 24 hours;
7 days;
30 days;
the entire period.

Agent types

The agent type filter offers:

all types;
Sales;
Support;
Notifications;
Assistant;
Education;
Surveys;
General.

Types of metrics

The metric type filter includes:

all types;
built-in;
by agent type;
custom.

Which metrics are compared

Rankings, trends, and individual metric pages can use both event metrics and additional discussion attributes that have a meaningful good and problematic direction. For example, urgency and issue severity are compared across agents and periods together with other indicators.

A primary discussion category is not included in this comparison: it groups topics in the AI summary. Use the summary for the topic list and the Insight tabs for changes in evaluative attributes.

The count shown next to a metric is the number of its observations, not necessarily the number of unique dialogs. For an event metric, these are message evaluation results; for an additional discussion attribute, they are detected topic fragments. One dialog can contain several observations.

Tabs

Anomalies - unusual changes in values;
Trends - the direction of change;
Normatives - comparison with target values;
History - individual metric results.

Problem metrics

the problem metrics block shows where a value is bad or requires attention;
displays the number of problematic ones relative to the total;
selecting one opens the detailed metric page.

Agents requiring attention

the block shows agents with problematic performance;
selecting one opens the Insight page for that agent.

If Insight is empty

check whether the agent’s metrics collection is enabled;
check whether the agent version with metrics is published;
check whether there were dialogs in the selected period;
check the filters;
check that the metrics are included in the analytics;
check the history of metrics extraction.