Insight
Insight helps you find weak points in customer conversations: which indicators are getting worse, which agents need adjustments, and which topics keep coming up. Start with the summary instead of reading every conversation, then examine the metrics, agents and dialogs that need attention.
In the sidebar's Analytics group, open Insights. The section page brings together the AI summary, metric comparisons and individual evaluations.
How it differs from Statistics
Insight and the AI summary help interpret quality: they highlight problem metrics, agents requiring attention, trends, anomalies, metric extraction history and explanations of changes.
Statistics answers questions such as "how many messages and leads", "what credits were spent on" and "how workload is changing". Insight focuses on quality: "which agent needs attention", "what changed in evaluations" and "what customers ask about".
Starting an analysis
- Read the AI summary for conversation topics and improvement suggestions. Its analysis period is shown at the bottom.
- Set the period and filters for metric comparisons.
- Choose a comparison method using the evaluation tabs.
- Open a metric or agent from the results to examine it in detail.
A completed summary does not change when you switch filters. To get conclusions for a different period, refresh it separately.
Filters
In the filter bar, choose:
- The period you want to examine.
- The agent type, such as support or sales. The comparison will include agents of that type.
- The metric type: built-in, agent-type metrics or custom project metrics.
Agent and metric types affect the metric and agent lists on comparison tabs. History uses the period, search and sorting. Type filters do not change the AI summary or automation recommendations.
Which metrics are compared
Rankings, trends and individual indicator pages can include both event metrics and additional discussion attributes with a meaningful good or bad direction. For example, urgency and issue severity can be compared across agents and periods alongside other indicators.
The main discussion category is not part of this comparison: it groups topics in the AI summary. Look for topic lists in the summary and changes to evaluative attributes on the Insight tabs.
The number of observations is not necessarily the number of unique dialogs. Event metrics count message evaluations; additional discussion attributes count identified topic fragments. A dialog can contain several observations.
Tabs
Choose a tab based on the question you want to answer:
- Anomalies: which agents or metrics stand out from the others;
- Trends: what changed compared with the previous period;
- Normatives: which values deviate from the metric's norm;
- History: which individual results make up the indicators.
Metric details
In Problem metrics, click a metric card. The metric page compares individual agents' values, their baseline and deviations from it. The period, mode and selected types are preserved when you navigate.
Read the percentage sign together with the metric's direction. For example, lower answer quality is worse, while a higher error rate is worse. The deviation in details refers to the selected mode's baseline; "Vs. previous period" on the overview card separately shows change over time.
The colored score helps you spot the extent of a deviation. Its percentage is not the share of bad dialogs or the accuracy of the agent's answers.
Click an agent in the list to move from one metric to all of that agent's indicators.
Agent details
In Agents requiring attention, click an agent card. The agent page shows its indicators and metrics ordered by deviation score.
Available actions on this page:
- Adjust the period if needed.
- Compare metric values with their baseline and open the required metric to see other agents' results.
- Click the agent name or avatar to open its settings.
- The back to Insight button returns to the lists with the selected conditions.
The list shows up to 50 metrics; when there are more, the loaded count appears next to the total. Opening settings requires permission to edit agents.
Automations requiring attention
The Automations requiring attention block uses the latest AI summary. Its counter shows how many automations have issues out of the published automations included in the analysis.
Each automation recommendation contains its name, priority, identified issues and suggested changes. Click Open workflow to go to the editor. If a recommendation refers to a particular step, the editor opens and highlights it.
If there are no recommendations, the block explains why: no issues were found, there are not enough observations, there are no published automations, or the data could not be retrieved. To check again, refresh the AI summary.
When Insight is empty
"No data" and "No problems" mean different things. The first means no suitable observations are available; the second means the comparison did not identify undesirable deviations.
Open the metric history and check for evaluations in the required period. If there are none, check agent metrics and publication of its settings. Anomalies need at least two agents sharing a metric.