Control Center

The Control Center is an advanced analytics screen for people who want to understand the quality of the bot's work, not just the number of conversations handled. The main question this screen answers is: "What exactly should I do to make the bot help better?".

You open it from Analytics in the side menu; the screen itself is titled Control Room. It's available to administrators and managers on the Starter plan and above.

The five metrics at the top

• Total escalations: how many conversations were handed off to your team in the selected period.
• AI resolution rate: the percentage of conversations the bot closed on its own, without involving an agent.
• AI confidence on escalations: how confident the bot was that your knowledge base had a good answer, counted only for conversations that were escalated to a person. Low values are desirable here: they mean the bot correctly hands over when the documentation has no good answer. High values suggest the escalation happened for other reasons, such as the customer's emotions or an explicit request for a person.
• Prevention rate: how effectively new knowledge base articles prevent further escalations on the same topic.
• Customer satisfaction: the share of positive ratings out of all rated conversations.

Each tile also shows the trend versus the previous period (green for up, red for down) with the percentage change.

Period filter

In the top right corner of the screen you can choose the number of days the metrics are calculated for: 7, 14, 30, or 90 days.

Charts

Weekly comparison

A bar chart juxtaposing the number of conversations resolved by the bot with the number of escalations across consecutive weeks. It shows whether the ratio is improving or getting worse.

AI confidence on escalations — breakdown

A donut chart that splits escalated conversations into three groups based on how confident the bot was that the knowledge base had a good answer:

• High: the bot had a likely answer but the conversation was still escalated. Check why: often the customer simply asked for a person.
• Medium: the bot found something related, but wasn't sure. Here it's worth checking whether the relevant articles are complete.
• Low: the knowledge base had no good answer, so handing over was the right call. Here it's usually worth expanding the knowledge base.

Escalation reasons

A table listing the reasons conversations were handed off to the team. Columns:

• Reason: a description of the category (e.g. "question outside the knowledge base", "customer asked for a human", "individual topic").
• Count: how many times this reason occurred.
• Percent: its share of the total number of escalations.
• Trend: the change versus the previous period.
• Avg. confidence: how confident the bot was when such a case occurred.

This table is the fastest way to identify what's missing in the knowledge base. If the biggest escalation reason is "question outside the knowledge base" and the topics keep repeating, adding a few articles will dramatically cut the number of conversations handled by the team.

AI insights

Below the charts you'll find AI insight cards, that is, automatic recommendations generated from an analysis of recent conversations. They may suggest, for example:

• specific topics worth writing an article about,
• low-performing articles to rewrite,
• recurring types of customer questions.

Each card shows its impact (High, Medium, or Low), the estimated reduction in escalations, and supporting evidence such as Sample questions, Missing topics, and Suggested additions. A summary above the cards shows the total potential reduction.

What's next

Once you've identified gaps in the knowledge base, go back to the Knowledge base section and add articles. After a few days, take another look at the Control Center to see how the prevention rate and confidence breakdown have changed.