Run your workspace

Analytics

The analytics dashboard is where you measure whether the agent is actually working: conversation volume, deflection, lead conversion, response latency, and the knowledge gaps that are quietly costing you answers.

Headline metrics

• Conversations — distinct chat sessions started by visitors.
• Messages — total visitor turns, with average depth so you can see how engaged sessions actually are.
• Deflection rate — conversations the AI handled end-to-end (no human takeover, no low-confidence flag) divided by total. This is the headline KPI for ROI.
• Leads captured — completed lead forms, with the conversion percentage against engaged conversations.
• Response time — median latency between the visitor's message and the first agent token.

Gap detection

Every time the agent answers below its confidence threshold, the system stores the question. Semantically similar questions are clustered, and the resulting gaps report ranks them by frequency. Each entry shows a representative question, sample conversations, and a one-click action to add a source that fills the gap — so the next visitor asking it gets a real answer.

Filters and exports

• Per-agent filtering — compare metrics across agents when you run more than one.
• Time range — 24 hours, 7 days, 30 days, or a custom window.
• CSV export — download the filtered data for spreadsheets and BI tools.

Citation tracking

The dashboard surfaces which knowledge sources the agent cites most often and — just as useful — which ones are never used. Unused sources are often outdated, mistitled, or duplicated; the citation panel is the fastest way to spot them.

A/B experiment monitoring

When you run an A/B test on a behavior rule, the conversion lift between variants shows up here directly so you can call a winner without exporting anything.