AI Agent analytics

• Conversation trends, message totals, leads, countries, devices, domains, and top pages.
• Answer quality signals such as knowledge answers, fallback answers, blocked answers, and greeting turns.
• Response time, AI token metadata, estimated cost, and whether a real AI call happened.
• Training health, failed sources, top unanswered questions, contacts, and visitor context.

Answer kinds include knowledge, fallback, greeting, and blocked. Fallback heavy behavior usually means the training set needs better coverage.

• Use unanswered questions to decide which website pages, manual Q&A, or files need stronger coverage.
• Review source references in transcripts to understand which approved content shaped an answer.
• Use knowledge gaps and generated FAQs to prioritize improvements.
• Connect lead analytics with conversation context before changing forms or follow up rules.

Visitor analytics use coarse context such as country, region, city, device, browser, OS, language, timezone, screen, viewport, referrer, and page URL.

Analytics are available with the product. Token and cost tracking are reporting signals; what counts against your plan is the number of AI replies your AI Agents send.