Agent analytics

Filtering

An agent's Analytics tab is filtered by channel instance (not just channel type — if you have two Telegram bots on the same agent, you can pick between them), a date range (7/30/90 days, all-time, or a custom range), and conversation status (open, escalated, human, resolved). Filters are reflected in the URL, so you can bookmark or share a specific view. Use Export CSV to pull the filtered conversation data out.

KPIs

• Customers and Conversations in the selected range.
• Grounded answers % (and the inverse fallback %) — how often the agent answered from retrieved knowledge vs. fell back.
• Average response latency.
• A by-channel breakdown table.

Charts

Sessions per day by channel, token spend per day (prompt + completion), requests per day, and conversations per channel.

Outcome metrics — not live yet at this level

CSAT, resolution time, resolution rate, and escalation-to-human rate are shown as placeholders on this page ("coming soon") — they aren't instrumented per-agent yet. Resolution rate and escalation rate are live at the organization level instead — see Organization analytics.

History and customer memory

Further down the page, History shows per-customer conversation transcripts with citations, and Customer memory shows the long-term facts the agent has learned about a given customer — admins can erase a customer's memory from here if needed.