The numbers that change decisions.

Responsiveness

Average first-response and handling time, computed against your working hours — a message answered 40 minutes after opening hours counts as 40 minutes, not 14 overnight hours. Period-over-period volume shows whether chat is growing with the business.

Load & people

• Per-agent stats — who handled what, how fast, with what outcomes. Coaching input, not surveillance.
• Visitor chat heatmap — day × hour grid showing when visitors actually talk. Schedule shifts and automations against reality, not habit.

Automation quality

• AI deflection rate — the share of visitor questions resolved by grounded AI answers without a human. The single number that tells you whether the knowledge base investment is paying.
• Tracked event summary — your product's events (via the event API) with visitor and company counts.
• CSAT & NPS trends — satisfaction over time, with detractor follow-through visible. See CSAT & NPS.

Reading them together

• Deflection up + CSAT flat = automation working. Deflection up + CSAT down = the AI is answering questions it shouldn't — tighten the knowledge base.
• Heatmap peak with slow first response = staffing gap, not a process problem.
• Volume up + handling time flat = the team is scaling well; consider raising the auto-reply threshold.

Exports

Conversation data exports to CSV/JSON for anything deeper — your warehouse, your BI tool, your quarterly review deck.