/Docs
Start free

Analytics

What happened in support, by the numbers: volume, response times, resolution, CSAT, AI involvement, topics, tickets, visitors and contacts. Every chart drills through to the conversations it counts.

Overview

Analytics lives in the sidebar with nine pages under it. All of them share one shell: the date range, the filter chips and the export button sit at the top and apply to whatever page you are on. The range and filters are kept in the URL, so a view survives navigation and a link to it can be pasted to a colleague. Analytics is in the Essentials plan and above; scheduled reports and the advanced pages are in Plus. See Plans and billing.

The nine pages

Performance
The front door: CX score, automation rate, involvement rate, resolution rate, first and human response times, time to handle, against their targets, with the funnel from conversation to resolution.
Messages
Conversations and messages over time, by channel, with a weekday and hour heatmap of when they arrive.
Response time
First response and human response, as a series and a distribution, per weekday and hour.
Operators
Each teammate's conversations, replies, response time and ratings over the range.
Tickets
Tickets raised and solved, time to solve, and where they sit by state and priority.
AI
What the AI agent handled: conversations it was involved in, resolved alone, handed off, and how visitors rated its answers.
Topics
What customers write about, discovered from the opening messages and named automatically, with volume per topic and the ratings behind each.
Visitors
Where widget visitors come from and the pages they chat from.
Contacts
How the contact list grows, and by which channel.

Range and filters

  • The range picker offers presets (today, 7 days, 30 days, this month, and so on) and a custom span up to 366 days. Series are bucketed in your own timezone, so the first and last day of a range hold the right conversations.
  • Filter chips narrow every conversation-shaped number at once: channel, status, assignee, rating, topic, AI involved. A chip means the same thing on every page.
  • Click any bar, slice or row to open the inbox filtered to exactly the conversations it counted.
  • A figure with nothing behind it shows as a dash, never as zero: no ratings in the range is not 0% satisfaction.

Targets

Under Settings, then Analytics, give each headline metric a target: automation rate, involvement rate, resolution rate, CX score (percentages), first response, human response, time to handle and ticket time to solve (minutes). The Performance page then shows a pass or fail beside each figure, and the emailed report says which targets were missed. Without a target a number can only be compared with last period.

Scheduled reports

  1. 1
    Settings, then Analytics, then New report.
  2. 2
    Name it, add the recipients (any email addresses, comma separated), and pick the pages it should summarise.
  3. 3
    Choose the frequency: daily, weekly or monthly. The list shows when each report next runs.
  4. 4
    Save. The email carries the headline figures for the period, the targets hit and missed, and a link into analytics for the same range.
ℹ
Reports are sent by the scheduler, which checks every fifteen minutes for reports that are due, so a weekly report lands within the quarter hour after its time.

Export and API

The Export button at the top of any page writes what that page shows, for the current range and filters, to a CSV. For a warehouse or a dashboard of your own, the public API returns the performance summary for a date range with an API key that has the analytics scope.

How the numbers are made

  • AI involvement counts every conversation the agent replied in, including those later taken over by a person; the handoff is what the involvement rate is for.
  • CX score and CSAT come from the rating cards visitors answer: after a chat is closed, after the AI resolves one, or when an automation asks.
  • Topics are discovered, not configured: opening messages are grouped by meaning and each group is named once. Groups of fewer than three are left out as noise. Classification runs on the scheduler, so a new conversation joins its topic within the hour.
  • Daily totals for long ranges come from a nightly rollup, so a year of history draws as quickly as a week.

Updated . Something wrong or missing? Tell us.