Run your workspace
Results & gaps
Results (/app/results; it was called
Analytics, and /app/analytics still forwards here) shows
whether your AI employees are working: how many real conversations
they had, how many visitors saw the chat, the leads they produced, and
the questions they could not answer. The figures, chart and
per-AI-employee table that used to open the Dashboard live here;
Home keeps this week's four numbers.
Period
Choose Last 7 days, Last 30 days
(the default) or Last 90 days at the top
(?period=7d|30d|90d). Every figure, the chart, the funnel
and the top lists follow it, and each figure is compared with the
period of the same length just before it.
Last 7 days is today and the six days before it, from midnight; 30 and 90 days work the same way. Home's This week, the Conversations list and Unanswered questions use the same stretch, so a number opens a list that shows exactly it. The period before is exactly as long, down to the minute: early in the morning today has only just started, and so has the stretch it is compared with.
One conversation count
A conversation is a chat where the visitor wrote
something. It is the same count everywhere: Home, the Conversations
list, the AI employee's pages, this page, the conversation export and —
since card #831 — the platform admin's dashboard, workspace and AI
employee lists, where the widget sessions are shown next to it
(Conversation::engaged). Your own tests in the playground
never count. Your plan's monthly conversations count the same way: a
conversation is counted once a month when the visitor's first message
is answered, never for a page load.
When the chat loads for a visitor it also starts a widget session, even when nobody types: one per visitor per day (a visitor who comes back within 24 hours of the session's start continues the same one). Those sessions are counted apart, as visitors who saw the chat. Before card #668 this page counted them as conversations, which is how one workspace saw 16,135 "conversations" here against 58 on Home. Page loads are counted only in the funnel below, as Shown.
The figures
| Figure | What it counts |
|---|---|
| Conversations | Chats where the visitor wrote something, started in the period. |
| Visitors who saw the chat | Widget sessions started in the period, whether or not the visitor typed: one per visitor per day, however many pages they open. |
| Messages | Messages in the period, from visitors, the AI employee and your team. |
| Leads | Visitors who left their details. A lead sent from a test chat (the preview in your dashboard) stays in Leads but is not counted here, just as test conversations are not. |
| Lead conversion | Leads divided by conversations. |
| Messages / conversation | Messages divided by conversations. |
Widget funnel
Three steps, each a part of the one before:
| Step | Counted when |
|---|---|
| Shown | The widget appeared on a page load (the widget.ready beacon). |
| Opened | The chat opened on that page load — a click, an auto-open, a trigger or a question from an inline block, once per page load. |
| Conversations | Opened, and the visitor wrote something: the conversation count above. |
The share next to each step is of the step before. A low Opened share points at the teaser or auto-open settings; a low Conversations share points at the greeting and starter questions. The opened beacon exists since card #502; when a window reaches back before it, a step can measure larger than the one before it, and the share shows — instead of a number above 100%.
How visitors ask
Every visitor message records how it was asked (card #792), and Results counts them for the period:
| Way | Counted when |
|---|---|
| Typed | The visitor wrote the question in the chat. |
| Answer buttons | The visitor tapped one of the buttons under an answer. |
| Suggested questions | The visitor tapped a question offered before the chat began — your starter questions, the page's own questions, or Ask about this product. |
| From the page | The question came from a chat block placed on the page (shown only when there are any). |
Messages sent before this was recorded have no way attached and are left out, so the card fills from 15 October 2026 on. In a conversation, a visitor message that was a tap says so under the bubble — tapped an answer button or tapped a suggested question.
Answer quality
Volume says how much your AI employee talks; this card says how well (card #796):
| Figure | What it counts |
|---|---|
| Thumbs down | The share of answers visitors rated that got a 👎, and how many were rated. |
| Weak knowledge match | The share of answers whose knowledge match sat under that AI employee's own threshold (Behaviour › How it answers › How sure it must be). These answers leaned on the page or on general knowledge — open them in Conversations, where each answer shows its knowledge match, and add what was missing. Counted from 15 October 2026: before that, an answer given on a page the widget could read showed a match of at least 85% whatever it found, so older answers would look falsely fine. |
| Products recommended most | The five products whose cards the assistant showed most, with how often each card was clicked. Counted from 16 October 2026 on. |
| Corrected before sending | How many product facts in answers contradicted your product records — the product catalog built from your product pages, or your product register — and were corrected before the visitor saw them (card #801), with the latest corrections: what the draft said, struck through, and what was sent. Each corrected answer is also marked in its conversation. Counted from 17 October 2026. |
Trend and top lists
- Traffic and capture trend — conversations per day (the blue area) and leads per day (the dotted line), with the number of AI employees that had conversations, published AI employees, tracked pages (different page addresses the chat was seen on in the period; recounted at most every ten minutes, because on a busy site it means reading thousands of addresses) and open gaps.
- Top AI employees — up to six, by conversations, with their status, leads and conversion.
- Top pages — the pages with the most conversations, and the leads they produced.
- Revenue attribution — conversations that clicked through to a product, and orders attributed to them. The full report is Revenue at the top (Revenue attribution). Below it, Requests via your AI employee (card #836): per week, the clicks on the AI employee's buttons to your own pages and forms — a quote form, a booking page — and the leads left in the chat. When quotes and bookings happen on your own form, those clicks are the requests your AI employee sent you. Test chats never count.
- Content gaps — the open questions your AI employees could not answer, most asked first. Content gaps at the top opens the full list (Content gaps).
- Visitor satisfaction (CSAT) — twelve weeks of thumbs up and down, and recent low-rated conversations (CSAT).
Advisor at the top opens the Advisor Mode sessions (Advisor Mode). Conversations and Leads are menu items, so the page no longer repeats them as buttons.
Export
Conversations are exported from the Conversations list: Export (owners and admins) exports the conversations of the AI employee and period the list is filtered to, as CSV or JSON. See Conversation export.