Analytics
Analytics answers one question: are your assistants actually doing their job? The page leads with five outcome tiles, then splits the detail across five tabs — Overview, Quality, Voice, Messaging, and Usage & Cost.
Filters and date range
Two controls sit above the tiles and apply to the figures on the page.
- Assistant — defaults to All Assistants. Pick one to narrow the figures to that assistant.
- The date range buttons — Today, 7d, 30d, 90d, Year, Custom. The page opens on 30d.
Choosing Custom reveals two date pickers and an Apply button. Nothing changes until you click Apply. The end date cannot be in the future, and a custom range is limited to 365 days and cannot start more than two years ago.
Every figure with a change indicator compares your selected range against the period of the same
length immediately before it. Volume metrics show a percentage change; rate metrics such as
Resolution and Self-served show a change in percentage points (pts). If either period had
no activity, no indicator is shown rather than a misleading ↓100%.
Note
The Assistant filter narrows chat and call figures. SMS, email and appointment figures are counted for the whole workspace regardless of which assistant is selected. The Needs review feed on the Quality tab is also not narrowed by the Assistant filter — it always spans every assistant.
The hero row
| Tile | What it measures |
|---|---|
| Conversations | Every interaction in the range — widget chats, calls, SMS messages, and emails that were sent or delivered |
| Resolution | Share of graded conversations and calls that met the assistant's goal, judged by an LLM |
| Self-served | Share of chats handled without handing off to a human agent |
| Satisfaction | Positive share of the replies visitors actually rated |
| Credits Spent | Credits used across your assistants in the range |
Resolution and Self-served carry a second line: how many conversations were graded, and how many were handed to a human.
Resolution needs a goal
Resolution stays blank — a dash and Not measured yet — until an assistant has a Goal set. The goal is a plain-English sentence describing what success looks like, on the General settings tab of the assistant editor. Once it is set, every conversation and every call is graded against that goal when it ends: achieved or not met, plus a grade from 1 to 5.
Conversations that ended before you set a goal are not graded retroactively, so the tile fills in as new conversations complete.
Satisfaction and self-served
Satisfaction counts only replies a visitor rated with the Helpful / Not helpful buttons in the widget, so it reads as a dash until someone rates something. A handful of ratings is not a survey — treat it as a signal, not a score.
Self-served counts a chat as escalated when the assistant handed it to a live agent. Handoffs that go out by email or to another system are not counted here.
Overview tab
Six small tiles sit above the charts:
| Tile | What it shows |
|---|---|
| Total Messages | Every message across the conversations in range |
| Avg Messages / Chat | Average message count per conversation |
| Completion Rate | Share of conversations that ended cleanly, rather than staying open or erroring |
| Active Contacts | Contacts with at least one interaction in the range |
| Call Objective Met | Only shown when calls in the range recorded an objective |
| Appts Completed | Only shown when appointments were booked in the range |
Below them:
- Channel Activity — daily volume as a line per channel. Only channels with activity are plotted. Hover a point for exact counts.
- Peak Hours — conversations by hour of day, bucketed in your own time zone. Your browser sends its offset with the request, so the histogram matches your working day, not UTC.
- Conversation Status — Completed, Active and Errored counts with a total. When appointments exist, a second block breaks them into completed, cancelled and no-show.
- Assistants — one row per assistant with conversations and credits. The Share bar is that assistant's slice of the estimated cost for the range. The dot beside each name is bright when the assistant is enabled and muted when it is switched off.
- Daily Breakdown by Channel — the same daily data as stacked bars, so you can see the mix.
Quality tab
Four cards, in the order you should read them.
Resolution shows a donut split into Resolved and Not met, with the resolved count, the not-met count, the average grade out of 5, and how many conversations were graded. Without a goal set, this card is an activation prompt instead — Turn on resolution grading — with a Set a goal button that takes you to your assistants.
Satisfaction shows the positive share of rated messages, a split bar, and the like and dislike counts. If any conversations are flagged, a link underneath reports how many need review and opens Conversations. When the workspace is busy the card notes that feedback is Sampled — most recent 2,000 conversations.
Needs review is the practical one: a feed of conversations where a visitor disliked a reply or the assistant fell back on an "I don't know" style answer. Each entry carries a Disliked or Fallback badge, the detected language, how long ago it happened, the visitor's last message and the reply that was flagged. Click one to open Conversations. This is the fastest way to find gaps in your knowledge base.
Languages lists the detected language of conversations in the range with a count for each.
Voice tab
Empty until you have call activity. Once calls exist:
| Tile | What it shows |
|---|---|
| Total Calls | Calls in the range |
| Avg Duration | Mean call length |
| Inbound / Outbound | Split by direction |
| Connect Rate | Share of calls that actually connected |
| Objective Met | Share of calls that met their objective, or a dash when no call recorded one |
| Top Disposition | The most common outcome |
Two charts follow: Call Volume Over Time, stacked inbound against outbound per day, and Disposition Breakdown, a donut of every call outcome in the range.
Messaging tab
Three sections, each of which only appears when that channel had activity. If none did, the whole tab shows a single empty state.
- SMS — Sent, Received, Failed and Opt-outs, over a daily sent-versus-received line chart. Failures and opt-outs are coloured when they are above zero, so they are hard to miss.
- Email — Sent, Delivered, Open Rate, Click Rate and Bounce Rate, over a daily sent / opened / clicked line chart.
- Appointments — Booked, Completed, Cancelled, No-show and Completion Rate, over a daily bookings bar chart.
Usage & Cost tab
Where the credits went.
| Tile | What it shows |
|---|---|
| Total Tokens | Model tokens consumed in the range, with the change against the previous period |
| Credits Spent | Credits deducted across your assistants |
| Avg Session Duration | Average time from a conversation's first message to its last |
| Success Rate | Share of finished conversations that ended cleanly rather than erroring |
The charts:
- Token Usage Over Time — input and output tokens per day. The input/output split is an estimate; only the combined total is measured directly.
- Sessions Per Day — conversations started each day.
- Conversation Length — a histogram of how long conversations lasted, from first message to last, capped at one hour. This is conversation length, not the assistant's reply speed.
- Credit Share by Assistant — a donut of credits spent per assistant.
- Conversation Length Percentiles — P50 (median), P95 and P99.
- Usage by Assistant — a table of conversations, tokens and credits per assistant, with each one's share of total credits and a credits total in the footer.
Paid feature. Conversation Length and Conversation Length Percentiles need a paid account. Until then both cards read Available on paid accounts — add credits to unlock. See Billing and credits.
Exporting
Export in the page header offers Export as CSV and Export as JSON. Both honour the current assistant filter and date range.
The export is a conversation-level record — one row per conversation, with the date, assistant name, session ID, message count, token count, status and duration in seconds. A single export is capped at the 10,000 most recent conversations in the range.
Paid feature. Export requires a paid account. Add credits to unlock it.