Before you start
Before you start, sign in to the dashboard and select the organization you want to work in.- You need a role with analytics access. To choose outcomes or change the layout, you need the Admin role.
- You need at least one deployment that has handled a conversation. Until then the page shows Waiting for first events.
- Read Analytics overview for the scope bar and the Conversations drawer. This page does not repeat them.
Open Analytics and set the scope
1
Open Analytics from the sidebar
The page opens with its cards in the order your organization saved.
2
Choose a range, a scope, a channel, and Live or Test
Every card follows the scope bar. For example, choose 28d, then Support agent, then All channels, then Live.
Key metrics
The Key metrics card shows the numbers your team chose, each with its change from the previous period. An Admin changes the tiles with Customize. By default, the card shows:
A tile that counts end users or conversations names its unit, Visitors or Conversations, in its help.
Below the tiles, the Units: line lists the units the card uses, such as visitors and conversations. Hover it to read what each unit counts. Other tiles are in the gallery. See The Key metrics gallery.
Each default tile has a line chart. Other outcomes you ticked, and results you chose to show, are in the gallery under Customize. Select a number to list its conversations in the drawer.
Each web link, phone number and email address written in a flow message gets one row under Links. A web link is counted by its site and path, without the query. A link whose path gets a value from a variable is not counted as shown, so one person’s data never becomes a row. Shown and click counts cover web chat only, because the widget records the clicks.
The row shows the click rate and the clicked and shown counts, such as 75% and 3 / 4. Add the row to put its click rate on the card, with the clicked and not clicked split below it. Open shown under the link to add the Shown result on its own, or to pick it as an outcome. Links in code and images are not counted as shown.
Other phone and email link clicks count as the Phone link and Email link results, and the number or address is not recorded. This includes links that get their number or address from a variable, and links in an AI reply. A web link in an AI reply keeps its click count but has no shown count. You can tick each result as an outcome.
Selecting either the launcher or its callout message counts as a click-open for Click rate. Dismissing the callout does not count as an open.
- Open rate also includes automatic opens and visits to the agent link.
- Click rate leaves out browsers that loaded only the agent link, because that page has no launcher.
The Key metrics gallery
While you customize the card, Add metric opens the Key metrics gallery. Search it with Search metrics and outcomes, or show one group with All, Your outcomes, Visitors, or Conversations. Select Add on a tile to put it on the card. A tile already on the card reads Added. The Visitors group appears only when Channel is set to All channels or Web. Its tiles count end users on your website:
The Visitors group also holds Visitors, Opened the widget, Open rate, and Click rate. The Conversations group holds Conversations and these tiles:
Waiting for an agent and Open now are live. They show the current state and ignore the date range.
The tile reads Now with All channels and the default traffic, or Now · ignores filters when you pick a channel or narrow the traffic. With All deployments in Scope, both tiles include voice calls. The email report leaves out the CSAT, Waiting for an agent, and Open now tiles.
The CSAT drawer has these parts:
- Average rating, Rated 4 or 5, and Response rate, the share of surveys sent that were answered.
- Ratings, the ratings split by score. View rated 1 or 2 opens those conversations.
- By who resolved it, the ratings of conversations that the AI resolved and that your team resolved.
- Totals holds Outcomes, the conversations with at least one outcome, and Outcome rate, their share of all conversations.
- Shown without counting holds results that show on the card without counting as an outcome. Add more adds tools, gambits, and events there.
Choose which outcomes count
1
Select Customize on the Key metrics card, then Add metric
The gallery opens. Your outcomes is the first section.
2
Select New outcome
Tick each result that counts as this outcome. Results are grouped under Actions, Fields captured, Buttons clicked, and Gambits. Each row shows how many conversations reached it in the period.
3
Select Done, then Done again
The first Done returns to the gallery. The second Done closes it.
4
Select Save on the card
The outcome shows on the Key metrics card and ends the Funnel.
Funnel
The Funnel card follows every end user from the page load to a conversation with an outcome. The steps are Visitors, Opened the widget, Started a conversation, Conversations, and Outcome. Between steps, the funnel shows what dropped out, such as Did not open, No conversation, and No outcome. One end user can start more than one conversation, so Conversations can be higher than Started a conversation. The web funnel counts browsers linked to a recorded widget open. Its conversation count can differ from the Conversations tile on Key metrics, even with Web selected. Select a step to open its conversations. This works on these steps:- Conversations, one step for each channel when there is more than one
- Each step you added, and Dropped
- Each outcome, and No outcome
- Add step adds a gambit or an outcome that conversations must reach before they count. A gambit with buttons is always its own step.
- A gambit belongs to one deployment. When you add one with All deployments selected, the scope changes to that deployment. A step for a gambit in another deployment shows when you select that deployment in Scope.
- Narrow counts only end users who saw a page or site, or arrived from a referrer or campaign. It also sets which kinds of traffic count, under Bots and crawlers.
Trends
The Trends card charts Visitors, Conversations, and Outcome over time. The Period control switches between three views:
By version needs one deployment in Scope. Until the agent is published a second time, it reads Available after the second publish.
Topics
The Topics card lists the topics your customers talk about, with each topic’s share of conversations. See Find topics in conversations.Busy hours
The Busy hours card is a grid of weekdays by hour in your timezone. The tabs above the grid pick the metric. A line under the grid names the busiest hour.- Visitors is always a tab when Channel includes web.
- The other tabs are the metrics an Admin ticks under Show metrics in the card’s options menu. By default these are Handoffs and Unattended. An Admin can also tick Opened the widget, Conversations, and Outcome.
- The ticked metrics save with the layout, so every member sees the same tabs. With no tab to show, the card reads No metrics shown.
AI performance
The AI performance card shows how the conversations that started in the period ended, who handled them, and how fast the AI replies. Conversations that ended with no message from the end user are left out. The left side leads with the share of conversations resolved. A bar below splits the conversations into Resolved by AI, Resolved by human, Unresolved, Visitor left, and Open. Select a part of the bar to open its conversations. The Who handled it table has a row for each way a conversation was handled:
The columns are Resolved, Unresolved, Visitor left, Open, Total, and CSAT. Open counts conversations that have not ended yet. Select a row or a count to open those conversations. No answer found in KB, below the table, counts the conversations where the knowledge search found nothing.
The right side shows the typical reply, the time that half of the AI replies beat. A bar splits the replies into Under 2s, 2–10s, and Over 10s. The Most time goes to row names the step that takes the largest share of reply time.
The AI replies drawer
Select Reply time details, or the typical reply, to open the AI replies drawer. It measures the time from an end user’s message to the AI’s complete reply.
The steps are these:
Model and Agent work show separately only when at least 95% of the replies carry model timing, in this period and in the previous one. Otherwise they show together as Model and agent work. A reply that runs no model, such as a greeting from a Text gambit, never carries model timing. An agent whose flow speaks first often shows the combined step for this reason.
Since October 27, 2026, Tars times a streamed reply in the widget until the reply is complete. Before that date, the timing stopped when the reply started to stream. A range that includes earlier days can show slower replies, but the AI is not slower.
Team
The Team card covers your team’s work in live chat. It shows how the AI’s handoffs ended, and how each human agent performed. Human messages, resolutions and ratings count even in conversations the AI never handed off, such as one an agent joined directly.- First response leads the card. It is the typical time from a handoff to a human agent’s first message.
- A bar splits the handoffs into Resolved, Picked up, not resolved, and Unattended.
- A live row, marked Now or Now · ignores filters, counts the conversations waiting for an agent and the open ones. It also shows when the oldest waiting conversation started.
The Team row counts each conversation once. A human agent no longer in the organization shows as Former agent.
The Agent table needs a role that can read live chat. Select a row to open that human agent’s conversations.
Team details opens the Team drawer. It compares this period with the previous one and charts first response and resolved conversations over time. It also splits pickups by first response time.
If the range has no handoffs and no human agent activity, the card reads No handoffs in this window. A message, resolution or rating from a human agent fills the card even without a handoff.
Before October 27, 2026, Tars did not record a human agent taking over a parked conversation. In earlier days, those conversations count as Unattended and have no First response. Tars did not correct the earlier data.
Labels
The Labels card has a bar that compares the labels by conversations, and a table with one row for each label. A conversation counts under every label it has now, so one conversation can count under two labels.
Select a row or a part of the bar to open its conversations. The card needs a role that can read live chat. Without one, it reads Needs live chat access.
With no labels, the card reads No labels yet. Create labels in the live chat settings. See Set up labels.
Fields captured
The Fields captured card lists each field that a form, a button, or the AI Agent gambit saved. The columns are Field, Asked, Answered, and No reply. No reply counts the conversations where the field was asked and the end user did not answer it. An empty card reads No fields captured in this period.Sources
The Sources card ranks where Visitors and Conversations come from. A page path with and without a trailing slash is one page. Switch the view with Pages, Sites, Referrers, Campaigns, and Bots and crawlers. The Pages, Sites, Referrers, and Campaigns views show Visitors, Conversations, Outcome, and Rate. On Sites, a Since column shows the first day the widget was seen on that site. The Bots and crawlers view shows Kind, Visitors, Opened the widget, Conversations, and Share. Share is the kind’s Visitors divided by all Visitors.Campaigns
The Campaigns card lists outbound campaigns in the period. The columns are Campaign, Sent, Delivered, Replied, and Reply rate. Select a row to open the conversations that the campaign started. Select the campaign name to open the campaign.Components
The Components card covers each component in your agents: how often it was shown and answered, meeting bookings, and the options people pick. Only published agents are counted. See Component gambit.
The table lists each component with Shown and Result. Result is the one number that matters for that kind. It is booking conversion for the Meeting Scheduler and submission rate for a form. For cards it is click-through, for choices the response rate, and for inputs the completion rate.
Open a row for its breakdown. It shows what happened each time the component was shown, such as Booked a meeting, Typed a reply rather than using it, or Moved on without answering. For a choice, Offered, Picked, and the share of all picks show per option.
Option labels from the agent never reach the report as written. Email addresses, phone numbers, card numbers, and links are masked, together with your organization’s custom rules and, when it is on, name detection. See Personal data protection.
Tools and integrations
The Tools and integrations card covers tool calls, failures, and connections that need attention. Each call counts once, after retries. Test runs show under Test in the scope bar. Voice calls, and calls made by App tools from inside CodeKit, are not included. The summary shows Tool calls, Success rate, Failed calls, and Conversations affected. Each shows its change from the previous period. Conversations affected counts distinct conversations with at least one failed tool call.
Used by names up to two agents with their flow steps, and folds the rest into a count. Hover it to list each step with its calls, and select a step to open it on the builder.
Select a row to open its panel, with the breakdown by action and by agent and the Recent calls. Select a call to open its details. See Monitor tool calls.
Countries and Devices
Countries ranks where people are by conversations. Devices compares phone with desktop. Neither card is on the page by default. An Admin adds them with Add chart.WhatsApp messages
The WhatsApp messages card counts the replies your agents and human agents sent on WhatsApp, and how many of them Meta charged for. It is not on the page by default. An Admin adds it with Add chart. The summary shows these numbers:- Replies sent: replies delivered on WhatsApp.
- Billed by Meta: replies Meta charged for, and their share of the replies.
- Per conversation: the average number of replies in each WhatsApp conversation.
Change the layout
The Components and Tools and integrations cards are part of the default layout. An organization that saved its layout before they existed adds them with Add chart.1
Hover a card and open its options menu
The menu offers Move up, Move down, and Remove.
2
Select Add chart below the last card to add a card
The Add a chart dialog lists every card that is not on the page.
3
Select Save in the bar at the bottom of the page
Every member of the organization now sees the new layout. Select Discard to drop your changes instead.