AI Insights

This one is about letting you see what the AI’s been doing on your behalf. AI Insights turns your closed AI conversations into a performance picture:

• resolution rate
• handover rate
• stuck queries
• customer sentiment
• topic breakdown and
• a priority chart that tells you what to fix next.

It lives in Reports → AI Insights, on all plans.

What changes for your team

You’ve had two ways to tell how the AI agent was doing:

1. Read conversations yourself.
2. Wait for CSAT, which only covered the customers who chose to respond.

Neither was a real answer.

AI Insights gives you one. It covers every closed ticket that was first assigned to your AI agent, and it breaks the picture into parts you can act on.

7 things on the page

1. Performance Overview. The Overview tab opens with every AI-handled ticket sorted into three outcomes:

Outcome	What it means
Resolved by AI Agent	Closed by the AI, no human handover
Reassigned to Human Agent	Handed to a person, who closed it
Stuck Queries	The AI visibly failed and the conversation never recovered: it said it couldn’t help, didn’t know the answer, never replied properly, or a handover it attempted didn’t happen

Reading the three together tells you how much volume the AI is carrying, and how much of the remainder is a handover by design versus a genuine gap you can close.

2. Customer Sentiment. How customers felt about the handling of their issue overall, from first message to close, including any human agent involved. Every closed conversation is read as one of three: Negative, Neutral or Positive. The card shows each as a bar with its share of conversations, and next to it how many points that share moved against the previous period of the same length, so you see the direction as well as the level.

Most conversations read Neutral. That’s expected: a customer who asked a question, got the answer and left is transactional, not unhappy. Negative means the customer showed real frustration — anger, sarcasm, repeated unresolved complaints. Being handed over to a person is a normal part of support and does not on its own make a conversation Negative. Positive means the issue was resolved and the customer showed they were satisfied; a clear “that worked” counts, an explicit thank-you isn’t required.

This is not a survey score. It’s derived from every closed conversation, not only the customers who chose to respond. That makes it a different measure from CSAT, which remains on its own tab.

3. Ticket Flow Overview. A chart mapping every AI-handled ticket across four columns, left to right: Channels → AI Handling → Topics → Customer Sentiment. Each band’s width is the ticket count taking that route, and each block shows its count and share of the total. The final column splits into Positive, Neutral and Negative, so you can trace which channels and topics feed each.

Hover a block to highlight every route through it. The rest dims. The layout stays put.

4. Pinning. Click a block to pin it. A pinned highlight survives the pointer leaving the chart, so you can scroll, read it properly, or capture it.

Pins combine two ways:

• Same column: widens the view (pinning two channels keeps tickets from either).
• Different columns: narrows it (pinning a topic and an outcome keeps only tickets with both).

If a combination has zero tickets, the selection stays pinned and reads 0 (0%) rather than silently resetting. An impossible filter is visible, not confusing.

5. Topics come from your own list. The Topics column is driven by what you define on Settings → AI Topics, not by a fixed classification. Each topic has a name and a definition, and the definition is what conversations are matched against. So the wording of your definitions directly determines how your reports group conversations.

Every project starts with a default set of ten common support topics — order status and tracking, returns and refunds, payment issues, product information, and so on — written with ecommerce support in mind. Treat them as a starting point to edit, not a classification to accept.

Every ticket gets exactly one topic. Anything that doesn’t fit your active topics lands in Others, a system-maintained catch-all. It shows in your reports but isn’t on the AI Topics page, because it’s managed for you.

A steadily growing Others count is worth reading as a signal. It means conversations are arriving that your topic definitions don’t describe.

6. Topic Breakdown. The Topic Breakdown tab ranks your topics three ways:

View	What it answers
Reassignment Topics	Which topics get handed to a person most often
Negative Sentiment Topics	Which topics end badly most often
Stuck Queries	Which topics the AI cannot resolve

Select a topic to open its overview: total ticket count, the same Customer Sentiment breakdown as the Overview page — Negative, Neutral and Positive shares with their movement against the previous period — a trend chart across the period, and a Recommended Next Step where one is available.

Below that, a sample of the actual tickets, each with the customer, the date, and a Customer’s Concern summary. View All Tickets opens the full list, with your date range and channel selection carried across. Every Ticket ID in that list opens the conversation in the Inbox in a new tab, so you can read the ticket and return to the report exactly where you left it.

The concern summaries are the point. You can read the pattern behind a number without opening every conversation.

7. Focus Areas. The Impact vs. Volume Matrix on the Focus Areas tab plots every topic on one chart. Volume runs left to right, impact runs bottom to top, and each dot is sized and coloured by priority (Low through Critical).

Impact is calculated from how often a topic’s tickets get escalated, get stuck, or carry negative sentiment. The top-right corner is high volume meeting high impact.

That’s your shortlist.

Hover any dot for the numbers behind it: Volume, Reassignments, Negative CX, Stuck Rate, and the resulting Impact Score. A priority ranking is never something you have to take on trust.

Because the chart is built from your own conversations against your own topics, the answer differs by business. The same page tells one team their weak point is product quality and another that theirs is order tracking. The second is something you can usually fix with better AI configuration. The first is not.

Before and after

Capability	Before	Now
AI performance	No consolidated view across channels	One page: resolution, handover, stuck queries, sentiment, attributed sales
Handover visibility	Visible only ticket by ticket	Handover rate as a headline figure, with the topics driving it, ranked
Unresolved conversations	Indistinguishable from ordinary handovers	Stuck Queries separated as its own outcome, with its own ranking and ticket list
How customers felt	CSAT only, limited to customers who answered a survey	Negative, neutral and positive shares across every closed conversation, including the human-handled part, with the movement against the previous period
What conversations were about	Manual reading, or tags applied after the fact	Automatic assignment against topics you define, starting from a default set, with a catch-all for the rest
Reading the chart	Static, one filter at a time	Hover to isolate, pin to hold, combine pins across columns to narrow
Deciding what to fix	Judgement from separate reports	A single priority chart ranking topics by volume against impact
Getting to the tickets	No route from a number to the conversations behind it	Every topic opens its ticket list with each customer’s concern summarised, and every ticket opens in the Inbox

How to set it up

Start with your topics (you must do this first)

1. Go to Settings → AI Topics.
2. Review the topics already listed. Every project starts with a default set of ten common support topics plus Others. Each has a name, who created it, and a status toggle.
3. Read each definition and adjust it to match how your customers actually write. The definition is what conversations are matched against, so it does more work than the name.
4. Use + Create New Topic to add anything specific to your business the list doesn’t cover.
5. Disable any topic you don’t want conversations grouped under.

Topics must be active before AI Insights has anything to report on. With none active, the page is empty.

Read the overview

1. Open Reports → AI Insights.
2. Set the date range. Narrow to one or more channels if you want; the selection follows you into every topic’s ticket list and export.
3. Read Ticket Flow Summary for the three outcomes against the total.
4. Check Customer Sentiment for the share of Negative, Neutral and Positive conversations, and how many points each moved against the previous period.
5. Check Total Sales by AI Agent for sales value from orders placed while customers were talking to the AI.

Follow a group through the flow

1. Scroll to Ticket Flow Overview.
2. Hover a block to highlight every route through it. Click to pin it.
3. Add a pin in a different column to narrow. Add a pin in the same column to widen.
4. Read left to right: where those tickets came from, how the AI handled them, what they were about, how they ended.
5. Click a pinned block to release it, or click empty space to clear all.
6. Use Export to take the flow as a file (arrives by email).

Rank your topics

1. Open Topic Breakdown.
2. Switch between Reassignment Topics, Negative Sentiment Topics, and Stuck Queries.
3. Select a topic for its overview: total tickets, sentiment breakdown, trend, and the recommended next step.
4. Read the ticket sample for the concerns behind the number. View All Tickets for everything; click any Ticket ID to open that conversation in the Inbox.

Decide what to fix

1. Open Focus Areas.
2. Read the Impact vs. Volume Matrix. Top-right corner = highest priority.
3. Hover a dot for its Volume, Reassignments, Negative CX, Stuck Rate, and Impact Score.
4. Work on the critical and high-priority topics first.
5. For each one, decide: is the fix a configuration change (better knowledge, clearer instructions, a new automation), or is it outside the AI’s reach?

Availability

Plans	All plans
Where	Reports → AI Insights
Topic management	Settings → AI Topics

Prerequisites

• The AI agent must be enabled on the project.
• At least one topic must be active on Settings → AI Topics. With none active, the page has nothing to report.
• The person viewing needs analytics read access to the project.
• Every project starts with a default set of topics written for ecommerce support. Review them against your own business before relying on the report; add what’s missing and disable what doesn’t apply.

Limits

Only closed tickets are reported. AI Insights covers tickets first assigned to your AI agent, after they’ve been closed and evaluated. Open tickets and tickets a person handled from the start are outside its scope.

One topic per ticket. Each conversation is assigned exactly one topic. If it spans multiple, the AI picks the primary one.

Others is not editable. The catch-all topic is system-maintained. You can’t rename or redefine it, but you can read it as a signal that your definitions have gaps.

FAQ

Do I have to do anything to see data? One thing: review your topics on Settings → AI Topics. Every project starts with a default set, so the page populates as soon as tickets close, but the defaults are written for ecommerce support. Tune the definitions to your business, add what’s missing, and disable what doesn’t apply.

Where do I find it? Reports → AI Insights. It sits alongside your other reporting tabs.

Is this a replacement for CSAT? No. Sentiment and CSAT are different measures. Sentiment covers every closed conversation. CSAT covers only customers who responded to a survey. Both are available; use them for different questions.

Is a handover counted as negative sentiment? No. Being routed to a person is a normal part of support. A conversation reads Negative only when the customer shows frustration — anger, sarcasm, repeated unresolved complaints. Most conversations read Neutral: the customer asked, got an answer, and left.

What counts as a “stuck query”? A conversation where the AI visibly failed and never recovered: it said it couldn’t help or couldn’t reach a system, it didn’t know the answer to a question about your business, it never replied properly to a clear request, or it tried to hand over and the handover didn’t happen. A conversation that reached a person isn’t stuck, and neither is spam or a test message the AI declined to engage with. That’s what keeps a planned reassignment and a genuine failure on separate lines.

Can I change how topics are assigned? Yes, by editing the definitions. Go to Settings → AI Topics and reword a topic’s definition. The definition is what the AI matches against, so changing it changes how conversations are grouped going forward.

Does it cover tickets from all channels? Yes, as long as the ticket was first assigned to your AI agent. The Ticket Flow Overview shows channel distribution as its leftmost column.

Which plan is this on? All plans.

Does this add work for my team? The setup does, briefly. Review and tune your topic definitions once. After that, the page populates on its own as tickets close.

What’s next

AI Ecommerce. Connect your store so the agent answers, recommends and orders from your live catalog.

AI Sales UX Livechat. The same catalog, browsing and checkout, now in the website widget.