Analytics
Chatbot analytics that tell you what to fix next.
Chatbot analytics measure what your AI agent did with the conversations it handled: how many it took, how many it resolved alone, and where it stopped. Outlearn tracks that automatically from the first conversation and puts the auto-solve rate, the handoff reasons and the questions it could not answer on one screen.
Free plan, no card. Analytics is on from the first conversation, on every plan.
Last updated 9 September 2026.
Analytics · Overview
Sample workspace- Conversations
- 1,284
- Auto-solved
- 51%
- Time saved
- 96 h
- Satisfaction
- 4.6 / 5
One per visitor who sent a first message.
Closed with no human handoff.
Estimated, from automated resolutions.
Average rating left after a conversation.
Sample workspace. How 1,284 conversations ended: resolved by the agent, handed to a person, or closed without either.
- Auto-solved 51%
- Handed to a human 6%
- Ended without either 43%
Sample workspace. Why the agent escalated, as a share of the conversations it handed over. The green bar is the one you can close with better content.
- Fixable with a source or a snippet
- A person was always the right answer
Sample workspace. Questions the agent could not answer from any connected source, this month, most asked first. Green marks the two worth writing today.
- Recurring — write it
- Watch it
Sample workspace. What people came to ask about, grouped automatically, as a share of all conversations in the range.
Scroll a chart sideways to read the whole of it. A sample workspace, to show the shape of the dashboard. Our own production numbers are below, with their denominators attached.
Measured, not promised
The numbers we publish about our own agent
A vendor that quotes a resolution rate without a denominator is quoting marketing. Ours comes with the arithmetic, including the conversations that ended in neither outcome, in the AI support benchmark.
- 29,003
- conversations measured, Oct 2025 to Aug 2026
- 45.9%
- resolved automatically, no human involved (13,299 / 29,003)
- 5.7%
- handed to a human (1,662 / 29,003)
- 2.7
- knowledge sources connected by the average workspace
Reading the dashboard
Four AI agent metrics, and the question each one answers
Every support dashboard has numbers. These four are on the Overview because each of them changes a decision you were going to make anyway.
Conversations
Is anyone actually using it?
The count of conversations the agent handled in the range you picked. A conversation starts when a user sends their first message, so this is adoption, not page views.
Auto-solved
How independent is the agent?
The share of conversations resolved with no human handoff. The documentation calls it the most important metric, and it is the one that moves when you connect a source or close a gap.
Time saved
What is it worth?
An estimate of the hours your team did not spend, derived from the conversations the agent resolved on its own. It is the line a support lead takes into a budget conversation.
Satisfaction
Are the answers any good?
The average rating people left in the widget's rating popup. A high auto-solve rate with a low satisfaction score is a warning, not a win, and you can only see that if both are on the same screen.
The vocabulary
Customer service metrics, defined once
Half the arguments about a customer service metrics dashboard are arguments about denominators. Each of these has a page with the formula, a worked example and what the number hides — and the eight read together explains which pairs must never be read alone.
- First response timeFRT
- Average handle timeAHT
- First contact resolutionFCR
- Resolution rateDefinition
- Cost per ticketDefinition
- Deflection rateDefinition
- Containment rateDefinition
- Agent assistDefinition
All eight live in the support metrics glossary.
The part that changes something
Content gaps turn the dashboard into a backlog
A number tells you the agent missed. A gap tells you which question it missed and how often. That is the difference between a report and a to-do list.
The agent says it doesn't know
A content gap is a question the agent could not answer from any connected source. It says so rather than inventing something, and the question is logged.
The gaps arrive ranked
The Content Gaps panel lists what people asked and the agent could not answer in the date range you selected, most frequent first. That ordering is the backlog.
You fix the source, not the bot
The answer exists but is not reachable: check the connection, resync, check accessibility. It does not exist: write the article or the snippet. It is weak: improve it and resync.
Out of scope stays out of scope
Some questions should not be answered at all. Update custom instructions so the agent declines gracefully, and no content changes hands.
Because the agent reads the sources you already keep and re-reads them on a schedule, closing a gap means editing your documentation once. The conversations that still need a person go through the handoff with the full transcript, which is why the handoff panel can name a reason at all.
The difference
Most chatbot dashboards count. This one diagnoses.
A typical chatbot dashboard
Outlearn
A count of chats, and nothing about how many ended in an answer.
Auto-solved is the headline: the share of conversations resolved with no human handoff, on the same screen as the volume.
One deflection number that quietly counts everyone who closed the window.
Handoffs and their reasons are reported separately, so an escalation is visible instead of absorbed into a nicer number.
A report you export, read once and file.
Content Gaps is a work queue: every unanswered question names a source to add, fix or resync.
Aggregates only. You never see the conversation behind the number.
Search every conversation, filter by topic, contact or status, and read the messages, the actions it triggered and how it ended.
Analytics as an upgrade, priced per seat like everything else.
In the Analytics tab from the first conversation. Plans are priced by credits for the AI's work, never per seat.
Pricing, for the record: credits for the AI's work, never per seat, and analytics is not a tier. Free $0 with 100 credits, Starter $99/month with 2,000 credits, Growth $199/month with 4,500 credits, Scale $599/month with 15,000 credits. 14-day trial on paid plans, no card. Full detail on the pricing page.
Connect one source and watch the first week of numbers arrive.
Point the agent at the documentation you already keep, put it on your site, and the Analytics tab fills itself. Free plan, no card.
What support leads ask about chatbot analytics
What is chatbot analytics?
Chatbot analytics is the measurement of what an AI agent did with the conversations it handled: how many it took, how many it resolved on its own, how many it escalated to a person, and what it could not answer.
Outlearn tracks this automatically from the moment your first conversation happens, and reports it in the Analytics tab rather than in an export you have to assemble.
What does the Outlearn analytics dashboard show?
The Overview leads with four numbers for the date range you pick: Total Conversations, Auto-Solved, Time Saved and Satisfaction Score.
Under them, a chart plots total conversations against auto-resolved conversations by day or by week, and panels break the period down by Common Topics, Top Users, Handoff Reasons, Content Gaps and Conversations by Country. The range defaults to the current month and the whole view exports to PDF or CSV.
What is a good auto-solve rate for an AI chatbot?
There is no single right answer, and a vendor quoting one without a denominator is quoting marketing.
Our own first-party benchmark is published with its arithmetic: across 29,003 conversations from October 2025 to August 2026, 45.9% were resolved automatically with no human involved, and 50.8% of the 10,694 conversations in the last 90 days of that snapshot. Your number depends on what customers ask, what you connect, and whether that content is current.
How is auto-solve rate different from deflection rate and containment rate?
They are measured over different denominators, which is why teams argue about them.
Containment rate is the share of bot conversations that end without escalating to a human. Deflection rate is the share of all issues raised that are fully resolved without reaching a human agent, including the ones that never opened a chat. Resolution rate is a queue metric: tickets resolved in a period over tickets received. Each has its own definition page in our glossary, with the formula and what the number hides.
What are content gaps and what do I do with them?
A content gap is logged when someone asks a question and the agent cannot find a relevant answer in any connected source, so it says it does not know instead of guessing.
The panel ranks those questions for the period you selected. Four things fix them: reconnect or resync a source whose answer exists but is unreachable, write the missing article or snippet, improve a weak answer and resync, or update your custom instructions when the question was never in scope. Review the list weekly or monthly and start with the gaps that keep repeating.
Can I read the individual conversations behind the numbers?
Yes — Conversation & Training in the Analytics tab lists every conversation your agent handled.
Search it by keyword, or filter by topic, by the contact who started it, or by status — unresolved, resolved, read or unread, rated or not rated. Each record shows the user's messages, the agent's responses, any actions it triggered, and whether it ended in a resolution or a handoff.
Can I export the analytics, and what does it cost?
You can export the dashboard for the selected range as a PDF or a CSV, delivered as a link by email.
Analytics is not a paid add-on and it is not priced per seat: every plan is priced by credits for the AI work the agent does, and the free plan needs no card. Current prices and credit allowances are on the pricing page.
Keep reading
The AI customer service agent
Reads your live docs, answers with the source, takes the action mid-chat, and hands the rest to a person.
Customer service metrics: the eight worth tracking
What each one measures, the pair it must be read with, and how each gets gamed.
Support metrics glossary
One-sentence definitions with the formula, a worked example and what each number hides.
Chatbot to human handoff
The full transcript and an AI summary go with the conversation; the customer never repeats themselves.
Knowledge base chatbot
Reads Helpjuice, Zendesk, Confluence, Drive, SharePoint, Slack and your website in place, kept in sync, with the source shown.
AI chatbot for your website, inbox, Slack and Teams
One snippet installs the widget; the same agent answers email, Slack, Microsoft Teams, Google Chat and your app via API.