Chatbot analytics: KPIs, filters and conversion explained
Read every analytics card correctly, reproduce the calculation and turn one signal into a useful next action.
Chatbot analytics become useful when you read volume, answer quality and business outcomes together instead of treating one KPI as the full story.
Analytics can look intimidating because eight cards, two filters and two visualizations appear at once. This guide starts with the questions a beginner actually asks: Which visitors are included, what does each number count, and what should I do next?
The important rule is simple: never judge a single card in isolation. First lock the chatbot and date range, then connect activity, answer quality and business outcomes.
Privacy-protected two-click player
Chatbot Analytics Tutorial: KPIs, Filters & Conversion Rate
Learn chatbot analytics with real data: sessions, messages, answers with sources, unanswered questions, leads, bookings, conversion rate and trends.
The YouTube player stays blocked until you choose Play. Loading it connects your browser to YouTube and may transfer technical data to Google.
Open directly on YouTubeWhat you will have at the end
- A plain-English glossary for all eight KPI cards
- A repeatable chatbot and date-range comparison
- A verified conversion-rate calculation
- A drill-down from a KPI to the underlying records
Before you start
- Analytics access on a paid plan or active trial
- At least one chatbot; real visitor activity is recommended
- A business question such as “Did the new FAQ reduce unanswered questions?”
Activity, quality and outcome are different layers
Sessions and messages describe activity. Answers with sources and unanswered questions describe how knowledge was used. Leads, bookings and conversion describe outcomes.
The filters define who and when. The daily table shows the exact values behind the chart. Related Conversations, Questions, Leads and Bookings explain why a number changed.
01–10
Set it up step by step
Open Analytics and read the screen from top to bottom
Start with scope, then summary cards, then daily evidence.
Open Dashboard → Analytics. The top card controls the chatbot and date range. The eight cards summarize the selected data. Daily time series at the bottom contains the exact day-by-day values. If Analytics is locked, confirm that the account has a paid plan or an active trial.
- Do not interpret the cards before the filters are correct.
- Refresh reloads both the KPI cards and the daily series.
Choose one chatbot or the whole account
The chatbot filter defines which assistants contribute data.
Choose All for an account-wide operational overview. Choose one assistant when you want to diagnose its content, campaign or conversion flow. In the example, Analytics Lab isolates only the events created for the tutorial, so other assistants cannot distort the totals.
- Write the selected chatbot into every report or export.
- Do not compare assistants with completely different jobs as if they had the same goal.
Set a complete date range and refresh
From and To are inclusive calendar dates.
Set From and To to match the question you are answering, then select Refresh. For a launch review, use the same number of complete days before and after launch; for weekly monitoring, compare complete Monday-to-Sunday windows. Never compare a full seven-day week with the first two days of the current week.
- Record timezone and campaign changes outside the dashboard.
- A zero may simply mean the event lies outside the selected dates.
Read Sessions, User Messages and Assistant Messages together
These three cards describe volume and conversation depth.
A Session is one distinct visitor chat session in the selected scope. User Messages counts visitor turns; Assistant Messages counts bot replies. Divide user messages by sessions for a simple depth check. In the verified example, 5 user messages across 2 sessions equals 2.5 visitor messages per session. That shows depth, not satisfaction.
- A 1:1 user-to-assistant count is common, but not a quality guarantee.
- A sudden depth increase can mean interest or confusion; inspect conversations.
Interpret Answers with Sources and Unanswered carefully
These cards indicate knowledge use and likely gaps.
Answers with Sources counts assistant replies that used relevant indexed knowledge. Unanswered counts collected questions classified as No Information. A high source count proves that connected content was used, not that every answer was correct. A zero under Unanswered is encouraging only if question collection is enabled and real visitors asked meaningful questions.
- Open Questions to inspect the exact unanswered wording.
- Retest important answers against the cited source after a knowledge change.
Calculate Leads, Bookings and Conversion Rate
WebChatAgent conversion is leads divided by sessions.
Leads counts captured lead records. Bookings counts stored appointments. Conversion Rate is calculated as Leads ÷ Sessions × 100; bookings are not part of this percentage. In the verified example, 1 lead from 2 sessions produces the visible 50.0%. If there are no sessions, the dashboard safely shows 0.0% instead of dividing by zero.
- Judge outcomes against the assistant’s purpose: support bots may legitimately have few leads.
- Always show counts beside percentages; 50% from two sessions is fragile.
Use the daily table to verify totals and spot the exact day
Each row uses the same definitions as the cards.
Keep the Table view when exact values matter. The newest day appears first. Scan across one row to see whether sessions, knowledge usage and outcomes moved together. Add the rows across the selected range and they should reconcile with the headline cards, apart from no hidden rounding because the underlying counts are integers.
- Hover the column icons to confirm which metric you are reading.
- A single spike is a lead for investigation, not proof of a trend.
Switch to Chart for patterns, then return to Table for proof
The chart makes direction visible; the table preserves precision.
Select Chart to compare the shape of activity, knowledge and outcome series over time. Hide noisy lines when necessary and look for repeated direction across several days. Return to Table before quoting a number, because overlapping lines and chart scaling can make small changes look larger than they are.
- Use the chart to form a question; use the table and records to answer it.
- Keep the same filters while switching views.
Compare like with like and write down the hypothesis first
Change one dimension at a time: chatbot or dates, never both.
For a before-and-after test, keep the chatbot fixed and change only the date range. For an assistant comparison, keep the dates fixed and change only the chatbot. Match traffic source, weekday mix, assistant purpose and major campaigns as closely as possible. State the hypothesis before looking at the result, for example: “The new returns article should reduce Unanswered without reducing Sessions.”
- Avoid cherry-picking the best-looking date window.
- Record launches, source re-indexing and model changes as annotations.
Open the underlying records before changing anything
Aggregates tell you where to look; records explain what happened.
Use Conversations for session and message changes, Questions for No Information entries, Leads for captured contacts and Bookings for appointments. Read a representative sample and protect visitor privacy. Only then choose one reversible change, such as improving one source article, clarifying one prompt instruction or shortening one lead form.
- Inspect at least ten records when volume allows.
- Measure again over a comparable complete window.
Example & result
See the practical test and its result
Every tutorial includes a fixed input, the expected outcome and a transparent record of what was actually verified locally.
Practical example: Chatbot analytics explained: every KPI, filter and chart
This exact scenario was completed with the temporary tutorial account.
Exact test input
Select “Analytics Lab”, use one complete date range, refresh, then reconcile Sessions, User Messages, Assistant Messages and the daily table.
Expected result
The chatbot filter, headline cards and daily rows use the same scope; Conversion equals Leads divided by Sessions.
What was actually verified
The real run showed 2 sessions, 5 user messages, 5 assistant messages, 5 answers with sources, 1 No Information question and 1 lead. The cards and daily table matched; 1 ÷ 2 produced 50.0% Conversion Rate, while Bookings correctly stayed at 0.
Tips & tricks
Make the setup reliable
Test with realistic examples, record your baseline and change one setting at a time. That makes real improvements visible.
Add annotations outside the dashboard
Record launches, campaigns, model changes and source re-indexing so metric shifts have context.
Prefer rates plus counts
A high conversion rate from three sessions is not equivalent to the same rate from three thousand.
When something does not work
Troubleshooting
Check status, permissions and test data systematically before changing the model or prompt.
The expected option is missing
Confirm the account plan, feature permissions and selected chatbot. Paid or beta features can be hidden when prerequisites are not met.
The test result is inconsistent
Reset the test conversation, keep the input identical and change one setting at a time so the cause remains measurable.
Ready for a production-style test
Save the chatbot, date range, counts, calculated rate and hypothesis in one short report. Inspect the supporting records, make one reversible change and repeat the same measurement over a complete comparable window.