Understand dashboard KPIs

What it is

Your analytics live on the dashboard home — the first page you see after signing in. There is no separate Analytics section to find. Six summary cards sit across the top, a set of charts sits underneath, and a date-range control switches all of them at once.

The six cards:

Card	What it counts
Total Conversations	"Conversations across all your chatbots in the selected period."
Messages	"Total messages exchanged in the selected period."
Successful response rate	"Share of messages the bot answered successfully — answered with high confidence or rated helpful by the visitor."
Avg Response Time	"Average time the bot took to reply in the selected period."
Resolution Rate	"Share of conversations where the bot answered every question asked."
Handoff Rate	Share of conversations that were escalated to a person on your team

Each card also shows a change against the previous window of the same length — pick 30 days, and the comparison is the 30 days before that. When there is no earlier data to compare against, it says so rather than inventing a number.

When you would use it

Weekly, to see whether the chatbot is getting better or worse. Before a review, to have a defensible figure. After a launch, to check that traffic is actually arriving.

Not for judging a single conversation. For that, read the transcript — see Read a conversation.

Where to find it

Dashboard, the home item in the sidebar.

The date-range control only appears once you have at least one chatbot. Before that, the page shows onboarding guidance instead of empty charts.

Steps

1. Open Dashboard.
2. Pick a range: 7 days, 30 days, or 90 days. Everything on the page follows it.
3. Read the six cards top to bottom. Hover any of them for the hint that defines it.
4. Look at the change indicator on each. A number without a direction is a snapshot; a number with a direction is a trend.
5. Scroll to Quality and performance for the answer-outcome charts, then Breakdowns for Top Chatbots.
6. When a figure looks wrong, check three things before assuming a bug: the range, your timezone, and whether a chatbot filter is applied.
7. If everything reads zero, work through Analytics shows no data yet.

The definitions that actually matter

A conversation must contain at least one message

A visitor who opened the chat and never typed anything is not counted. This is deliberate: an empty session would otherwise draw a phantom bump on the chart every time someone clicked the bubble out of curiosity.

It also means your conversation count is lower than your widget-open count, and that is correct.

"Successful" has a precise definition

A message counts as a successful response when the chatbot answered it and either the answer was high-confidence, or the visitor gave it a thumbs-up. An explicit thumbs-up always counts, regardless of confidence — a real human saying "this was useful" outranks the chatbot's own estimate.

The same definition is used everywhere the figure appears, so the dashboard and a chatbot's own page will never disagree about what "successful" means.

The important caveat: a confident wrong answer still counts as successful. This metric measures whether the chatbot produced a real answer, not whether the answer was true. High success with unhappy customers is a real and common pattern — when you see it, go and read transcripts.

Resolution Rate is per conversation, not per message

It is the share of conversations where every question got an answer. One unanswered question in a ten-message conversation makes the whole conversation unresolved.

That makes it stricter than the success rate, and more useful as a "did we actually help?" figure. A high success rate with a low resolution rate means the chatbot handles most questions but reliably falls over on one kind — usually a topic missing from its knowledge.

Handoff Rate is not scored

The share of conversations escalated to a person. The dashboard deliberately shows no up-or-down arrow on this card, because there is no universally good direction. A rising handoff rate can mean the chatbot has got worse, or that you have just configured it to escalate properly. Read a few of those conversations before you conclude anything.

Charts underneath

• Conversations & Messages — both series over time, so you can see whether volume or depth is changing.
• Answer Quality — every message split into Answered, Partial answer, and No answer. A growing "No answer" slice is your knowledge gap, in one picture.
• Answer outcomes over time — the same three categories as a daily trend.
• User Feedback — Positive against Negative thumbs. Empty until visitors actually rate something: "Thumbs up/down from your visitors will appear here."
• Avg Response Time — daily, so you can spot a slow day rather than an average that hides one.
• Top Chatbots — your five busiest by message volume, each with its own success rate. This is how you find the one chatbot dragging the average down.

Limits and plan notes

Analytics is not a paid feature. Every plan gets the same dashboard. What you need is traffic, not an upgrade.

Test chat counts. Messages you send yourself from the owner's test drawer are included in every figure on this page. A long afternoon of testing measurably moves your numbers, especially on a quiet chatbot — see Analytics shows no data yet for how to keep launch-week figures honest.

Figures are cached for about a minute, so a message you just sent may take a moment to appear. Editing a chatbot clears the cache immediately.

Team members see only what they can reach. A member limited to selected chatbots gets analytics for those chatbots only, so their totals will legitimately differ from the owner's.

Costs are not shown. The dashboard reports usage and quality. Spending is on Settings → Billing — see How message credits work.

Common problems

A chat I just sent is not showing.

Wait a moment and refresh. Figures are cached briefly. If it is still missing after a minute, confirm the range covers the right day in your timezone — see Why your dashboard days start when they do.

My success rate is high but customers complain.

The metric says the chatbot answered, not that it was right. Open Conversations, sort to the thumbs-down and no-answer turns, and read them. The fix is nearly always knowledge or instructions rather than a setting.

Resolution Rate lags the other cards slightly.

It is calculated from per-conversation counters that finish a few seconds after the message itself is recorded. Refresh once.

A teammate's numbers do not match mine.

Either they are limited to a subset of chatbots, or you are looking at different ranges — or different timezones. Compare all three before assuming the data is wrong.

Handoff Rate went up and nothing is highlighted.

Correct. The dashboard does not judge that number for you. Read the escalated conversations and decide for yourself.