The chatbot Analytics tab

What it is

Analytics is the one place that answers "is this chatbot working?" without you having to check four other tabs. It is the third tab of a chatbot, after Playground and Conversations, and it holds everything numeric about that one chatbot.

It does three jobs:

1. Tells you whether it can answer well, with training status and knowledge size on the Health & configuration card.
2. Shows you what came back — volume, quality, and engagement over the range you pick.
3. Lists the specific questions it handled badly, in the Needs attention panel.

It is a read-and-decide page, not a place you configure anything. Actual settings live on the Settings tab; knowledge lives on Knowledge; the live widget preview lives on Playground.

When you would use it

• Weekly, to see which questions your chatbot is failing and close those gaps.
• After adding knowledge, to confirm the training run finished.
• When a visitor complains and you want to know whether it was a one-off or a pattern.

Where to find it

Open Dashboard → Chatbots → your chatbot → Analytics.

Steps

1. Go to Dashboard → Chatbots, open a chatbot, and select Analytics — the second of seven tabs.
2. Pick a range at the top: 7 days, 30 days, or 90 days. Every card below follows it; the trend arrows compare against the same-length window before it.
3. Scan the performance numbers: total conversations, total messages, successful response rate, and average response time, each with its all-time figure underneath.
4. Read the activity chart — messages and conversations per day for the range.
5. Open Needs attention. This is the most valuable part of the page — see below.
6. Check answer quality: the status split (answered / partial / no answer) and the outcomes-over-time chart.
7. Check Engagement: messages per conversation, positive feedback, negative feedback.
8. Read the Health & configuration card at the bottom. It shows current status, when training last ran, knowledge base size, and how many knowledge collections there are, with a shortcut to Manage knowledge.
9. When you want to try something yourself, use the Test Chatbot button in the chatbot header. It is a drawer, not an eighth tab — see Try Test Chatbot.

Needs attention — the part worth your weekly ten minutes

This panel collects the questions your chatbot handled badly, filtered two ways:

• Unanswered — questions where the chatbot could not find a grounded answer. Each entry shows the confidence it managed. These are gaps in your knowledge, and every one of them is a document you have not uploaded or a page you have not crawled.
• Negative feedback — replies a visitor gave a thumbs-down. These are usually not missing knowledge; they are wrong tone, an incomplete answer, or a visitor who wanted a human rather than a paragraph.

The workflow is: read the question, decide whether it is a knowledge gap or a behaviour gap, then fix it in the right place. Knowledge gaps go to Knowledge overview or a website crawl. Behaviour gaps go to Write instructions, or to Handoff to a human if what they actually wanted was a person.

When both lists are empty, the panel says so — which on a busy chatbot is genuinely good news, not a loading failure.

Training banners

Two banners appear on the health card when knowledge is mid-change, and they mean different things:

• Knowledge updated, re-sync in progress. The chatbot keeps answering with the previous content and picks up the new content shortly. No retraining action is needed from you.
• A knowledge update didn't finish. Some answers may be missing the newest content. There is a retry.

Neither requires you to deactivate the chatbot. Leaving it live during a re-sync is the intended behaviour — see Chatbot readiness.

What you will see

The range picker sits at the top. Below it: KPI cards, the activity chart, Needs attention, the two quality charts side by side, Engagement, and Health & configuration. A draft or inactive chatbot shows the activation note on the health card as well as on the Playground tab.

Limits and plan notes

Analytics is read-only, so viewing it needs nothing more than permission to view chatbots — Viewers see the same page without the buttons they cannot use.

Numbers are cached for about a minute, so a conversation that just ended can take a moment to appear. Opening Analytics does not spend message credits. See Message credits.

For numbers across all your chatbots, use the Dashboard home instead — see Dashboard KPIs.

Common problems

Analytics says the chatbot has never been trained.

Add at least one knowledge source and wait for it to finish. The health card updates once the run completes. See Dataset status.

All the numbers are zero.

Nobody has chatted yet, including you. Send a message through Test Chatbot and the counters start filling. See No data yet.

Needs attention will not load.

The panel shows its own error with a retry rather than breaking the page. If it persists, the rest of Analytics is still trustworthy.

The numbers do not match the Dashboard home.

The Dashboard home sums every chatbot in the account; this tab is one chatbot. Compare like with like.