Loading...
Every chatbot metric explained plainly — plus what to do about it.
Your bot keeps score of how it's doing, and you can see the whole scoreboard on the bot's dashboard/analytics page. Don't worry if numbers make your eyes glaze over — this guide explains every metric in one plain sentence, tells you what a good or bad number looks like, and says what to do about it.
Open your bot and go to its dashboard (analytics). Everything below lives there. You can switch the date range between 7 days, 30 days, and 90 days (30 days is the default).
What it is: how many separate chats your bot had. Good or bad: higher just means more people are using it — there's no "bad" number, it's a volume gauge. What to do: if it's near zero, check that the widget is actually installed and visible on your site (see the install guide).
What it is: the total number of messages sent across all conversations (bot + visitors). Good or bad: it's a volume number; read it alongside conversations, not on its own. What to do: nothing directly — it feeds the next metric.
What it is: on average, how many back-and-forth messages each chat takes. Good or bad: a low number can mean the bot answers fast; a very high number can mean visitors are struggling to get an answer. What to do: if it's climbing, your bot may be confusing people — improve its training so it answers in fewer turns.
What it is: the percentage of chats the bot finished without needing a human. Good or bad: higher is better — it means the bot is genuinely helping. A low rate means people keep needing a person. What to do: to raise it, add more knowledge and better answers in Training.
What it is: the percentage of chats the bot handed off to a human. Good or bad: the mirror image of resolution rate — lower is usually better, but some handoffs (sales, refunds) are healthy on purpose. What to do: if it's too high, tighten your training or loosen a too-strict confidence threshold in Guardrails; if it's too low and answers are wrong, you may want more handoffs.
What it is: how visitors felt, split into positive / neutral / negative / frustrated. Good or bad: lots of positive and neutral is great; a growing "frustrated" slice is a warning sign. What to do: if "frustrated" is rising, read those chats to find what's tripping people up, then fix the answers.
What it is: the questions people ask most often. Good or bad: not good or bad — it's a to-do list. What to do: make sure your bot answers each top question perfectly; these are your highest-impact fixes and great candidates for Quick Topics.
What it is: whether your bot's knowledge is up to date and trained. Good or bad: "trained/ready" is what you want; "needs training" means recent changes aren't live yet. What to do: if it says anything is pending, open Training and finish or re-run it.
Check these once a week, fix your top questions first, and your numbers will improve on their own.
Tags
Training Your Chatbot
Train your chatbot with documents, URLs, FAQs, and conversation review.
Handing Off to a Human Agent
When and how your chatbot passes a conversation to a real teammate.
Guardrails: Confidence, Limits & Requiring an Email
Safety rules made simple: confidence, message/credit limits, filters, and the email gate.
Chatbot Overview
Deploy AI-powered chatbots on your website, WhatsApp, or internally.