Home · Blog · Chatbot metrics: 8 numbers that show whether the bot is worth it
Chatbot metrics: 8 numbers that show whether the bot is worth it
The bot is running, messages are coming in — but is it working? “Lots of conversations” says nothing: a bot can answer briskly and bring in not a single request. Below are eight numbers that answer the question “is the bot worth it”, where to find each of them in CHBT and how to count what the ready-made reports do not show.
First — what counts as a result
Metrics make sense only relative to the bot's goal. Before looking at the numbers, answer in one sentence: the bot is needed to… collect requests with a phone number, book people for a service, answer common questions instead of an operator, pick products and send people to a marketplace. The answer decides which of the numbers below is the main one and which are supporting.
1. Incoming messages and their trend
How much people write to the bot and whether it is growing. In CHBT — “Analytics” → “Review”, the “Messages by day” chart for 14 days: incoming messages from customers, bot replies (including AI) and operator replies. Below the chart — this week against last week. A sharp drop in incoming messages is a reason to check the channel (whether the token broke) and the traffic sources.
2. Source: where people came from
The “By channels” block splits conversations for 30 days between Telegram, MAX, VK and the website chat. Within a channel the source comes from a label: links and QR codes from “QR campaigns” count Telegram bot starts by each label. How to set this up — in the article on the bot link with a start parameter.
3. Share of people who reached the goal
The main metric of almost any bot: of those who started a conversation, how many left a phone number, booked or bought. CHBT has no ready-made counter for flow blocks — and this is solved with tags. At the result point place an “Action” → “Add tag”, for example left-phone. Then:
- in “Contacts” a filter by the tag shows how many people got there (the list shows up to 500 records);
- “Export contacts” gives a CSV with tags — for an exact count and your own spreadsheet.
Share = contacts with the result tag / all new contacts for the same period.
4. Where people drop off
The same technique, just with more tags: one at each important step — step-service, step-time, step-phone. The difference between neighboring tags is the drop-off. If 80 people reach the service choice and 15 reach the phone step, the problem is in the step between them: too many questions, an unclear button or asking for the number too early. Typical causes — in the article “10 mistakes in bot flows”.
5. Operator response time
When the bot hands a conversation to a person, the customer waits. The “Operators” block for 30 days shows the median wait for a reply, the share of replies within 5 minutes and the ratings given by website visitors — for the bot and for each employee. The “Waiting for an answer” tile on the “Review” tab is a link: one click opens the list of conversations currently waiting for a person.
6. The load the bot took off
If the bot is meant to replace an operator, look at the ratio on the chart: bot and AI replies against operator replies. If the bot's share grows with the same incoming volume, the bot is taking over the routine. If the operator's share grows, the bot is not coping, and it is worth looking at which questions make it hand the conversation over.
7. Product selection funnel
For a sales bot CHBT has a “Sales” tab for 7, 30 or 90 days: how many buyers started the selection, for how many the bot completed it, who was shown a card with a price and who went to the marketplace. The “Where does the selection stop?” block hints at what a drop at a stage usually means: the card was shown but nobody clicked through — more often a matter of price or trust. The tab is filled by the AI sales assistant.
An honest limit: the buyer pays on the marketplace or in your store, and these orders are not matched with the bot's conversations. The funnel ends with the move to the marketplace.
8. Broadcasts: delivery and refusals
The “Broadcasts” card on the “Review” tab shows how many broadcast messages have been delivered in total. Each broadcast shows delivered, errors and “no channel”. Errors growing from broadcast to broadcast are a sign that people are blocking the bot. What stands behind each number — in the breakdown of why a broadcast is not delivered.
Example: a salon booking bot over a month
A salon sets the goal “booking with a phone number” and tags at the steps: step-service, step-time, booked.
- New contacts for the month — 420, of which 60% came from Telegram, 25% from MAX and the rest from the website chat (“By channels”).
step-service— 300,step-time— 210,booked— 95.- Share of people who booked — 95 out of 420, about 23%.
- The biggest drop-off is between choosing a time and booking: 210 → 95.
The conclusion is not “the bot is bad” but concrete: people leave at the phone step. The salon removes the email question, moves the phone request to after the time is confirmed — and a month later compares the same drop-off.
The numbers in the example are illustrative: this is what an analysis looks like, not an industry norm.
Mistakes that distort metrics
- The tag was placed in the wrong spot. If “booked” stands before “Data collection”, it also marks those who left without giving a phone number.
- Different periods are compared. A week with a broadcast and a week without one are different weeks.
- Test runs are counted. In CHBT the test chat does not get into the “Sales” funnel, but if you tested the bot from your phone like an ordinary customer, your tags will be in “Contacts” — exclude yourself from the count.
- The flow is changed and “before and after” is compared without a date. Write down when you changed what.
Metrics for different types of bots
You do not need all eight numbers at once. Which metric is the main one depends on the task:
| Bot | Main metric | Supporting |
|---|---|---|
| Collecting requests | Share who left a phone number | Sources, drop-offs by step, operator response time |
| Booking a service | Share who booked | Drop-off at choosing a time, share of cancellations (by tag) |
| Support and answering questions | Share of conversations without an operator | Operator response time, website visitor ratings |
| Product selection | Moves to the marketplace | Where the selection stops, assortment problems |
| Lead magnet and subscription | Received the material (tag) | Sources, broadcast delivery, share of errors |
Choose one main metric and look at it every week. The others are there to understand why the main one changed.
How to connect bot metrics to money
A business cares not about the share of requests but about how much the bot brought in. CHBT cannot count revenue directly: payment usually goes through a till, a CRM or a marketplace. But you can connect them:
- Deals in the CRM. “Action” → “Transaction in CRM” at the result step. The manager moves the deal to “Paid” — and the CRM shows how many deals from the bot reached money.
- Source tag + export. Export contacts with tags to CSV and match phone numbers with sales from your till or 1C.
- Payment in the bot. If a person pays right in the bot via the “Payment” block, set a tag after the “Paid” exit — then those who paid are visible in “Contacts”.
Divide the revenue from customers who came through the bot by the costs — the subscription and the ads leading to the bot — and you get an honest payback. Without this step any bot metrics remain “supporting”.
How to review metrics: once a week, in one table
| Metric | Where in CHBT |
|---|---|
| Incoming, week over week | “Analytics” → “Review”, chart |
| Channels | “By channels” |
| Sources within Telegram | “QR campaigns”, counters |
| Reached the goal | “Contacts”, filter by result tag |
| Drop-offs by step | step tags, “Export contacts” |
| Operator response time | “Operators” |
| Product selection | “Sales” tab |
| Broadcasts | “Broadcasts” |
The “Analytics” section is open on all plans and to all project members. Test runs from the editor do not get into the funnel, so testing a flow does not spoil the numbers. More about the section — on the chatbot analytics page.
Chatbot analyticsHow a bot affects conversionTags as a segmentation systemCRM and contacts