Home · Blog · How a bot affects conversion: 6 mechanics and how to measure the effect
How a bot affects conversion: 6 mechanics and how to measure the effect
The bot does not have a magic “increase conversion” button. There are six specific places in the customer journey where the bot changes the outcome: responds before a human, replaces form fields with buttons, filters out non-targeted ones before the call, reminds through “Pause,” returns by sending out a tag, and calls the operator on time. Let’s analyze each mechanic on the CHBT blocks and show how to measure the result - “Analytics”, counter tags, tags in deep links and A/B through the “Random thread” - because without measurement, any of the six will remain an opinion.
Where does conversion occur in a bot?
The client path in the bot consists of four transitions, and each has its own percentage of losses:
- Enter → first response. The person clicked /start, opened the widget or followed the link. The losses here are if the bot is silent or writes a sheet.
- First answer → choice. Pressed the button, answered the question. Losses - if it is not clear what to do next.
- Select → contact or lead. I left my phone number, signed up, created a deal. Losses - if the contact was asked too early or without explanation why.
- Application → transaction. The manager contacted, the client bought. Loss - if they responded within a day or the lead was not targeted.
Six mechanics below are distributed among these transitions: the first two work at the entrance, the third - on the lead, the fourth and fifth - for those who left, the sixth - at the last step. It's useful to decide in advance which transition you have the most leaky, and start there, rather than implementing everything at once.
Mechanics 1. Instant response instead of waiting
A person who wrote to the company’s messenger and did not receive an answer in a couple of minutes usually goes to ask elsewhere. The bot responds in a second at any time - this is the first and most measurable thing it gives.
In CHBT instant response is a script with a trigger “Command /start” for Telegram and MAX, “Open widget” for the site and “Any message” in case the person did not start with a command. In the first block “Message” there is a short greeting and 3-4 buttons; For more details on how to formulate it, see the article about bot's first message.
Disclaimer: an instant response increases conversion only if it is meaningful. "Hello! Wait for the operator to respond” - this is not a response, but the same queue. If typical questions are answered by a script, and complex questions are answered by an AI agent based on your knowledge base (on Basic and higher), you only have to wait for a person in rare cases.
On the Free plan, the script answers within 30 conversations per month: when 30 different people wrote in a month, the conversation with them continues, and instead of a script, the bot responds to each new interlocutor with a service message about the limit. For a bot with advertising traffic, this must be taken into account in advance - advertising brings in new people.
Mechanics 2. Buttons instead of form fields
A form with five fields requires five decisions at once. The bot asks one question at a time, and most responses are button presses rather than typing. This reduces the effort at each step and increases the percentage of people reaching the end.
How it works in CHBT: a “Message” block with buttons - one per option (“Manicure”, “Pedicure”, “Complex”). Clicking leads to its own branch; in Telegram these are inline buttons under the message, in MAX and widget - buttons under the message, in VKontakte - up to ten buttons, two in a row. What cannot be selected with a button (phone, time) is asked by the “Data Collection” block, checking the format - phone, e-mail or number - and asking again if there is an error.
Three rules that are visible in the results:
- no more than 4-5 buttons in one message - then people stop reading the options;
- always leave the “Other / ask a question” button, otherwise those for whom none of the options suited leave silently;
- text input - only where it is impossible without it: phone number, address, time. Every text question loses more people than a button question.
How to correctly build the chain “use → name → phone” is discussed in detail in the article about collecting applications via bot.
Mechanics 3: Qualify before the manager wastes time
The “lead → deal” conversion depends not only on the manager’s speed, but also on who got into the leads in the first place. If half are not your region, not your budget, or “just asking,” the manager wastes time on them, and responds to the target ones more slowly.
The bot qualifies before contact: two or three questions using buttons (city, type of task, deadline), after which the “Condition” block separates people into branches. The target is “Data collection” on the phone, the “Notify operator” action and the tag target, according to which the manager opens a deal in CRM; non-targeted - an honest answer “we don’t deliver to your city yet” and, for example, a link to a partner. The condition in CHBT tests a tag, variable, channel, time (hour), or day of the week with the operators equal to, contains, greater than/less than, in list; named branches, each with its own output, plus “Else”.
A side effect that often turns out to be the main one: answers to qualifying questions remain in the contact card - in tags or in their own fields. The manager opens the request and sees “Moscow, repairs, urgent” instead of a bare number. The conversation starts straight away with action, and this is also a conversion, although it is not visible in the bot’s reports.
Mechanics 4. Reminder through “Pause”
The man got to the question about the phone and got distracted. In form, this is a lost visit. In the bot - a contact in the messenger, who you can write to in an hour.
Implementation in CHBT - block "Pause" and the “Condition” block. After the question “Data collection”, set a Pause (presets - from a minute to a day, your value - in seconds), then the Condition “variable phone exists? If yes, the script continues with gratitude; if not, one short message: “If it’s more convenient, just write when to call” and a button. The second reminder is a day later, there is no need for a third.
Disclaimers:
- The pause after the deadline continues the script itself, but the bot writes in the messenger only to the person who has already written to the bot - in Telegram and MAX this is a platform rule, and the bot does not bypass it.
- The reminder should not repeat the question word for word: the person has already seen it and did not answer. Suggest another way - “write here”, “select time using button”.
- Reminders will only return some of those who are stuck, and the third or fourth reminder is unlikely to change this, but it may become a reason to block the bot.
The same combination of “Pause + Condition + tag” is the basis of reheating chains; the ready-made sequence “benefit → pause → offer → lead” is collected in a template "Auto Funnel".
Mechanics 5. Return via mailing tag
Some people won’t buy today, but they will buy in a month – if you remind them. The form on the website does not provide such an opportunity: the contact has gone to CRM, and then only a call. The bot saves the interlocutor and permission to write to him.
In CHBT the newsletter comes from the section “Mailmails” (Basic and higher): text with type variables {{name}}, audience selection by tags, delayed launch by date and time, sending through a queue taking into account Telegram limits. In order for the mailing to be targeted, and not “to everyone”, tags must be set in the script - with the “Add tag” action at the moment when a person has shown interest: looked-manicure, asked for price, didn't-leave-phone. The newsletter audience is specified by a list of tags - everyone who has at least one of them will receive it; “all but” cannot be excluded. Therefore, when a person has bought, remove the tag from him asked for price action “Remove tag” - and after two weeks the newsletter for this tag will reach exactly those who were interested and did not buy.
Limitations that affect the result: CHBT broadcasts do not have buttons or media - only text, so the call to action is in words (“write “I want” - and I will write it down”), and this word must be processed by a script with the “Keyword” trigger. How mailing works in a Telegram bot and what restrictions the platform has are on the page broadcasts via Telegram bot.
Mechanics 6. Handover to operator at the right moment
The last transition - from order to transaction - is almost always made by a person. The bot influences it in two ways: it does not interfere and transmits on time.
“Does not interfere” means that the person who wrote “I want to talk to the manager” does not receive a menu in response. In CHBT for this, the block “Operator” is placed on the “Another question” button in each menu and on a separate script with a “Keyword” trigger for the words “operator”, “manager”, “person”. The conversation goes to “Conversations”, the bot becomes silent, operators receive a notification - in the browser, via Web Push or by letter.
“On time” means that the transfer occurs after qualification, and not before. If the bot first collected the city, task and phone number and then transferred it, the operator responds to the completed request. If you said “hello” at the first time, everything is done by a human, and the bot didn’t help in any way. In the “Operator” block there is a field “Message to the operator” - write the context there: “The client chose repair, urgently, the phone is assembled.” The operator will see this before the first response.
Where exactly to draw the line between a bot and a human, what to configure in “Conversations” and how to return the conversation to the bot is the topic of a separate article: where the bot ends and begins person.
How to measure: “Analytics”, counter tags, tags in deep links
Without numbers, six mechanics are a set of hypotheses. CHBT has three measuring tools, and each has its own zone.
"Analytics". The “Overview” tab shows incoming messages by day for 14 days with changes from last week, bot and operator responses on the same graph, conversations over 24 hours, the number of conversations that are waiting for a response, and conversations by channel over 30 days. This is enough for the top level: whether the flow is growing and how much work the operators have. The section does not count how many people left a contact - these are tags and “Contacts”; There are also no counters for script blocks; you cannot see how many people reached the third question. Only the AI salesperson has its own funnel: the “Sales” tab counts customer sessions at the stages “Started selection”, “Collected selection”, “Showed offer”, “Proceeded to the site”.
Tags are like counters. The lack of block-by-block statistics is compensated for by the “Add Tag” action at key points in the flow. Five tags - start, chose-service, asked-phone, left-phone, operator — turn the “Contacts” section with a filter by tag into a funnel: 200 → 140 → 90 → 60 → 15. The difference between adjacent numbers is the transition loss. Export to CSV - to compare weeks.
Source labels in deep links. The link in the Telegram bot looks like https://t.me/ваш_бот?start=payload; payload - any label: yt_sep, vk_post12, qr_box. The “Deeplink” trigger in the script reacts to payload, and the first block is set to “Action” → “Add tag” with the name of the source. Then there’s the same funnel based on tags, but in terms of traffic channels: advertising from VK can get more starts, and QR from packaging can get more phones. In MAX, deep linking works through payload when the bot is launched; on the site, the role of the source is played by the page with the widget.
How to mark a database with tags systematically so that they do not turn into garbage is discussed in the article about tags as a segmentation system.
A/B test via “Random branch”: an honest experiment
When there are two wordings of a question or two orders of blocks, there is no point in arguing - you need to check. There is an action in the “Action” block "Random thread": list ports separated by commas (a, b), and the block has two outputs with equal probability. There are no scales - the division is always equal, so for the 50/50 test two ports are enough, and for 33/33/33 - three.
Experiment scheme:
- Immediately after the “Random thread” in each thread - “Action” → “Add tag”:
test-aandtest-b. - The threads differ in one element: the text of the question, the presence of a benefit before the contact, three buttons versus five. Don't change two parameters at the same time - you won't understand what worked.
- After the point where the branches converge, there is a common result tag, for example
left-phone. - After 1–2 weeks in “Contacts”: share
left-phoneamongtest-aagainst share amongtest-b.
Minimum volume - from 100 people per branch; at 20 per branch, a difference of 3 phones means nothing. If traffic is low, test the grossest distinction (contact before or after benefit), rather than nuanced wording. Leave the winning branch, delete the losing one - the version history of the script will allow you to roll back if something goes wrong.
You can start with one mechanic and one test on the Free plan: 30 conversations per month is enough to check the order of blocks in the first flow; “Random thread”, tags and “Analytics” are available on all Plans, broadcasts - from Basic for 650 ₽. Collect the script and put the first tag →
Template "Auto funnel"Help: conditions and branchesCollecting applications through a botPlans