Home · Blog · Push-button, AI or hybrid: how to choose the type of bot for the task
Push-button, AI or hybrid: how to choose the type of bot for the task
The question “what kind of bot is needed - a push-button one or with artificial intelligence” is usually asked from the wrong side. The type of bot is determined not by fashion, but by three things: how many unique questions you have, whether there is text for which the neural network could answer, and who sorts out errors. Below are three types of bots based on one characteristic, their cost of ownership at specific rates, a diagram for assembling each in CHBT and a checklist that tells you in five minutes where to start. Most often - from buttons, and AI is added in the third week.
One sign instead of ten classifications
Bots are divided into push-button, text, voice, “smart”, scripted, dialog - and this list is of little use, because the classifications mix up the channel, interface and logic. To select, one question is enough: Who decides what to answer to the client?
- If the answer was written in advance by the author of the script and the client selects it with a button or keyword, this is pushbutton (scripted) bot.
- If the answer is formulated by a neural network every time, based on your knowledge base, this is II- bot.
- If the script itself determines where to answer with buttons and where to give the question to the neural network, and in both cases it knows how to call a person - this is hybrid.
Everything else is details. There are no voice bots here: CHBT works with text in Telegram, MAX, VKontakte and widget on the site, and voice is a separate industry with telephony and other budgets. There will also be no “bot on a specific neural network” as a separate type: in CHBT, the model - GPT-5, Claude, Gemini, DeepSeek or Qwen - is selected in the agent card, this is an agent setting, not a bot type.
Within one project in CHBT all three types are assembled from the same blocks. The difference is whether the “AI agent” block is in the script and what happens when the client writes something that is not in the menu.
Push-button bot: predictability at the price of zero
A button bot is a tree: a greeting, a menu of buttons, each with an answer and the next menu or question. The client will never receive an answer that you did not write. This is both the main plus and the main minus.
Where is he the best? Recording and requests: the path “service → time → phone” does not require understanding free text. Qualification: three multiple choice questions. Cafe menu, selection of real estate according to parameters, distribution of materials, registration. Anything where there are a finite number of questions, and the clients’ wording is predictable.
Where does it break down? To the question “what if I’m allergic to paint, what do you recommend?” There is no such item in the menu, and the flow has two outputs: a safety trigger “Any message” with the text “I don’t understand, select a menu item” (bad: the person has already asked and is waiting for an answer) or the same trigger with the “Operator” block (normal: the person will answer, the bot will be silent in this conversation).
Cost. Zero over subscription: no credits, no response fees. It works on the Free plan in the “Steps” mode - up to 30 conversations per month, and on the Basic plan without a limit. Time is the only resource: two to three hours to assemble and 20 minutes a week to read the conversations to understand which questions are not on the menu.
How to assemble in CHBT. Trigger “/start” → “Message” with 4–6 buttons → each button has its own branch from “Messages” and, where necessary, “Data Collection” → at the end of each branch there is a “To the menu” button or an “Operator” block. The second flow with the trigger “Any message” → “Operator”. Ready-made frames - templates "Autoresponder FAQ", “Customer registration”, “Collection of applications”.
AI bot: free questions, knowledge base and answer price
The AI bot answers any question, formulated in any way you like, but only within the limits of what you gave it to read. In CHBT, an agent is created in the “AI Agents” section: name, role and character (instructions like “you are a salon consultant, you answer briefly, you name prices from the database, you don’t make things up”), a knowledge base - free text of up to 15,000 characters (it goes entirely to the model with each question, so only what is needed), choice of model and “temperature”. In the flow, the agent is activated by the “AI Agent” block: from this point it conducts a conversation itself, taking into account up to the last ten messages, until the transmission to the operator is triggered.
Pros. The client writes “how much will coloring for long hair with tinting cost” - and receives an answer from the price list, and not “select a menu item.” One agent closes hundreds of statements that would require hundreds of branches in a push-button bot. The database is edited with text in a minute - no need to rebuild the script.
Disadvantages that you need to know about before purchasing a plan.
- Need business materials. The database can be assembled from text, a link page and PDF, DOCX, TXT, MD, CSV files. Only the text is taken from PDF and DOCX, scans without a text layer are not recognized; the link loads one page, not the entire site.
- Each answer costs credits according to the price of the model: one for the default model, more expensive for powerful models. Credits 150 per month for Basic (650 ₽), 500 for Business (1290 ₽). A four-question conversation on the default model is four credits. When exhausted, the agent responds with a neutral phrase, and the script goes along the fallback branch - usually to the operator.
- The neural network sometimes makes mistakes, especially where the base is silent. The cure is escalation in the flow: the “did not understand” port of the “AI agent” block (triggered when the agent’s response sounds uncertain or like a refusal) leads to “Operator”, and the words “operator”, “person” are caught by a separate flow with the “Keyword” trigger.
- There is no list of “what the agent could not answer” in CHBT - you will have to read the conversations and add to the database manually.
How to assemble. Section “AI agents” → create an agent → insert a database → check the agent in the test chat directly in the account → in the script: trigger “/start” → “Message” with a greeting → block “AI agent” → “Operator” on the escalation branch. This is the whole flow: three blocks. How does a knowledge base answer work and what does its price consist of? in a detailed analysis of the AI bot.
Hybrid: buttons for the standard one, agent for the rest
A hybrid is not a “push-button bot plus AI somewhere nearby,” but a conscious division: the typical one goes through the buttons for free and predictably, the atypical one goes to an agent for a loan, and the one that is incomprehensible to the agent goes to a person. The diagram in CHBT on Canvas looks like this.
- flow 1, "/start" trigger. A greeting and a menu of 4–6 buttons on the most common topics: “Prices”, “Make an appointment”, “How to get there”, “Ask a question”. The “Prices” and “How to get there” branches are regular “Messages”. The “Sign up” branch - three blocks “Data collection” and the “Notify operator” action. “Ask a Question” thread – “AI Agent” block.
- flow 2, “Any message” trigger. Catches those who did not press the button, but immediately wrote. The first block is “Condition”: if the time of day is working - “AI agent”, otherwise - “Message” “we will answer in the morning” and “Operator”, so that the conversation is waiting for the employee in the inbox.
- Logging out of the agent. In the “AI Agent” block, the “did not understand” port is enabled - it is triggered when the agent is not sure of the answer - and leads to the “Operator”; the words “operator”, “person”, “manager” are caught in a separate script with the “Keyword” trigger. The conversation goes to the “Conversations” section, the bot becomes silent.
- Marking. Before “AI Agent” - “Action” → “Add Tag”
ai-question. A month later, using the filter in “Contacts,” you can see how many people have reached the agent, and whether it’s worth adding a button to the menu for them.
Hybrid economics using an example: let buttons close 80% of calls, and the conversation with an agent takes three answers. Then 150 credits of the Basic plan will be enough for about 50 conversations with an agent per month - that is, for a bot with about 250 calls. For a Business with 500 credits, it’s about three times as much. This is the answer to the question “why have buttons if there is AI”: buttons save credits and remove the risk of error where error is unacceptable - in prices and addresses.
Comparison by seven parameters
| Parameter | Pushbutton | II- bot | Hybrid |
|---|---|---|---|
| Minimum plan in CHBT | Free, 0 ₽ | Basic, 650 ₽ | Basic, 650 ₽ |
| Payment for answer | No | From 1 credit per agent response, at the price of the model | Only for answers agent |
| Time to launch | 2–3 hours from template | Evening, if there is a base in the text; 2-3 pm if not | Two pm |
| Predictability answers | Full | Depends on the base and model | Full in the menu, depends - in free questions |
| Reaction to a question outside the script | “I don’t understand” or operator | Base response or operator | Agent, then operator |
| Support | Editing texts and branches | Addition of the database based on the results of conversations | Both and other |
| Risk | Losing a client with a non-standard question | Incorrect answer about price or conditions | Minimum if correct division |
The “risk” line is more important than it seems. The button bot makes the mistake of being “silent” - this is noticeable and correctable. The AI bot makes the mistake of a “confident answer” - the client will come to you with it as a promise. Therefore, prices, terms, warranty conditions in a hybrid are included in buttons and text, and consultations are left to the agent, where variations in wording are acceptable.
Cost of ownership per year: three calculations
We calculate at real rates, without the cost of your time - you will calculate it yourself by the number of hours.
Push-button recording bot for the master. Up to 30 conversations per month - Free plan, 0 ₽ per year. Limitations: one operator, “Steps” mode, signature “CHBT” in the widget, if it is on the site. As soon as there are more than 30 entries per month - Basic: 650 × 12 = 7,800 ₽ per year. This is the price at which one saved post per month pays for the subscription.
AI consultant for online school. Basic: 7,800 ₽ per year plus 150 credits per month - 150 agent responses to default models. If the average call is four responses, that's 37 conversations with an agent per month; In addition, the agent responds with a neutral text and passes it on to the operator. If there are 100 or more inquiries with questions - Business: 1290 × 12 = 15,480 ₽ per year and 500 credits - 500 answers to the default model, that is, about 125 conversations with the agent. Plus two hours to prepare the base and an hour per month to supplement it.
Hybrid for an online store. Basic, 7,800 ₽ per year. Buttons close shipping and returns; the “Where is my order” button leads to the operator or to an “HTTP request” in your accounting system (the bot itself does not know the status; with Wildberries and Ozon it does not receive it); agent - questions about products according to the description from the database. 150 credits with three answers to the conversation - 50 conversations with the agent; the remaining 200–300 hits per month are made via buttons for free. The transition to Business is needed when the share of free questions is growing - and this can be seen from the tag ai-question from the previous section.
In all three cases, the cost does not depend on the number of channels: connecting Telegram, MAX, VK and a widget to one bot is already included in the price. The trial day of the Business plan is included in the “Payment” section - it is enough to test the agent on a real basis. Current conditions are on the page “Plans” page.
Five questions that choose your type for you
- How many different questions do clients ask? Take the 50 most recent requests. If 40 of them fit into 5-7 topics - buttons. If there are 20 topics and the wording is not repeated - agent.
- Do you have a text to respond to? Price list, description of services, FAQ in the document. Yes - an agent is available today. No - start with buttons and write the base at the same time: it will be useful in any case.
- How expensive is a mistake in an answer? Medicine, legal terms, financial conditions - just buttons and pre-verified text. Consultation on product selection - an agent is acceptable.
- Who reads “Conversations” and when? If there is an operator on duty, either type works. If not, the agent at least responds to the point while you are away, and a push-button bot without the required menu item will leave the person with “I don’t understand.”
- What plan are you willing to pay from day one? Zero - push-button on Free. 650 ₽ - all three options are open.
If the answers differ - for example, there is a lot, but a mistake is expensive - this is a hybrid: expensive in buttons, the rest to the agent. If they go the other way - so much the less, but there is no operator - a push-button bot with an honest “we’ll answer in the morning” is better in terms of time than an agent who has nothing to read.
How to change the type without rebuilding the bot
Choosing a type is not a death sentence. In CHBT, a bot of one type turns into another by adding or removing one block, and it is better to do this in this sequence.
- Push-button → hybrid. Go to Basic, create an agent in the “AI Agents”, test it in the test chat, add the “Ask a Question” button with the “AI Agent” block to the menu and put the agent in the “Any Message” flow. Don't touch the menu - it's already working.
- AI bot → hybrid. Look in the “Conversations” to see which questions are repeated, and put them in the buttons in front of the “AI Agent” block. Loans will no longer be spent on “how much does it cost” and “where are you located?”
- Any type → experiment. Make a duplicate of the script, change it, check it using “Check Script”, include it instead of the old one. If you don’t like it, the version history with a rollback will return to the previous one.
Practical advice for your first bot: build a push-button bot on the Free plan, one task at a time, read the conversations for two weeks, and make a decision about the agent based on tags and the number of questions that reach the operator. How to design the menu itself so that it includes more questions - in the article about the conversation map and branches; what is a bot in general and what task to start with - in basic analysis. There is a template for a hybrid with quick access to humans “Support”. Both diagrams above are assembled in the office in the evening: create a bot and try →
Plans: Free, Basic, BusinessTemplate "Autoresponder FAQ"“Support” templateHow to Design a Bot ScriptChatbot for business: what is it?