AI chatbot for websites and messengers
We build AI chatbots for websites, Telegram, WhatsApp and VK. The bot answers questions from your knowledge base, clarifies the customer's task, collects contacts and hands the conversation to a manager with the full history. Unlike button-based bots, an AI consultant understands free-form questions and answers from your materials rather than the model's general knowledge.
Where the bot works
A widget on a website on any platform: 1C-Bitrix, WordPress, Tilda and others. A bot in Telegram and WhatsApp through official APIs. The same scenario and knowledge base serve all channels, and the conversation history is kept when a customer moves from the website to a messenger. For Tilda sites we connect the bot as an external service, since the builder does not allow server-side logic.
Answers from a knowledge base, not guesses
We assemble a knowledge base from website pages, price lists, regulations, managers' replies and documents. The bot finds relevant fragments and answers from them; when there is no answer in the base it does not invent one but offers to contact an employee. Before launch we check answers on a test set of real customer questions, and after launch we review the log regularly and extend the base.
Leads and hand-over to a manager
The bot collects enquiry details: the task, contact and preferred channel, then creates a deal in the CRM or sends a Telegram notification to the team. Complex questions, complaints and stop-list topics go to an employee together with the conversation history. Hand-over rules, working hours and message texts are agreed before launch.
What development includes
From scenario and knowledge base to the website widget and integrations.
- Conversation scenario, tone of voice and topic limits
- Knowledge base with document retrieval and updates
- Website widget or messenger bot with brand styling
- Lead capture, CRM integration and team notifications
- Testing on real questions and acceptance
- Conversation log, metrics and a report on enquiry topics
- Documentation and a guide for updating the knowledge base
How the work is organised
-
Step 1
You send a link to the website, materials for the knowledge base and examples of questions customers ask.
-
Step 2
We design the scenario: greeting, clarifying questions, answer and hand-over rules, contact collection.
-
Step 3
We build the bot, fill the knowledge base and check answers on a test set together with your team.
-
Step 4
We connect the widget and messengers, the CRM integration, launch and hand over the update guide.
Cost of an AI chatbot
The cost depends on the number of channels, the size of the knowledge base, scenario complexity, CRM integrations and widget styling requirements. Model usage is estimated separately. To get an estimate, send a link to the website, the list of channels and sample customer questions.
Frequently asked questions
How does an AI bot differ from a regular button bot?
A button bot follows a predefined tree. An AI bot understands a free-form question, finds the answer in the knowledge base and holds a conversation. Both can be combined: buttons for typical scenarios and AI for everything else.
What if a customer asks something off-topic?
The bot is limited to your business: for unrelated questions it politely explains that it only helps with the company's services and offers to contact an employee.
How does the bot hand a conversation to a manager?
By a rule in the scenario: a customer request, a stop-list topic or no answer in the base. The manager receives a notification in the CRM, Telegram or email together with the chat history.
Can the bot connect to amoCRM or Bitrix24?
Yes. The bot creates a deal or contact, writes the conversation history into the card and assigns a task to a manager. For other systems we use their APIs.
Who updates the knowledge base after launch?
Your team following the guide, or we do it as part of support. Updating requires no programming: add a document or page and run re-indexing.
How is model usage charged?
It depends on the number of conversations and the amount of text. Before launch we estimate the cost for the expected load and choose a model that keeps it predictable.