Technical Task: Implementation of AI Assistant in Bitrix24 Open Lines

1. General Goal

Automation of initial communication with clients in messengers and social networks (Telegram, WhatsApp, Viber, Facebook, etc.) using an AI agent. The bot should consult, "warm up" the lead, handle objections, fill in data in CRM, and assign tasks to managers.

2. Technological Stack

• CRM: Bitrix24 (Cloud or Box version with activated Marketplace).

• Automation Platform: n8n (installed on a private server for data security).

• AI Model: GPT or Gemini Pro new versions (via OpenRouter or direct APIs).

• Knowledge Base (RAG): Vector storage (e.g., Pinecone or PostgreSQL with pgvector) for uploading instructions, price lists, and catalogs.

3. Functional Requirements

3.1. "First Contact" Mode

• The bot automatically engages in dialogue upon a new inquiry in the Open Line.

• Identification: The bot recognizes the client, creates a Deal/Lead in Bitrix24, and links the correspondence history.

• Flexibility: Support for both text and voice messages (voice-to-text conversion for analysis).

3.2. Consulting and RAG System

• The bot must answer questions based exclusively on uploaded documents (PDF, Excel, TXT) to avoid hallucinations.

• Example: a client asks about the availability or price of a service—the bot queries the table (via SQL or vector search) and provides the current price and description of the service.

3.3. Warming Up and Handling Objections

• Scenario: The bot does not just respond but engages in dialogue, asking clarifying questions ("should I schedule you, when do you plan, etc. (we will describe in more detail).

• Handling Objections: If the client says "too expensive" or "I'll think about it," the bot uses the scripts embedded in the prompt to address these objections.

3.4. CRM Automation (Field Filling)

• During the dialogue, the AI should recognize entities (name, phone number, budget, type of service) and automatically fill in the corresponding fields in the Deal and contact card in Bitrix24.

• The bot should independently move the Deal to the next stage of the funnel (e.g., from "New" to "Qualified").

3.5. "Reminder" Function (Follow-up)

• If the client stops the dialogue, the bot should send a reminder message after a set time (e.g., 2 hours or 1 day) to clarify the relevance of the request.

3.6. Handing Over the Dialogue to the Manager

• As soon as the client is ready to purchase or requests a "live" operator, the bot:

  1. Disconnects from the dialogue (automatic bot disconnection function when intercepted by an operator).

  2. Creates a Task in Bitrix24 for the responsible manager with a brief summary of the dialogue and a list of further actions.

4. Implementation Algorithm (Step-by-Step)

1. Setting up n8n: Creating a Workflow with a Webhook node to receive messages from Bitrix24.

2. Integration with Bitrix24:

   • Creating a local chatbot via REST API.

   • Setting access rights (Users, CRM, Chat, and Telephony).

3. Connecting Channels: Connecting the Telegram bot via BotFather to Bitrix24 Open Lines. - Channels are already connected

4. Developing the Prompt: Writing a system instruction (System Prompt) that defines the role of the bot (e.g., "You are the expert consultant of company X"). - the prompt has also been developed and the bot trained for operation

5. Setting up Memory: Implementing a Memory node (Simple or Postgres) to retain the context of the dialogue (so the bot remembers the client's name).

5. Expected Results

• Instant response to the client 24/7.

• Relieving the routine workload from the sales department at the qualification stage.

• Fully filled CRM card even before the first call from the manager.