PROCESS

1. AI Prompt Design

The first step was designing conversational prompts that allow the AI agent to communicate naturally instead of following a rigid decision tree.

The prompts enable the assistant to:

• understand user intent and conversation context;

• remember information already provided;

• avoid asking duplicate questions;

• perform BANT qualification;

• calculate a Lead Score;

• generate structured JSON for downstream automation.

2. Workflow Automation

The entire business logic was implemented in n8n using modular workflows.

The automation includes:

• receiving messages from website visitors;

• maintaining conversation memory;

• interacting with AI models;

• creating and updating CRM records;

• analyzing completed conversations;

• generating executive reports automatically.

Each workflow operates independently while exchanging data through the CRM, database and reporting tools.

3. System Integration

The platform integrates multiple business services into a single ecosystem.

Connected components include:

• AI models for qualification and analysis;

• CRM for lead management;

• PostgreSQL for conversation history;

• Google Sheets for analytics storage;

• Telegram for automated notifications and executive reports.

4. Data Flow

The complete process follows an end-to-end automation pipeline:

Visitor

↓

AI Qualification Agent

↓

Lead Scoring (BANT)

↓

CRM Record Creation

↓

Conversation Memory Storage

↓

Daily AI Conversation Analysis

↓

Analytics Database

↓

Weekly Executive Report

↓

Telegram Delivery

5. Continuous Improvement

Conversation analytics are continuously used to improve the qualification prompts.

The system identifies recurring objections, communication gaps and successful conversation patterns, allowing the AI assistant to become more effective over time without changing the workflow architecture