Building a WhatsApp Lead Qualification Bot with n8n and LLMs

For the agente_seguros_ai project, I built a WhatsApp bot that qualifies insurance leads automatically — 24/7, no human needed unless the lead is hot. Here’s how the architecture works.

Why n8n

n8n is a self-hostable workflow automation tool (think Zapier, but open source and deployable on your own VPS). It handles the WhatsApp webhook, conversation state, and routing logic without writing a single line of code for those parts.

I only write code for the parts that need it: the LLM prompt, the database writes, and the lead scoring.

The Flow

WhatsApp message
    ↓
n8n webhook trigger
    ↓
Load conversation history from PostgreSQL
    ↓
Build prompt with context + user message
    ↓
Groq API (Llama 3.1) generates response
    ↓
Parse structured data (name, phone, interest)
    ↓
Update PostgreSQL + score lead
    ↓
Send reply via WhatsApp Business API
    ↓
If hot lead → notify agent via Telegram

The Prompt Strategy

The system prompt has two modes:

1. Qualification mode: The bot asks 4–5 questions naturally (not as a form) to gather: full name, type of insurance needed, current coverage, budget range.
2. Handoff mode: Once qualified, the bot says a human advisor will contact them and stops generating AI responses.

The trick is injecting the conversation history as context so the model doesn’t repeat questions already answered.

Handling State in PostgreSQL

Each conversation has a state field: qualifying, qualified, handed_off, cold. n8n checks this before deciding whether to call the LLM or just send a static message.

Lessons Learned

• Groq is fast enough for real-time chat — typical latency is 400–800ms including the WhatsApp API round-trip.
• Structured output matters — ask the LLM to return JSON for extracted data, not free text.
• Test edge cases manually — users who answer in voice notes, send images, or write in regional dialect.
• n8n’s error handling is weak — wrap your custom function nodes in try/catch and log to a separate table.