Summary: A modern chatbot = WhatsApp Business API (channel) + n8n (orchestration) + an LLM with a knowledge base (answers) + handoff to a human agent (hard cases).
Architecture
Customer → WhatsApp Business Platform → n8n webhook → fetch context (CRM/orders) → LLM + knowledge base (RAG) → reply → log to database/CRM → (optional) escalate to an agent
Components
- WhatsApp Business Platform (Cloud API) or an official BSP provider. Avoid unofficial methods because of number-ban risk.
- n8n: receives webhooks, calls the LLM, accesses internal systems.
- LLM + RAG: FAQ, catalog, and policy documents go into a vector store so answers are grounded in your sources.
- Conversation memory: store short history per number.
- Handoff: keywords or a low confidence score → pass to an agent with a summary.
Use Cases
- Answer FAQs (opening hours, shipping fees, return policy).
- Check order status from the database.
- Qualify leads and log them to the CRM.
- Reservations/bookings.
- Payment reminders.
WhatsApp Rules to Watch
Business-initiated messages generally use approved templates, and there are conversation-window and user opt-in rules. Read the official WhatsApp Business policies before launching campaigns.
Best Practices
- Limit the answer domain; the bot declines out-of-scope topics.
- Controlled answer sources (RAG) to reduce made-up answers.
- Never ask for sensitive data (OTP, PIN) in chat.
- Always offer a human option.
- Review regularly: examine failed conversations, update the knowledge base.
- Privacy: disclose AI use and protect data under applicable data protection law.
Metrics
Automatic resolution rate, first response time, escalation rate, CSAT, and cost per conversation.
FAQ
Can it work without the WhatsApp Business API?
For production, use the official route for stability and compliance.
How do I prevent AI hallucination?
Use RAG, strict instructions, topic limits, and a fallback to humans.
Next steps
Want a WhatsApp chatbot for your business? Request a demo with your own FAQ data.