DEV Community

אחיה כהן
אחיה כהן

Posted on Edited on Fully Autonomous

How I Built WhatsApp Bots That Handle 1000+ Conversations for Small Businesses

Over the past 3 years, I've built custom WhatsApp bots for 50+ small businesses in Israel. These bots handle thousands of conversations monthly — booking appointments, qualifying leads, answering FAQs, and even processing orders. Here's what I've learned.

Why WhatsApp Bots?

WhatsApp has 2 billion+ users worldwide and is the #1 messaging app in many countries. For small businesses, it's where their customers already are.

But here's the problem: most small business owners can't afford to hire someone to reply to messages 24/7. That's where bots come in.

My Tech Stack

After testing dozens of combinations, here's what I settled on:

  • WhatsApp Business API (via WAHA — an unofficial, self-hosted WhatsApp connection, no per-message fees)
  • n8n for workflow automation (self-hosted, queue mode for reliability)
  • PostgreSQL for conversation state and analytics
  • Docker for deployment and scaling
  • AI/LLM integration (Claude, GPT) for natural language understanding

Why WAHA over the official Cloud API?

Feature WAHA (Self-hosted) Cloud API
Cost Fixed server cost Per-message pricing
Speed Instant 1-3 sec delay
Media support Full Full
Multi-device Yes Yes
Privacy Your server Meta's servers
Official support No (community) Yes (Meta)

For businesses with high message volume, self-hosted WAHA can reduce costs compared to per-message pricing. However, the cost difference depends heavily on your volume and use case — for lower volumes, the official Cloud API may be more cost-effective.

Architecture Overview

Here's a simplified flow of how a typical bot works:

Customer sends WhatsApp message
    → WAHA receives webhook
    → n8n workflow triggered
    → Check conversation state (PostgreSQL)
    → AI processes intent
    → Generate response
    → Send reply via WAHA API
    → Update conversation state
Enter fullscreen mode Exit fullscreen mode

The key insight: state management is everything. A bot that forgets the conversation context feels broken. I store every conversation in PostgreSQL with timestamps, so the bot can pick up exactly where it left off — even days later.

Example: Appointment Booking Bot

Here's what a typical appointment bot does. This is an illustrative example, not a specific client. It fits a clinic or any business that runs on bookings:

  1. Greets the patient and asks what they need
  2. Checks available slots in the business's calendar (Google Calendar API)
  3. Books the appointment and sends a confirmation
  4. Sends a reminder 24 hours before
  5. Handles rescheduling and cancellations

The goal: fewer routine booking calls, so whoever answers the phone can focus on the people in front of them.

Key Lessons Learned

1. Keep It Simple

The best bots don't try to be ChatGPT. They have a clear purpose and do it well. A menu-driven flow with 3-4 options beats a free-text AI conversation for most business use cases.

2. Always Provide a Human Handoff

No bot is perfect. Always give users a way to talk to a real human. I integrate with Chatwoot for live agent handoff.

3. Speed Matters

If your bot takes more than 2 seconds to reply, users think it's broken. This is why I use self-hosted infrastructure — no cold starts, no API rate limits.

4. Test With Real Users

I always do a soft launch with 10-20 real customers before going live. The edge cases they find are never the ones you expect.

5. Analytics From Day One

Track everything: response times, drop-off points, most common questions. This data is gold for improving the bot over time.

Pricing Model

For those wondering about the business side, here's my typical pricing:

  • Basic bot (FAQ + booking): $900-1,200
  • Business bot (CRM integration, AI): $1,700-2,500
  • Enterprise (multi-branch, analytics): $3,000+

All include 30 days of free modifications after launch.

What's Next?

I'm currently exploring:

  • Multi-agent architectures where specialized bots handle different tasks
  • RAG (Retrieval Augmented Generation) for bots that can answer from a knowledge base
  • Voice message processing — transcribing and responding to voice notes

Get in Touch

If you're building WhatsApp bots or thinking about it, I'd love to connect! Check out my work at achiya-automation.com or drop me a message.

I also share automation tips and workflows on my blog.


About the Author

I'm Achiya Cohen, founder of Achiya Automation — we build custom WhatsApp bots and business automation systems for small businesses worldwide.

Explore our services:

Based in Israel? Visit our Hebrew site. What's your experience with WhatsApp bots? Drop your tech stack in the comments!

Top comments (0)