AI
Building a WhatsApp AI Assistant That Never Loses an Enquiry
How we built an AI assistant for Queen Touch Technology that answers customers on WhatsApp and the website in their own language, understands voice notes, and logs every conversation into a CRM — and what small businesses should copy from it.
Most small businesses don't lose customers because their service is bad. They lose them in the gap between a customer's message and the reply — a WhatsApp message read at 10 p.m. and forgotten, a voice note nobody had time to play, an enquiry that never reached the person who could quote it.
At Queen Touch Technology we closed that gap for ourselves first, with an AI assistant on our own WhatsApp number and website chat. This is how it is put together, and the design choices that matter more than the AI model.
The building blocks
The assistant is not one product. It is a small set of tools, each doing one job well:
- Chatwoot as the inbox, connected to the official WhatsApp Business (Cloud) API and our website chat — one place where every conversation lives.
- n8n as the workflow engine that receives each message, decides what to do and calls the other services.
- A large language model (DeepSeek) to understand the question and write the reply, grounded in our real service catalogue and price book.
- Speech-to-text (Whisper on Groq) so voice notes are transcribed and answered like typed messages.
- Frappe CRM, where every person becomes a lead, and every conversation becomes a note with a summary.
Answer in the customer's language
Our customers write in English, Tamil and a mix of both, and sometimes send voice notes instead of typing. The assistant replies in the language the customer uses. That single rule removed more friction than any feature we added.
Know when to hand over to a human
An AI assistant should not pretend to be the business owner. When someone asks for a quotation, a demo, a call or a person, the assistant says so, alerts our team on Slack and leaves a private note in the conversation. Quotations go through a human approval step before they are sent.
Memory turns chats into customer history
When a conversation is resolved, the assistant writes a short summary — what the customer asked, the outcome, the next step — into the CRM. The next time the same person writes, the assistant reads that history first. Existing customers with a problem get a support ticket automatically instead of a sales reply.
What small businesses should copy
You don't need our exact stack. You need the same principles:
- One inbox for every channel, so no message lives only on someone's phone.
- Answers grounded in your real services and prices — never invented ones.
- A clear handover to a person for anything that needs judgement or money.
- Every enquiry saved to a CRM, with a summary you can read in ten seconds.
- Use the official WhatsApp API, not unofficial tools that put your number at risk.
Technology like this is only valuable when it creates real impact — in this case, no enquiry left unanswered. If you want to build something similar for your business, Queen Touch Technology builds these systems for clients in India and abroad.
Written by S. Sathish Kumar, Founder & CEO of Queen Touch Technology.