WhatsApp AI chatbots for Indian businesses: what works in 2026

Every week someone asks us for "a ChatGPT on our WhatsApp". The technology is genuinely ready. Whether the project succeeds depends almost entirely on decisions made before a single line of code is written.

Start with the conversations you already have

Before we design anything, we export a few months of real WhatsApp chats and read them. In almost every business, 30 to 50 questions make up the large majority of enquiries: fees, timings, location, availability, delivery time, "is this in stock", "do you have EMI".

Those are the questions an AI assistant should own completely. The rare, complicated or emotional conversations — a complaint, a negotiation, a medical concern — should go to a person, fast, with the full chat attached.

The three guardrails that matter

• Prices and offers are never generated. They are looked up from a fixed table. A language model that "helpfully" invents a 20% discount is the fastest way to lose trust in the whole project.
• The assistant only answers from your content. Website, brochures, policies and price lists are indexed. If the answer is not in there, it says so and offers a human.
• Handover is one message away. Typing "talk to someone" should always work, at any point, without a menu.

Hindi, English and Hinglish

Indian customers switch languages mid-sentence, and modern models handle this well. We test with real customer phrasing — "fees kitni hai for the weekend batch?" — rather than polished English, because that is what actually arrives at 11 pm.

What it really costs to run

There are two running costs. AI usage is usually smaller than people expect: for a typical SME handling a few thousand conversations a month it is a few thousand rupees. WhatsApp conversation charges are set by Meta and depend on volume and conversation type. We always show clients a usage estimate before launch and a live cost dashboard after.

Measure resolution, not "engagement"

The number that matters is how many enquiries were fully answered without a human, and how many qualified leads reached your team. For an admissions assistant we launched for a Delhi coaching institute, roughly two thirds of enquiries were resolved by the assistant, and counsellors spent their time only on families ready to enrol.

The part everyone skips

Someone has to read a sample of transcripts every week for the first two months. That is where you find the questions the assistant handles badly, the policy nobody wrote down, and the product detail missing from your website. A chatbot that is reviewed monthly keeps getting better. One that is launched and forgotten slowly gets worse.

Arjun Rao

AI & Automation Lead at Corewebbers, New Delhi. Builds the LLM chatbots, RAG pipelines and WhatsApp automations our clients run on.

Ask a question