WhatsApp bots for customer support in India: a practical guide

For most Indian businesses, "where do customers contact us" has a one-word answer. WhatsApp beat the contact form years ago, and for anything with a booking or an order attached — a ferry ticket, a hotel stay, a delivery — it is where the "where is my…" message arrives first.

The problem is not getting customers onto WhatsApp. It is that WhatsApp, left alone, is a phone with no queue. Messages land on whichever number is on the poster, someone answers when they can, and nothing about it is measurable. This guide is about turning that into support you can actually run.

Step one: make every message a ticket

Before any bot, every inbound WhatsApp message needs to become a ticket in a helpdesk — with a category, a priority, an owner and a clock. This is the unglamorous part and it is the whole foundation. A bot on top of an unowned inbox just deflects into a void.

What "a ticket" should carry from a WhatsApp message:

• The phone number matched to a customer record, if one exists
• The booking or order it refers to, if the message contains a reference
• A category and priority, assigned by rules or by AI on arrival
• An SLA timer that runs against your business hours, not the clock on the wall

Once that exists, you can measure first-response time on WhatsApp the same way you would on email. Most businesses are surprised the first time they see the number.

Step two: decide what the bot is allowed to answer

A WhatsApp bot is a knowledge base with a very direct interface. It should answer exactly the questions your knowledge base can answer with a citation, and hand over everything else.

The right first list is short and boring:

1. Status questions — "is my booking confirmed", "where is my refund". These need a lookup, not a conversation.
2. Policy questions — cancellation windows, luggage, vehicle rules, check-in times. These come straight from documents you already have.
3. Simple actions — resend the ticket, resend the QR, reschedule to a listed alternative. Each is one API call.

Everything with a complaint, a group, an exception or a named person in it should go to a human on the first message. Not because the bot cannot try, but because the cost of getting it wrong on WhatsApp — where the customer is one tap from a screenshot — is higher than the cost of a hand-over.

Step three: the hand-over is the product

The single thing that decides whether customers accept a bot is what happens when it stops. A good hand-over has four properties:

• It is fast. Under a minute to a named person during business hours, and an honest "we'll reply by 9am" outside them.
• The person has the conversation. They do not ask the customer to repeat anything.
• The bot says so. "I'm connecting you to Sana, who can see this chat" beats silence.
• The customer can trigger it. "Talk to a person" should work at any point, in any wording.

If your bot vendor cannot show you the agent's screen at the moment of hand-over — the summary, the history, the suggested next step — the hand-over is not built yet.

The WhatsApp-specific rules

Running support on WhatsApp Business means working inside Meta's rules. The ones that matter operationally:

Consent. You can reply to a customer who messaged you within a 24-hour window. Outside it, you can only send approved template messages. Build your flows so that the customer's message opens the window, and use templates for proactive notifications like booking confirmations and reminders.

Templates are pre-approved. A "your booking HL-2041 is confirmed for 07:30 tomorrow" template takes a day or two to get approved and then works forever. Plan the templates you need — confirmation, reminder, reschedule offer, refund initiated — before launch, not after.

Quality rating. If customers block or report your number, your ability to send drops. Do not use the support number for marketing blasts. Keep the two separate.

A verified business. The green tick and the business profile matter more on WhatsApp than on any other channel, because the alternative looks like a scam.

What to automate first, and what to leave

A sensible first month looks like:

• Week 1: every WhatsApp message becomes a ticket; humans answer everything; you measure.
• Week 2: the bot answers status and policy questions above a high confidence threshold; everything else hands over.
• Week 3: self-service actions — resend, reschedule — go live for customers with a matched booking.
• Week 4: you look at hand-over reasons and the low-confidence cluster, add to the knowledge base, and lower the threshold where the evidence supports it.

Notice what is not on the list: complaints, refunds with a dispute, group enquiries, anything about a specific staff member. Those stay human. A bot that knows what it is not for is the one customers keep using.

Request a demo of Conversational AI →

The question to ask before you buy

The best test of a WhatsApp support bot is not a demo conversation. It is this: send a message the bot should not handle, and watch what the agent sees. If the agent gets the whole chat, a one-line summary, the customer's booking and a suggested reply — you have a support tool. If the agent gets a notification with a phone number, you have a chatbot, and your customers will find its limits before you do.