Intro

In India, WhatsApp is where business already happens. Customers ask prices, send photos, confirm bookings and chase orders — all on WhatsApp, all needing someone to pick up the phone.

A WhatsApp AI agent handles those conversations automatically. It reads the message, checks your calendar or order system if needed, and replies properly.

Here’s how it works, what it costs, and where to start.

 How It Works

One loop:

1. Customer messages your WhatsApp number
2. AI reads it and works out what they want
3. It fetches what it needs — price list, calendar, order status
4. It replies in normal language

What makes it your agent is what you feed it: your real hours, prices, policies and tone. A bot that says “Press 1 for sales” is a menu. An agent that says “We’re open till 8 today, and the ₹2,500 package includes delivery” is answering the question.

 What It’s Used For

• Repeat questions — hours, pricing, location, delivery areas. Usually most of your inbox.
• Qualifying enquiries — asks budget and timing so your callback is worth making.
• Bookings — checks a calendar and confirms a slot. Common for salons, clinics, consultants.
• Order status — the highest-volume message any delivery business gets.
• Quote follow-ups — day 3 and day 7 nudges nobody does manually. Usually where the recovered revenue is.
  

 The 24-Hour Rule

This decides what’s possible, so understand it before building anything.

Inside 24 hours of a customer messaging you, you can reply freely in your own words. Every new message from them resets the clock.

Outside that window, you must use a message template pre-approved by Meta. That’s why “follow up with last month’s enquiries” isn’t as simple as it sounds.

Two consequences: get explicit opt-in, and expect a low daily messaging limit at first that grows as your quality rating improves.

 What It Costs

Three separate costs:

1. WhatsApp conversation charges — Meta bills per conversation, varying by category. Utility is cheaper than marketing.
2. The AI — billed by usage. Usually the smallest of the three at small-business volumes.
3. The platform — a no-code tool subscription, or development time.

For a few hundred conversations a month, this typically costs far less than the staff hours it replaces. At high volume, model the conversation charges first — they scale.

 Two Ways to Build It

No-code — tools like n8n or Make connect WhatsApp to an AI and your other systems by dragging boxes. First agent live in days.

Custom build — a small server, giving full control over logic, memory and human handoff. Better for multiple agents or complex rules.

Start no-code. You can rebuild later if you need to, and most businesses never do.

 Where It Goes Wrong

• Building everything at once. One job, working properly, beats five done badly.
• No escape hatch. Every agent needs a clear route to a human. Trapping people in loops costs you more trust than having no agent.
• Unsupervised irreversible actions. Bookings, refunds and payments get confirmed, not guessed.
• Feeding it nothing. Generic input, generic answers.
• Skipping testing. Yourself, then one friendly customer, then a small group. Then everyone.

 Where to Start

1. Write down your 20 most-asked questions with real answers. This is the actual work — do it even if you never build the agent.
2. Build an FAQ agent first. No integrations needed. Proves the loop works.
3. Add one lookup — order status or calendar, whichever matters more.
4. Add follow-ups last, once templates are approved.

Most businesses stop at step two and get most of the benefit.

 Do You Actually Need This?

Count how many WhatsApp messages you answered last week, and how many were questions you’d answered before.

Mostly repeats? Automation pays for itself.

Handful of messages a week? Not yet. Answering personally is faster and converts better. Your problem is enquiry volume, not enquiry handling — which is what  SEO 
and  paid ads  are for.