WhatsApp AI chatbot for small business in India: options and monthly cost

Last updated 31 October 2026 · Prices seen 31 October 2026 on each company’s own pricing page · Sources

I have set up WhatsApp bots for kirana and clothing shops, a dental clinic, two coaching centres, a cloud kitchen and a hardware distributor. The ones that paid for themselves were rarely the clever ones. They answered the five questions customers ask every day, passed everything else to a person within minutes, and were checked every week by someone who cared. The ones that failed usually started with “let’s put AI on WhatsApp” before anyone had read a month of old chats. This guide follows the order I now use with clients: what you actually need, what Meta allows, what it costs, then how to build and run it.

What to remember

Prices seen 31 October 2026 on Meta’s WhatsApp rate card and each provider’s own pricing page, before GST. Dollar prices are shown in dollars, as the company bills them.

The three levels of WhatsApp bot

“WhatsApp AI chatbot” is used for three quite different things. Knowing which one you need saves most of the money.

Level 0: the free WhatsApp Business app

The WhatsApp Business app is free to download and has light automation built in: a greeting message for new chats, an away message outside business hours, quick replies (WhatsApp says up to 50 saved replies, triggered by typing “/”), labels to sort chats, and a catalogue that can hold up to 500 items. There are no Meta message fees for normal chats in the app. The limits: one main phone plus a few linked devices, no real branching flows, and no link to your billing or CRM.

In May 2026 Meta launched Business AI in the WhatsApp Business app in India. It answers customer questions about products, prices, offers and delivery in Indian languages, learns from your catalogue, Facebook Page and website, and lets you take over the chat at any time. Meta says it is for eligible businesses, set up under Tools → Your Business AI, and WhatsApp’s help pages describe a gradual rollout with a waitlist. Meta’s India announcement does not state a fee, so check inside your app before you rely on it.

Level 1: a menu or button bot on the API

This is what most providers call a “chatbot”: you draw a flow with buttons and lists (“1. Price list 2. Track order 3. Talk to us”) in a no-code builder. It needs a WhatsApp Business API number through a provider such as AiSensy, Interakt, WATI or Gallabox. It is predictable, cheap to run and good for repeated questions. It cannot understand a question that does not fit your buttons.

Level 2: an AI chatbot or AI agent

An AI bot reads the customer’s own words, in English, Hindi or Hinglish, and writes a reply from the information you give it: your price list, policies, FAQs and catalogue. A stronger “AI agent” can also look things up, such as order status in Shopify or a free slot in your booking calendar, and hand over to staff. Options include provider add-ons (AiSensy AI Agent, Gallabox AI credits, WATI AI features, respond.io AI Agents), bot platforms like Botpress connected to your WhatsApp number, or Meta’s own Business Agent. The risk is that AI can state a wrong price or invent a policy, so it must be fenced in.

How a WhatsApp chatbot actually works

It helps to picture the path one message takes, because every cost and every failure sits somewhere on it.

1. The customer writes “price kya hai?” to your business number. That message is free and opens the 24-hour service window.
2. Meta passes it to your provider (AiSensy, Interakt, WATI, Gallabox and the rest) through the WhatsApp Business Platform. This is why a real bot needs an API number and the normal app cannot run one.
3. The bot decides. A menu bot matches the message to a button or a word you set up. An AI bot reads the words, looks up the facts you gave it, and drafts a reply. An agent can also check an order in Shopify or a slot in your calendar.
4. The reply goes back as a service message. The first 1,000 each month per number are free, then each one is charged.
5. If the bot is unsure, a person takes over in the provider’s shared inbox, ideally with the whole chat in front of them so the customer does not repeat themselves.

When something goes wrong, it is almost always at step 3 (the bot had the wrong or old facts) or step 5 (nobody was watching the inbox). The software is rarely the problem.

Meta’s rules for chatbots and AI on WhatsApp

WhatsApp is strict, and breaking its rules can get templates paused or a number restricted. Four rules matter most for a bot.

1. AI for your business is fine; a general AI assistant is not

Meta’s WhatsApp Business Platform terms (last modified 23 September 2026) say that providers of AI technologies, including large language models and general-purpose AI assistants, are “strictly prohibited” from using the platform to offer those technologies when they are “the primary (rather than incidental or ancillary) functionality being made available”. Meta’s developer documentation puts the start date at 15 January 2026 and allows exceptions only where Meta is legally required to, which does not include India.

In plain terms: a saree shop whose bot answers “is this available in red?” is using AI incidentally to serve its own customers, which the terms allow. A start-up offering “chat with our AI about anything on WhatsApp” is not. The same terms let you hire an AI company as your solution provider, but WhatsApp chat data must not be used to train or improve AI models, except a model fine-tuned only for you. Meta also decides which side of the line you fall on, so keep your bot clearly about your business.

2. Opt-in before you message anyone

The WhatsApp Business Messaging Policy (last updated 23 September 2026) says you may contact people only if they gave you their phone number and agreed to receive messages from you, and that you are responsible for how you collect that opt-in. Meta recommends separate opt-ins for different kinds of messages, for example order updates versus offers. A number from a bill book or a bought list is not opt-in.

3. The 24-hour window and templates

When a customer messages you, a 24-hour customer service window opens. Inside it, your bot or staff can send free-form replies. Outside it, you can only send templates that Meta has approved in advance, such as an order update, a reminder or an offer, and each delivered template is charged. Meta can pause or reject templates that customers block or report.

4. A human must be reachable

The same policy says you may use automation inside the 24-hour window, “but must also have available prompt, clear, and direct escalation paths”, and lists examples such as transfer to a human agent in the chat, a phone number, email, a web support page, a store visit or a support form. A bot that loops customers forever breaks the policy and loses sales.

What it costs: three ₹ examples

A WhatsApp AI chatbot bill has three parts: Meta’s message fees, the provider’s plan, and AI usage. These examples use Meta’s India rates from 1 October 2026: marketing ₹0.8631, utility ₹0.115 and service ₹0.115 per delivered message, with the first 1,000 service messages per number each month free. AiSensy bills its own published rates (marketing ₹1.09, service and utility ₹0.145), so we use those where the example uses AiSensy. All prices exclude GST unless stated; we add 18% GST to the full bill.

Assumptions: each chat gets 4 replies from the bot or staff; Examples B and C send 1 utility message per chat (such as a booking or delivery update); marketing messages are 300, 2,000 and 5,000 a month. Your numbers will differ, so use the WhatsApp API cost calculator with your own volumes.

Why the AI line matters. If half the replies in Example B are written by AI, that is 6,000 AI replies a month. At Meta Business Agent’s published rate of about 4–5 US cents per message, that alone would be US$240–300, or roughly ₹21,000–₹26,400 at an assumed ₹88 per dollar, more than double the rest of the bill. In Example C, 40,000 AI replies at that rate would be roughly ₹1.4–1.76 lakh. Provider AI add-ons with included messages are usually cheaper at small volumes, but read the overage price before you switch AI on for everyone.

Two cheaper paths are worth comparing: answer simple questions with buttons and use AI only when the customer types something the menu cannot handle, or stay on the free app and test Meta’s Business AI if your business is eligible.

WhatsApp chatbot providers compared

Prices below are from each provider’s official pricing page, seen 31 October 2026. Plans in US dollars are shown in dollars; your card may also be charged GST and currency conversion. WhatsApp message fees are extra on every plan unless stated.

One thing the table hides: user limits. Gallabox Basic is a hard cap of 3 users and WATI Growth includes 3 users with no extra seats, so a shop with four people on the counter is really shopping for the next plan up. For a deeper comparison of onboarding, support and markups, read our guide to WhatsApp Business API providers in India.

Setup checklist (no coding)

1. List your top 20 questions. Scroll your last month of chats and note what customers ask. Usually price, stock, timing, location, delivery and order status cover most of it.
2. Decide the level. If 20 questions cover most chats, try the free app’s quick replies first. If you need staff sharing one number or flows, choose an API provider.
3. Get a number ready. Use a number that is not active on the normal WhatsApp app, or plan to move it. Have your business name, GST or shop documents and website ready for Meta verification.
4. Connect through the provider’s signup. Most providers connect your Meta Business account and WhatsApp number in their own screens. Add a payment method so service replies beyond 1,000 are delivered.
5. Build the main menu. Five buttons or fewer, with “Talk to a person” always visible.
6. Feed the AI only approved facts. Upload your current price list, return policy, timings and FAQs. Tell it not to promise discounts, refunds or medical or legal advice.
7. Write templates and get them approved. Order confirmation, reminder and one offer template are enough to start.
8. Collect opt-in properly. A checkbox at checkout, a QR code in store with a “Hi” message, or a website button, with clear words about what you will send.
9. Test with 20 real messages. Include Hinglish, spelling mistakes, angry messages and voice notes.
10. Review weekly. Read chats the bot could not answer, update facts and track your monthly bill against the calculator.

Build it yourself or use a provider?

People who can code often ask me whether to skip the provider and connect straight to Meta’s Cloud API with their own AI. You can, and for some businesses it is the right call. For most shops it is not, because the hard part is not the code. It is template approvals, the shared inbox for staff, opt-in records, and someone fixing it at 9 pm on Diwali week.

My rule of thumb: if nobody in the business can explain what a webhook is, use a no-code provider and spend the saved time on the bot’s facts. You can always move later; your number and opt-in list stay yours, so ask the provider how a migration works before you sign.

How long it takes to go live

From my own setups, a menu bot for a single shop usually takes one to two weeks of calendar time, though only a few hours of actual work. Expect delays in these places:

• Business verification and display name. Meta checks your business details and the name customers will see. Mismatched names between your GST papers, website and chosen display name are the most common hold-up I see.
• Moving the number. If the number is already on the normal WhatsApp Business app, plan when you will switch, and save any chats you need first.
• Template approvals. Order updates are usually approved quickly; offers worded like spam get rejected or pushed into the marketing category.
• Writing the facts. This is the step owners underestimate. A clean price list, timings and return policy can take longer than the whole technical setup.

An AI bot adds a week or two of testing on real chats before I let it answer customers on its own.

Sample flows for four Indian businesses

These are starting points. Replace the prices and timings with your own, and keep the language your customers actually use.

Clothing store

Menu: 1. New arrivals 2. Check size and stock 3. Track my order 4. Return or exchange 5. Talk to us

Customer: “Bhaiya ye kurti M size mein hai? Kitne ka hai?”
Bot: “Haan ji, ye kurti M size mein available hai. Price ₹899 hai, COD bhi chalega. Order karne ke liye ‘Order’ likhiye, ya photo bhejiye agar koi aur design dekhna hai.”

AI instruction: “Answer only from the catalogue and stock sheet. Never offer a discount. For exchanges, share the policy and pass to staff.”

Clinic or coaching centre

Menu: 1. Book an appointment / demo class 2. Timings and fees 3. Location 4. Reschedule 5. Talk to the front desk

Customer: “Kal shaam ko doctor available hain kya?”
Bot: “Kal shaam 5 se 8 baje tak slots hain: 5:30, 6:15 aur 7:00. Kaunsa slot book karun? Agar emergency hai to turant clinic ko call kijiye.”

For a coaching centre the same flow books a free demo class and sends a utility reminder the evening before. Guardrail: a clinic bot must never suggest a diagnosis or medicine; it books, reminds and hands over. Meta’s terms say the platform is not designed for regulated health data, so keep reports and prescriptions out of the bot.

Restaurant or cloud kitchen

Menu: 1. Today’s menu 2. Order for delivery 3. Table booking 4. Where is my order? 5. Talk to us

Customer: “Order kab tak aayega? 40 min ho gaye”
Bot: “Sorry for the wait! Aapka order #2145 nikal chuka hai, rider 10 minute mein pahunch jayega. Agar 15 minute mein nahi aaye to ‘Help’ likhiye, manager aapko call karenge.”

Order status needs a link to your ordering system. Without one, the bot should collect the order number and pass the chat to staff.

B2B distributor

Menu: 1. Price list 2. Place a repeat order 3. Outstanding and ledger 4. Dispatch status 5. Talk to sales

Customer: “Sir 10 carton Parle-G ka rate bhejo, aur pichla payment update kar do”
Bot: “Namaste! Aaj ka rate PDF bhej raha hoon. 10 carton ka order confirm karna ho to ‘Confirm 10’ likhiye. Payment update ke liye aapka ledger sales team check karke aaj shaam tak bhejegi.”

Guardrail: credit limits, special rates and outstanding amounts should come from your accounting data or a person, never from the AI guessing.

To write greeting, away and quick-reply messages for your flows, use the free AI prompt generator.

Writing the facts and instructions your AI bot uses

“Training” an AI bot for a small business mostly means giving it a short, current set of facts and firm rules. Good providers let you upload documents or paste text, and Meta’s own Business Agent learns from your catalogue, Facebook Page, website and the price lists you upload. What you put in decides what comes out.

• One fact sheet, one owner. Prices, stock rules, timings, delivery areas and charges, return policy, payment options. One person updates it, and it carries a date at the top.
• Write answers, not brochures. “Delivery in Pune city ₹40, free above ₹999, same day if ordered before 2 pm” works. A paragraph of marketing copy does not.
• Tell it what it must never do. No discounts, no refund promises, no medical, legal or tax advice, no guessing a price that is not in the sheet.
• Tell it when to hand over. Complaints, bulk orders, anything about money owed, and any customer who asks twice for a person.
• Give it your customers’ words. Paste twenty real questions, in the Hinglish and spellings people actually use, with the answer you would give.

A short instruction that has worked well for my shop clients: “You are the assistant for [shop name]. Answer only from the facts below. If the answer is not there, say you will check with the team and offer to connect them. Reply in the customer’s language and keep replies under three lines.”

Voice notes, photos and messages a bot will not expect

Indian customers send voice notes, screenshots of products from Instagram, photos of a broken part and “?” on its own. Plan for these before launch, not after the first angry review.

• Voice notes. Some providers now offer AI voice features, for example Gallabox lists AI voice on WhatsApp on its Advanced plan. If yours does not, the bot should reply “Please type your question or tap Talk to us” and flag the chat for staff.
• Photos and screenshots. A menu bot cannot read them. Route them straight to a person with a note of what the customer tapped last.
• Late-night messages. Say clearly when a person will reply, and keep the promise. “Our team replies from 10 am” beats a bot pretending to be awake.

How to tell if the bot is working

I ask owners to look at five numbers every week for the first two months. None needs a dashboard; most providers show them, and a notebook works too.

If the owner never reads the failed chats, the bot quietly gets worse as prices and stock change. That weekly fifteen minutes is the real cost of running one.

Meta AI in WhatsApp is not your business chatbot

Many owners confuse three things. Meta AI is the assistant anyone can chat with inside WhatsApp; it does not answer for your shop. Business AI is Meta’s agent inside the free WhatsApp Business app, which answers your customers from your own catalogue and information, for eligible businesses (see Meta’s India announcement). A provider bot runs on the WhatsApp Business Platform with your own flows, staff inbox and links to your other systems. If your chats are few and your catalogue is simple, test Business AI first; it may be all you need.

If you are also choosing the rest of your software, our guides to choosing a CRM and building a lean tool stack show where a WhatsApp bot fits, and AI tools for small business covers the AI subscriptions around it.

Data safety and consent

• Collect the minimum. Name, phone and order details are enough for most bots. Do not ask for Aadhaar, full card numbers, OTPs or bank passwords in chat.
• Know where chats go. Ask your provider which AI model answers your customers, where data is stored, how long chats are kept and how to delete them. Meta’s terms ban using WhatsApp business data to train shared AI models.
• Keep opt-in records. Save when and how each customer agreed to messages, and honour “STOP” quickly.
• Plan for the DPDP Rules. India’s Digital Personal Data Protection Rules, 2025 were notified on 13 November 2025, and most duties on businesses start 18 months later, around May 2027. Customer chat data is personal data, so use providers with clear data terms.
• Tell customers it is a bot. A simple “I’m the shop’s assistant; type ‘person’ any time to reach our team” builds trust.

Mistakes to avoid

• Letting AI quote prices from memory. Give it a current price list and tell it to say “let me check with the team” when unsure.
• Sending offers to every saved contact. Without opt-in you risk blocks, reports and paused templates, and each marketing message costs money.
• No exit to a human. It breaks Meta’s policy and loses angry customers.
• Paying for AI before buttons are tried. Many shops handle most chats with five buttons.
• Ignoring the per-message cost of AI. Check the AI overage price, not just the monthly plan.
• Forgetting GST and markup. Provider rates can sit above Meta’s; AiSensy’s published service rate is about 26% above Meta’s ₹0.115, and GST is added on top.
• Building a general “ask anything” AI. That is the use Meta bans. Keep it about your business.
• Not testing Hinglish. “Rate kya hai”, “kitne ka” and “price batao” should all reach the same answer.
• Letting the fact sheet go stale. The most common failure I am called in to fix: the bot is fine, but last month’s price list is still loaded.
• Launching to everyone on day one. Switch the bot on for new chats first, keep staff watching for a week, then widen it.
• Choosing on the plan price alone. User caps, AI credits and message markup decide the real bill. Work it out for your volumes with the WhatsApp Business API pricing guide.

Related guides

• WhatsApp API cost calculator
• WhatsApp Business API pricing in India
• WhatsApp Business API providers in India
• ChatGPT, Gemini and Claude prices in India
• AI tools for small business in India
• AI for accountants and CAs in India
• ChatGPT for Excel
• How to choose a CRM for a small business in India
• Free CRM software in India
• Build a lean business tool stack

Frequently asked questions

Disclaimer: This guide is general information, not legal or tax advice. WhatsApp rules, Meta rates and provider prices change often, and examples use stated assumptions. Confirm current prices and policies on the official pages before you buy or send messages.