What a WhatsApp AI agent actually costs in India.

You are paying three bills, not one

Almost every argument about WhatsApp pricing is really an argument about which of these three layers somebody is quoting. Separate them and the comparisons get easy.

When a vendor quotes you a single all-in rate per message, they have merged all three and made the mix decision on your behalf. That is not automatically a bad deal, but you cannot tell whether it is a good one, because you cannot see which layer the margin is sitting in.

Meta bills per message now — not per conversation

This is the single most important correction on this page, and most of the pricing guides currently ranking for this query get it wrong. On 1 July 2025 Meta retired the conversation-based model — the one where a 24-hour window was opened and billed as a unit, with 1,000 free service conversations a month. It no longer works that way.

Under the current model you are charged for each delivered template message, and for nothing else. Non-template messages inside an open customer service window are free, with no monthly cap. If you are reading a cost calculator that asks you how many conversations you expect, it was written for a model that has been dead for a year, and its total will be wrong in both directions.

A second change worth knowing: Meta moved India billing to INR on 1 January 2026. Rates are published in rupees and charged in rupees, so the FX gap between the rate card and your statement is gone.

The India rate card, and why category is everything

Meta revises these, and revises them often — the India marketing rate rose about 10% on 1 January 2026, and a fresh rate card took effect on 1 July 2026. So the figures below are deliberately given as a band rather than to four decimal places. Anyone quoting you an exact paise figure on a page they wrote months ago is quoting a number that has since moved. Check Meta’s live rate card before you commit a budget to any of them.

Read those three numbers again, because the ratio is the whole story. A marketing message costs roughly seven times what a utility message costs, and that ratio has held across every revision Meta has made. Your bill is decided by your category mix far more than by your volume. Ten thousand utility messages cost you somewhere around ₹1,250. Ten thousand marketing messages cost around ₹8,700. Same send count, same customers, seven times the invoice.

The ratio is the durable fact here, not the paise. Rates move every few months; the gap between a promotional message and a transactional one does not. Build your budget on the mix and you will be roughly right for years. Build it on a rate you copied off a blog and you will be wrong by the next quarter.

Add 18% GST on top of all of it, and on your platform fee as well. When you compare two quotes, confirm whether each one is inclusive or exclusive — 18% is comfortably enough to flip which vendor looks cheaper.

The messages that cost nothing at all

This is the part that separates a well-designed setup from an expensive one, and it is almost never in a vendor’s pitch — because it reduces the bill they are quoting you on.

1. 1. Every reply inside the service window is free. When a customer messages you, a 24-hour window opens. Inside it you can send unlimited free-form replies — text, images, price lists, PDFs — at no per-message charge. An agent that answers questions well is spending nothing to do it. The cost only starts when you reach out to someone who has not written to you.
2. 2. Utility templates inside that window are also free. This one is genuinely under-documented. An order confirmation or dispatch notice sent while the customer’s window is still open costs you nothing. The same template sent two days later costs the utility rate. Sequencing your notifications to land inside live conversations rather than as cold sends is a real, repeatable saving.
3. 3. Free entry points give you a longer window. Conversations that start from a Click-to-WhatsApp ad or a Facebook page call-to-action open a 72-hour free window instead of 24. Every message inside it — templates included — is free. If you already run Meta ads, routing them into WhatsApp is cheaper per conversation than routing them into a landing page.

The strategic conclusion is unglamorous and correct: the cheapest WhatsApp operation is one where customers message you first. Every rupee of message cost you pay is essentially the price of having to start the conversation yourself. Design for inbound — a number on the packaging, a QR on the invoice, a Click-to-WhatsApp ad — and most of your traffic moves into the free column permanently.

How a bill goes up without you sending more

Given the seven-to-eight-times gap between marketing and utility, category drift is the most expensive thing that can happen to a WhatsApp setup — and it happens silently.

Meta re-reviews template categories after approval. A template you wrote as transactional, submitted as UTILITY and had approved as UTILITY can be moved to MARKETING weeks later if the copy reads as promotional on a second look. No email arrives. Nothing in the console changes colour. Your send volume is identical and your invoice is several times larger. We have had this happen to a storefront template of our own, which is why our admin console now flags the mismatch between the category we intended and the category Meta currently reports.

If you manage templates by hand, make that comparison a monthly check — it is the only way you will notice. The mechanics of why it happens, and the delete-and-recreate trap that makes it hard to undo, are in why your WhatsApp template keeps getting rejected.

What we charge, and what moves it

A fixed build fee for the agent, plus a monthly fee to run, monitor and improve it. Meta’s message charges are yours and pass through at cost — we do not mark up Layer 1, because a provider whose revenue rises with your message volume has an interest in your message volume rising.

We scope the build against a number you already track — enquiries answered, orders taken, tickets closed, documents processed — so that at the end of a quarter you can tell whether it paid for itself, rather than admiring a dashboard. If we cannot find such a number with you during scoping, that is a signal the project is not ready, and we would rather say so than build against a metric invented for the invoice.

What actually moves the build number, in rough order of weight:

• How many systems it has to touch. An agent reading one product sheet is days. An agent that checks live stock in your ERP, holds an item, raises a payment link and writes the order back is meaningfully more work. This is the largest single driver.
• Whether it writes, or only reads. Answering questions is a well-understood problem. Taking an action that changes your data — reserving stock, issuing a refund, booking a slot — needs validation, failure handling and an audit trail, because a wrong write is worse than no answer.
• How much of your policy is written down. If your delivery zones, refund rules and escalation thresholds already exist as documents, we ground the agent on them. If they live in one person’s head, the first part of the project is getting them out of it — worth doing, but it is time.
• Language and channel spread. Hinglish and regional-language handling is largely a matter of testing rather than extra build. Adding channels beyond WhatsApp is not.

Five questions to ask any vendor

You do not need to be technical to price this well. You need to know which layer each number belongs to. These five questions separate a transparent quote from an opaque one, whoever you are talking to.

1. 1. Do you mark up Meta’s per-message rate, and by how much? Many providers add 10–30% on top of Layer 1. That is a legitimate business model, but it should be a stated number, not a discovery you make from a statement.
2. 2. Are your prices inclusive or exclusive of 18% GST? Ask about both the message charges and the platform fee. Two quotes on different sides of this are not comparable.
3. 3. What happens to my number and my chat history if I leave? The WhatsApp number should be registered to your business, not to the vendor. If leaving means losing the number your customers have saved, the switching cost is the real price.
4. 4. Is there a per-seat, per-contact or per-bot charge? These scale in ways that are invisible at pilot size and painful at production size. A per-contact fee on a growing customer list is a bill that rises whether or not you use the product.
5. 5. How does the setup use the free windows? If a vendor cannot explain the difference between a template send and a service-window reply, they are not optimising your bill, because they do not know where it comes from.

Where to go from here

If you are still deciding what to build rather than what to pay, chatbot vs AI agent is the honest comparison — the two have very different cost profiles because a scripted bot deflects conversations that an agent would close. If you are setting up commerce from scratch, how to sell on WhatsApp covers the journey from first message to paid order, and what a WhatsApp AI agent actually is covers the mechanics of how one takes an order.

For the full range of what we build beyond WhatsApp — support agents, document processing, data pipelines — see our services. And if you would rather skip the reading, the fastest way to price this is to tell us the number you want moved. The scope falls out of that conversation in about twenty minutes.