If your WhatsApp budget was modelled on a 24-hour conversation fee, it is two pricing models out of date. And if someone has told you that Meta's new AI policy bans your customer service bot, they have misread it — the opposite is true, and the distinction is worth money to you.

Four mechanics decide whether a WhatsApp programme makes money in India. None of them are messaging strategy. They are which product you are on, which category Meta assigns your template, how recipients behave, and which Indian law actually applies. Our performance and automation work now starts with those four, because they determine the economics before any copywriting does.

What changed in WhatsApp pricing, and when?

Two structural changes in eighteen months, both of which invalidate older arithmetic.

1 July 2025 — conversation pricing ended. Meta deprecated the model where a single fee covered unlimited messages inside a 24-hour window, and moved to per-delivered-message pricing. You are now charged per message that actually reaches a device, at a rate set by the message's category and the recipient's country. Meta's pricing documentation sets out the current model.

1 October 2026 — free service replies ended. Previously, replying to a customer inside the 24-hour service window cost nothing. From October, service messages and utility replies sent inside that window became chargeable per message, at rates that vary by market and broadly track utility and authentication rates. Vendor documentation of the change explains it clearly, though as always verify against Meta's own page before you budget.

That second change matters more than it looks, and it corrects something we believed until we checked: inbound response is no longer free. It is still the cheapest thing you can do on the platform, by a wide margin. But a support-heavy operation that modelled on zero marginal cost per reply now has a per-message line item it did not have in September.

There is a third change worth knowing because it changes your cost model rather than your rate. Meta's own Business Agent — its native AI layer — is priced on compute consumed rather than per message, at two dollars per million tokens from August 2026, which works out to a few cents per exchange depending on how much context each conversation carries. That is a different shape of cost to a per-message rate, and it is not interchangeable with one when you are forecasting.

The practical consequence of all three is simple. Any WhatsApp proposal, quote or internal projection written before mid-2025 is describing a pricing model that no longer exists, and anything written before this October is missing a cost line. If a provider has quoted you on "conversations," ask them which model they are actually billing you on.

What is the Category Tax?

Here is the mechanic that decides your unit economics, and the part most businesses have never read.

Every template you send falls into a category, and the categories are priced very differently. In India, the aggregated rates circulating in the market put marketing messages around ₹0.86 each and utility or authentication messages around ₹0.115 — roughly a seven-fold difference for the same act of sending a message. Those exact rupee figures come from Business Solution Provider rate-card aggregations rather than from Meta's own published text, so treat them as indicative and confirm your own rates with your provider.

But the rate is not the interesting part. This is: you submit a template, and Meta's systems decide its category.

Put a discount code inside what you intended as a shipping update, and Meta's categorisation logic will reclassify it as marketing — and your cost per message multiplies. You are notified through the template_category_update webhook, which means if nobody is monitoring that webhook, your costs can change without anyone noticing until the invoice arrives.

That is the Category Tax: the single largest driver of your WhatsApp cost is a classification decision you do not make. The operational consequence is specific — keep utility templates ruthlessly clean of anything promotional, and monitor the categorisation webhook as a cost control rather than a technical nicety.

It also settles the question the plan for this article started with. Broadcast promotion delivers the worst return on WhatsApp and inbound response the best, and you do not need any argument about open rates to get there. Marketing templates cost roughly seven times what a service reply costs. The asymmetry is in the rate card.

One more lever sits alongside the category, and it arrived recently. Meta opened a Marketing Messages API in October 2026 that lets advertisers set a maximum price they are willing to pay for a marketing message, with delivery allocated against that bid. For businesses sending marketing at volume this converts a fixed rate into something closer to an auction — which is familiar territory if you already run paid media, and unfamiliar if you have treated WhatsApp as a flat-rate broadcast channel.

How do you actually lose a WhatsApp number?

Not by sending too much. By sending things people do not want.

Your number carries a quality rating — green, yellow or red — driven primarily by recipients blocking and reporting you. Volume is not the input. A business can send a hundred thousand well-targeted utility messages a day and hold a green rating; a few hundred irrelevant promotional templates that trigger the block button will start the decline.

The enforcement mechanism is template pacing, and the sequence is documented. When you launch a marketing or utility template, Meta sends it to a small sample first. If negative feedback appears, the template is paused automatically: first pause three hours, second pause six hours, third instance permanently disabled. During a pause the API returns error 132015, and any automated sequence depending on that template needs to stop rather than retry.

Separately, your outbound scale is tiered. Messaging limits run 250, 2,000, 10,000, 100,000 and unlimited unique recipients per 24 hours outside the service window. New accounts start at 250. Reaching 2,000 requires business verification plus delivering 2,000 quality messages in 30 days, and subsequent automatic escalation requires using at least half your current limit within seven days while holding quality.

And the two products behave differently. Throughput on the free WhatsApp Business App is hard-capped at 20 messages per second; the Cloud API starts at 80 and scales automatically to 1,000 once you send 100,000 messages in 24 hours while holding a yellow or green rating. If someone has pitched you "WhatsApp automation" without establishing which product you are on, they have not scoped the job.

That distinction decides more than speed. The Business App is a phone application with broadcast lists and quick replies — adequate for a clinic, a boutique or a single-location showroom handling tens of conversations a day, and free. The Cloud API is infrastructure: it needs a provider, a verified business, template management and someone monitoring webhooks, and it carries per-message cost from the first send. Businesses frequently buy the second while needing the first, which is how a ₹2,000-a-month problem becomes a ₹40,000-a-month retainer.

Three things are worth internalising about how enforcement actually behaves in India. Negative feedback compounds across your templates rather than sitting in isolation. Quality decline is gradual and visible before it is terminal, so a monitored rating gives you weeks of warning. And a number that has been permanently restricted is not something a provider can argue back — which is why the cheap discipline of relevance is also the only real insurance.

Does Meta's new AI policy ban your chatbot?

Almost certainly not, and this is the most misreported development on the platform.

Meta announced the change around 18 October 2025 and it took effect 15 January 2026. TechCrunch's account is the clearest: Meta added an "AI providers" section to the Business API terms barring general-purpose chatbots. What is prohibited is using the API to offer AI technology where that technology is the primary function rather than an incidental one, judged at Meta's sole discretion.

What remains permitted is a business using AI to serve its own customers. TechCrunch's own example is a travel company running a customer service bot.

So the parties actually affected were the AI companies themselves — OpenAI, Perplexity and Microsoft all announced their WhatsApp assistants would stop working. If you run a bot that answers questions about your products, bookings, orders or admissions, you are the permitted case, not the prohibited one.

Two honest caveats. Meta has not published a rigid technical boundary between a permitted business-specific assistant and a prohibited general-purpose one, and it reserves sole discretion — so a bot that will cheerfully discuss anything a customer asks about is closer to the line than one scoped to your business. And Meta now prices its own Business Agent on compute consumption rather than per message, which is a different cost model to understand if you use Meta's native AI rather than your own.

The practical rule: scope the assistant to your business, keep a human handover path, and do not build something that functions as a general assistant with your logo on it.

Which Indian law actually applies to WhatsApp?

This is where Indian businesses are being sold something they do not need.

TRAI's DLT regime does not apply to WhatsApp. The TCCCPR framework — sender header registration, template pre-approval on a distributed ledger, preference scrubbing — governs commercial messaging routed through telecom operators. WhatsApp is an over-the-top service, and the Telecommunications Act, 2023 kept over-the-top services outside that perimeter. TRAI has consulted on whether to extend regulation to over-the-top messaging as spam migrates there, and industry bodies have formally contested the extension on the basis that these platforms are already regulated as intermediaries. Those consultations are proposals, not law.

So if a provider is selling you a "DLT compliance package" for WhatsApp templates, you are paying for compliance with a regime that does not reach the channel. Ask them to cite the instrument.

What does apply is DPDP, and Meta's own policy. Under the Digital Personal Data Protection Act, 2023 — whose Rules were notified in late 2025 with phased compliance — consent to message someone must be free, specific, informed, unconditional, unambiguous and withdrawable. It cannot be bundled, so you cannot make a service conditional on accepting promotional messages. Withdrawal has to be as easy as consent was, which in practice means a working opt-out inside the chat. And the burden of proving valid consent sits on you, which a boolean opt_in = true field in your database does not discharge — you need a record of what text the user actually saw, when.

One correction on the penalty, because the figure circulating is wrong. The DPDP Schedule sets ₹250 crore for failing to take reasonable security safeguards and ₹200 crore for failing to notify a breach or for children's-data breaches. Consent failures are not a named entry — they fall under the residual "breach of any other provision," capped at ₹50 crore. The Schedule is published in full. Still a serious number, and worth getting right rather than repeating a figure that is four times too high.

Meta's own opt-in policy operates on top of all this and requires active opt-in naming your business and what the person is agreeing to receive. Because Meta enforces algorithmically off block and report rates, its rules bite faster than any Indian regulator will.

So where does the return actually sit?

Three functions, ordered by the economics rather than by anyone's engagement chart.

Inbound response. Cheapest per message even after October's change, highest intent, and the one place where speed demonstrably helps. The useful discipline is answering the actual question fast, not deploying a questionnaire — which is the argument we make at length in our piece on lead qualification.

Utility messaging tied to a real event. Order confirmations, dispatch updates, appointment reminders, payment links, document requests. Priced in the cheap band, genuinely wanted by the recipient, and therefore protective of your quality rating rather than corrosive to it. For retail and ecommerce businesses this is where the channel earns its keep; for education it is admissions logistics rather than admissions promotion.

Re-engagement, narrowly. A stalled conversation with a person who enquired is a reasonable use. A quarterly blast to a list you have had since 2023 is the expensive category, aimed at people who have forgotten you, with your quality rating as the stake. Note also that Meta tightened the marketing cooldown from one message per user per 24 hours to one per 48 hours in September 2025, which caps frequency regardless of your plans.

A word on the numbers you will be shown. Vendors quote 90-plus percent open rates and figures like "21 times higher conversion under five minutes." That 21x figure appears to be the qualification multiplier from a 2007 United States telephone lead-response study, recycled and relabelled as WhatsApp data — we traced it while writing our article on qualification. Treat both claims as unverified. The case for WhatsApp does not need them, because the case is in the rate card.

When WhatsApp automation is the wrong purchase

Four situations where we would tell you to spend elsewhere.

You are buying it to broadcast. If the plan is promotional volume to a large list, you have chosen the most expensive category on the platform and staked your number on how strangers react to it. Email or performance media will usually cost less and risk less.

Your contact list is old and unconsented. Under DPDP a large list without provable, specific, unbundled consent is a liability rather than an asset. Auditing and re-permissioning it is the first job, and it may shrink the list to a fraction — which is the correct outcome, not a failure.

You expect automation to reduce headcount. In Indian high-ticket categories — real estate, education, healthcare, considered B2B — conversations still need a person to close. Automation moves the work earlier and makes it faster; it does not remove the salary. Budget for both.

You are building on a platform you do not control. This is the honest structural objection and we would rather state it. Meta has changed the pricing model, started charging for a previously free category, and restricted what can be built on the platform — three times in eighteen months. Your unit economics and your automation layer both sit at the discretion of a company that has demonstrated it will move them, and the vendor selling you that layer carries none of that risk. Build on WhatsApp because the channel is where your customers are, not because you believe the terms are stable.

What we would do first. Establish which product you are on and what your current rates actually are per category. Audit your templates for anything that invites re-categorisation. Put the categorisation webhook on a monitor. Re-permission your list against DPDP's test and keep the evidence. Scope any assistant to your own business with a human handover. Then measure cost per outcome by category rather than messages sent — the general principle behind measuring marketing honestly.

Want your WhatsApp economics checked? Send us your monthly send volume by template type, your current provider and rates, how your opt-in is captured, and whether you are on the Business App or the Cloud API. We will come back with what the October pricing change did to your costs, which templates are at risk of re-categorisation, where your DPDP consent evidence is thin, and whether your automation plan sits inside Meta's permitted scope. If the honest answer is that you are being sold DLT compliance you do not need, we will tell you that too. Talk to Midgrow, or message +91 9632398896 with your sector and monthly volume. Pricing is set out in what agencies do not tell you, engagement structures are on our services page, and the content side sits with our social work.

Frequently asked questions

How does WhatsApp Business pricing work now?

Per delivered message, not per conversation. Meta deprecated conversation-based pricing on 1 July 2025, so you are charged for each message that reaches a device, at a rate set by the template category and the recipient's country. Any budget built on a flat 24-hour conversation fee is two pricing models out of date.

Are WhatsApp customer service replies still free?

No. From 1 October 2026, service messages and utility replies sent inside the 24-hour customer service window became chargeable per message, at rates varying by market and broadly aligned with utility and authentication rates. The window still exists; the replies inside it are no longer free. They remain the cheapest messages you can send.

Why do WhatsApp marketing messages cost more than utility messages in India?

Because Meta prices by category, and marketing sits in a far higher band. Market rate-card aggregations put Indian marketing messages around ₹0.86 against roughly ₹0.115 for utility and authentication — about a sevenfold difference. Those exact figures come from provider aggregations rather than Meta's published text, so confirm your own rates.

Can Meta change my template's category and increase my cost?

Yes. You submit a template but Meta's systems assign the category, so a utility template containing anything promotional — a discount code in a shipping update, for example — can be reclassified as marketing at the higher rate. You are notified via the template_category_update webhook, which is worth monitoring as a cost control.

What gets a WhatsApp business number banned?

Recipient behaviour, not volume. Quality rating is driven mainly by blocks and reports. Meta's template pacing sends a new template to a small sample first and pauses it automatically on negative feedback — three hours for a first pause, six for a second, permanent disabling on a third. Well-targeted high volume is safer than low-volume irrelevance.

Did Meta ban AI chatbots on WhatsApp?

Only general-purpose ones distributed by AI providers. The terms change announced in October 2025 took effect on 15 January 2026 and bars using the Business API where AI technology is the primary function rather than incidental. Businesses using AI to serve their own customers remain permitted — OpenAI, Perplexity and Microsoft were the parties whose assistants stopped working, not ordinary business support bots.

Do I need DLT registration for WhatsApp marketing in India?

No. TRAI's TCCCPR framework and its distributed-ledger template approval govern commercial messaging routed through telecom operators, and the Telecommunications Act, 2023 kept over-the-top services outside that perimeter. TRAI has consulted on extending regulation to over-the-top messaging but those remain proposals. Providers selling DLT compliance packages for WhatsApp are selling compliance with the wrong regime.

What is the penalty for WhatsApp messaging without proper consent in India?

Up to ₹50 crore. The DPDP Act's Schedule does not list consent failure separately, so it falls under the residual "breach of any other provision" at that cap. The higher figures often quoted — ₹250 crore and ₹200 crore — apply to security-safeguard failures and to breach notification or children's data respectively.