What you actually pay for WhatsApp Business messaging

Most of the confusion about WhatsApp costs comes from explanations that were accurate once. The older model was conversation-based: sending a template opened a conversation window, and every message exchanged inside that window was covered by a single flat fee. Under that model, batching messages into an open window was a genuine cost-saving tactic.

That model no longer applies. WhatsApp Business Platform pricing changed structurally on 1 July 2025, when Meta moved from conversation-based pricing to per-message pricing. Charges now attach to individual delivered template messages rather than to a window that covers an unlimited number of them. Anything written or half-remembered from before that date describes billing that has been retired, which is why a business can follow perfectly sensible advice and still get an invoice it did not predict.

The practical implication is direct: you can no longer estimate a campaign by multiplying an assumed per-conversation rate by a recipient count. Two templates sent to the same person are two billable messages, not one conversation. Any spreadsheet still built on the older assumption will understate the cost of a sequence and overstate the saving from bundling messages together.

A charge is triggered when Meta successfully delivers a template message to a recipient's device — not when the business presses send. If a message fails to deliver because the number is wrong, has been deactivated, or no longer has WhatsApp attached to it, there is nothing to bill. This is why a broadcast's real cost is a function of how many recipients it reached rather than how many it was addressed to, and why a delivery report is the document that explains an invoice.

Free-form replies sent inside an open customer service window are a different case. They are session messages, not templates, and they have not been billed under the template categories at all. A business answering a customer's question inside that window has not been paying per reply, which is a large part of why keeping conversations alive inside the window has been such favourable economics.

The same is true of qualifying free entry points. A customer who arrives by clicking a Click-to-WhatsApp advertisement opens a window that Meta treats favourably, so the resulting conversation does not carry the cost profile of an outbound template send. The advertisement itself costs money, of course, but that is ad spend on a separate line rather than a messaging charge.

Each template message is billed according to its category, and the category — not the length of the message or the number of variables in it — is what determines the rate. Marketing covers promotional content and sits at the higher end. Utility covers transactional messages that follow from something the customer already did: confirmations, dispatch notices, receipts, reminders. Authentication covers one-time passcodes. Service refers to those free-form in-window replies, which are not templates.

Rates also vary by the recipient's country, which matters more than it first appears. A contact list spanning several markets does not have a single per-message price, so an estimate built by multiplying one rate across the whole list will be wrong in proportion to how mixed the list is. The three inputs that actually determine a send's cost are the number of recipients reached, the category of the template, and the countries those recipients are in.

Meta also operates volume-based tiers that can improve per-message rates for utility and authentication messages as sustained usage grows. These are a different system from the messaging limits that cap how many unique people a business may message, and the two are easy to confuse. A messaging limit increase is not a discount, and reaching a better rate tier does not raise the ceiling on how many customers can be contacted.

Two things that have been free stop being free. From 1 October 2026, service messages — the free-form replies sent inside the open customer service window — and utility templates sent inside that same window begin carrying a charge. Utility templates sent outside the window are already billed under the per-message model, so the change is specifically about the messages that have been free because they happened inside a live conversation.

The date is confirmed. The rate is a separate matter: Meta has said rates would be published by 1 September 2026, roughly a month before the change takes effect. Any specific figure circulating before that rate card appears is an estimate, and a business planning around one is planning around a guess. It is also unsafe to assume the new charge will mirror the existing out-of-window utility rate, because an in-window message and an out-of-window template are functionally different sends today and there is no published basis for expecting identical pricing.

The useful preparation is not restructuring workflows around an unknown number. It is measuring how many service messages and in-window utility templates your own number actually sends in a typical month. That figure is checkable today, and it is the input the eventual rate will be multiplied against. Cutting off free-form replies before the charge exists gives up a real capability against a price nobody has seen.

A defensible estimate works forward from three verifiable inputs rather than backward from a remembered rate. Take the recipient list and break it down by country. Decide the category of each template in the sequence. Count how many separate template messages each recipient will actually receive, since under per-message billing a three-step sequence is three charges. Then look up Meta's current published rate for each country-and-category pair on the day you are estimating.

The last step is the one most often skipped, and it is the one that keeps the estimate honest. Meta revises rates with advance notice measured in months rather than years — as little as one month's notice for a rate card update by its own stated policy — so a rate checked a few campaigns ago may already be stale. Building the lookup into the launch routine, rather than treating price as a constant learned once, is what separates a bill that matches expectations from one that does not.

It is also worth separating messaging cost from total cost when deciding whether a campaign worked. Ad spend that drives Click-to-WhatsApp conversations, the staff time to answer the resulting conversations, and the per-message charges are three different numbers, and a campaign judged on only one of them is being judged incompletely.

Rates move; a few mechanics have not, and those are the ones worth building a process around. Charges apply to delivered messages, not attempted ones. The category a template is submitted under determines its billing treatment, and that category is enforced against the content rather than accepted on the label. Free-form replies inside an open window are treated differently from template sends outside one. Those three have held through the shift from conversation-based to per-message pricing, and they will survive the October change too — what changes then is which messages are billable, not how billing is structured.

So the durable version of cost control is not a set of figures to memorise. It is a short discipline: choose the honest category before submitting a template, count the actual number of messages a sequence sends rather than the number of people it targets, check the current rate for the specific markets involved before any send at real volume, and reconcile against the delivery report afterwards rather than against the send count.

Meta's own pricing documentation is the only source worth trusting for the rupee figure on the day a campaign goes out. Every summary of it, including this one, is a description of a structure rather than a quotation of a price.