01 — What happened
On 1 October, Meta resumed charging for two kinds of WhatsApp message that had been free.
The first is the service message: a plain, non-template reply sent inside the 24-hour window that opens when a customer writes to you, free since 1 November 2024. The second is a utility template sent inside that same open window, free since 1 July 2025. Both are now billed per message, on delivery.
There is a free tier. Each business phone number gets 1,000 service messages a month, charged from the 1,001st, with no roll-over and a monthly reset. Beyond the tier, service messages are priced on the recipient’s country rate card, and Meta updated forty of those rates on the same date. Businesses without a payment method on file by 30 September keep delivery inside the free tier and lose it beyond.
Separately, on 1 July 2026 Meta introduced the Meta Business Agent — its own AI answering on a business’s behalf, billed per token at a single global rate of $2.00 per million tokens, which Meta estimates at roughly four to five cents a message. No free tier, no volume tiers.
Three things did not change. Messages from customers to businesses are still free. You are charged only on delivery. And the free entry point window still applies: when someone reaches you by tapping a Click to WhatsApp ad in the mobile app, that window may stay open for up to seven days, with marketing, utility, authentication and service messages all free inside it. Business Agent messages are charged even there.
This is the WhatsApp Business Platform — the API businesses connect to a CRM or help desk; Wati, a Meta solution provider, notes it does not touch the free WhatsApp Business app used on its own. And any guide written before last month will tell you replying inside the 24-hour window is free. It was. It is not now.
02 — Why it matters
The marginal cost of talking to a customer moved from zero to a positive number.
That sounds small. It is not. For two years the rational design under a free channel was to talk more: more follow-ups, more check-ins, more bots that ask one question at a time because the extra turns cost nothing. Every one of those turns is now a line on a bill.
The free tier matters more than the rate. It is a threshold, not a discount. Below 1,000 service messages a month nothing has changed. Above it, every additional message is billable at full price.
The third point is the one most people will miss. The cheapest reply and the most expensive reply now sit on the same screen. A reply typed by a person — or generated by your own AI through the API — is a service message: free inside the tier, then standard rate. Meta’s Business Agent is billed per token from the first message, with no free tier, and is charged even inside the ad-entry window where everything else is free. On the two things you can check without opening a rate card — the free allowance and the ad window — the platform’s own built-in AI is the most expensive way to answer a customer.
None of this is unusual. It is the same shape as software vendors moving from per-seat to per-resolution pricing: a unit of work that used to be bundled gets counted, once there is enough volume to be worth counting.
03 — The economics
A fixed free allowance means every saving comes off the top, and the top is the only part you pay for.
Take an illustrative service business handling 400 customer conversations a month on WhatsApp. Assume each takes six replies from the business side: acknowledge, ask for detail, confirm, quote, chase, close. That is 2,400 service messages: a thousand free, fourteen hundred billable.
Now assume conversation design cuts it to three replies: a structured intake step collects the detail in one go, the quote goes out as a template, the confirmation fires automatically. Total 1,200 messages, 200 of them billable.
Halving the message count cut the billable volume by 86%.
The leverage runs the other way too. At 2,400 messages a 10% reduction removes 240, all billable — a 17% cut in cost. The same 10% at 900 a month saves nothing, because nothing was charged. The budgeting question is your volume relative to 1,000.
The rate itself will get more attention than it deserves. Service prices follow each market’s rate card and vary widely by country; pull your own figure from Meta rather than a blog post. What does not vary is the Business Agent: one global price, per token, from the first message. Whatever your market’s service rate, the Business Agent’s cost curve starts at the origin and the service message’s does not.
As to why now: Meta’s Family of Apps “other revenue” — the line that holds paid messaging — was $1,007m in the second quarter of 2026 against $583m a year earlier. Meta does not break WhatsApp messaging out within it, and the results release gives no explanation for the rise. The direction tells you what kind of line this is. Metering is how it grows.
The free entry point window is the exception that explains the rule. A conversation beginning with a tap on a Click to WhatsApp ad is free for up to seven days, across every category except Meta’s own agent. Meta has made the conversations it already earns from the cheapest ones to have; organic inbound — somebody who saved your number from a card or your website — is the expensive one. That is consistent with everything else the platform has done this autumn, including putting its own page between your ad and your website.
04 — What businesses should do
Find out which side of 1,000 you are on. Count service messages per business phone number, not conversations, for a full month. If you are under a thousand and not growing quickly, this is not a project — confirm a payment method is on file so delivery does not stop at the tier, and go back to work.
If you are over, count replies per conversation before you count money. Find the median number of business-side messages it takes to resolve a request. If that is six and could plausibly be three, you have a conversation design problem that has acquired a price. The price is the smaller issue.
Collect structured information once, and template the predictable. The commonest cause of a long thread is asking for five things one at a time; a form, a booking link or an in-chat flow replaces four billable replies with one. Order updates and confirmations belong in templates — billed too, but one that arrives complete beats four messages of back-and-forth.
Decide the AI question on your own numbers. Build an agent against the API and its replies bill as ordinary service messages, inside the free tier. Meta’s Business Agent is far easier to switch on, and has no free allowance at all. At low volumes it is the more expensive choice; at high volumes with short answers it may not be. Model both against your actual thread lengths.
Do not change channel in a hurry. Email, SMS and the telephone all cost something, in fees or staff time. The number that matters is cost per resolved request, not cost per message — the distinction that makes hours saved a poor way to value an automation.
05 — Where technology fits
For most businesses reading this, the honest answer is nothing at all. Under a thousand replies a month this change costs you zero, and a project to optimise a bill of zero is a bad project.
Above the threshold, the useful work is rarely an AI layer. In rough order of return:
A structured intake step — a form, a booking link, a short in-chat flow — replacing the question-and-answer sequence at the start of most threads. Ordinary web work, and it removes more messages per unit of effort than anything else.
An integration that writes the conversation into your CRM, so the next person to pick it up does not re-ask what the customer already answered. Repeated questions are the second commonest cause of long threads, and they are a systems problem rather than a messaging one.
Then, last, an AI agent — for a question set you can name, have counted, and have watched a person answer the same way fifty times.
An agent that takes five messages to settle what a form settles in one has made the bill worse while looking like progress.
06 — RevenueStack’s take
Our view: the rate is not the story, and the bill is not the real loss. The habit is.
Two years of free service messaging taught a lot of businesses to build customer operations on a cost that was always provisional. Free channels carrying commercial volume get metered. Organic reach, click-to-message, API calls, model tokens — the sequence is the same every time, and it arrives as a date and a rate card rather than a conversation.
So build as though the meter is coming, because on any channel you do not own, it is. What you own is the structure that keeps a conversation short, the record of what the customer already told you, and the system that makes the next reply unnecessary. Those survive a rate card change. An operation that depends on messages being free does not.
And watch 1,000, not the per-message price. A free tier is a statement about who the pricing is aimed at, and this one sits where it exempts the smallest businesses and bills everyone who has built a real operation on the channel. Rates get adjusted quietly and often. Free tiers get moved deliberately.
Sources
- WhatsApp Business Platform — Pricing — Meta, 2026-10-21
- WhatsApp Business Platform — Updates to Pricing — Meta, 2026-10-21
- Meta Reports Second Quarter 2026 Results — Meta, 2026-07-29
- WhatsApp Service Message Pricing Changes Explained (2026) — Wati, 2026-08-13
