WhatsApp Reply Costs: Check Your API Invoice

WhatsApp API replies can add operating costs, especially when a bot sends many short messages. Provider notices describe service-message changes from 1 October 2026, including 1,000 free replies per number each month. Check the terms against your account, Meta’s rate card, and your provider contract before budgeting. The rupiah figures below are illustrative rather than universal rates.

Gupshup’s notice describes the changes. The dates and allowances below are provider-reported terms; confirm how they apply to your account. Obtain written confirmation for your account.

If your WhatsApp volume is promotional blasts, almost nothing changes. If it is support chat and order updates — which is most Indonesian businesses that ever bothered to connect the API — a line that read zero on every invoice for the last two years may now carry a cost.

What changes do providers report?

Meta sorts approved message forms into three categories — marketing (promotions), utility (receipts, reminders, order status), and authentication (one-time codes) — and prices each one differently. The provider notices cited here describe charges for two previously free message types. Confirm category rates against your current account terms.

The timeline matters if you are arguing with a vendor about who should have warned you. Meta notified providers on 10 August 2026, published the country rates on 1 September 2026, and the change takes effect 1 October 2026 — dates that Zendesk's provider notice lays out plainly: "Starting October 1, 2026, Meta will charge per message for non-template replies sent during this window."

The reported allowance is per phone number, not per company. respond.io's breakdown (published 3 September 2026) is specific: a business running several WhatsApp numbers gets 1,000 free on each number, the counter resets every calendar month, and unused allowance does not carry over.

What is a service message, and why was it free?

A service message is any message you type yourself rather than send from a pre-approved form — a plain text reply, a photo, a PDF, a row of quick-reply buttons. Meta calls the pre-approved forms templates, and every template has to be submitted and categorised before you can send it.

The 24-hour window is the part people forget. When a customer messages your business number, a timer starts. For the next 24 hours you can reply freely in your own words. Once it expires you can only send an approved template, which is why the "Hi, we noticed you didn't finish checkout" message always sounds like it came out of a form — it did.

Free-form replies inside that window have been free since November 2024. That is what ends. The window itself is unchanged; only the price of speaking inside it moves.

What does a reply cost in Indonesia?

Provider notices describe utility-country pricing for service messages. The table below uses ChatMaxima’s IDR figures as simulation assumptions, with an assumed effective tax of 11%. Actual costs can vary with billing currency, exchange rates, volume tiers, provider markup, and invoice tax treatment.

Use Meta’s rate card and your provider invoice as the final references. The examples assume all messages are delivered and qualify for the stated categories; check failures and allowance handling against delivery logs and billing reports.

Two details are easy to miss. Utility and authentication have volume tiers — the per-message rate drops as monthly volume crosses thresholds — while marketing discounts and rates must be checked against the current account rate card. And marketing costs 1.64× what utility costs, which makes a misfiled template category one of the more expensive clerical errors available to you.

Which price applies to which message?

How much does a chatty bot cost now?

Take an operator handling 200 conversations a day — a hotel front office, a property manager, a shop running orders over chat. That is 6,000 conversations a month.

Most chat flows built in the last two years were designed to feel human, which in practice meant many short bubbles: a greeting, a confirmation, the answer, a follow-up question, a closing pleasantry. Call it six typed replies per conversation.

• 6,000 × 6 = 36,000 service messages
• minus 1,000 free = 35,000 billable
• 35,000 × Rp356.65 = Rp12,482,750, or Rp13,855,853 with the assumed tax

Now rebuild the same flow to answer in two messages — one complete reply, one confirmation:

• 6,000 × 2 = 12,000 service messages
• minus 1,000 free = 11,000 billable
• 11,000 × Rp356.65 = Rp3,923,150, or Rp4,354,697 with the assumed tax

Same customers, same information delivered, roughly Rp9.5 million a month difference. None of that saving comes from a cheaper provider. It comes from deciding that "Baik kak 😊" is no longer a free gesture.

Why does message count suddenly matter more than message content?

Because the unit of billing changed and nobody redesigned around it. Under the old conversation-based model, a conversation cost what it cost regardless of how many times you spoke inside it, so splitting a reply into five friendly bubbles was free warmth. Per-message billing turns every bubble into a line item.

This hits automated flows hardest. A human agent types one paragraph. A bot sends an acknowledgement, the answer, and a separate menu — three messages that might fit in one. Typing indicators are a separate API feature; do not automatically count them as billable messages. The same bot is often wired to retry on delivery failure without checking whether the first attempt actually landed, which is how a webhook loop turns into a billing incident. The reliability patterns that stop duplicate sends are the same ones you already need for payments; I wrote them up in webhook reliability patterns.

What should you check on current invoices?

Seven items, in the order I would do them.

1. Count your last 30 days. Export the message log and classify every outbound message into four buckets: marketing template, utility template, authentication template, typed reply. Multiply the typed-reply count by Rp356.65 after subtracting the applicable free allowance per number, with a minimum billable count of zero. That is a simulation; reconcile it against delivered messages and actual billing reports.

2. Merge the bubbles. One reply carrying the booking code, the amount, the deadline, and the next step. Test whether guests can understand and act on the combined reply; fewer messages reduce this simulation's cost.

3. Review separate filler replies. Acknowledgements and closing questions can add messages. Count the items in delivery and billing reports; typing indicators are not automatically billable messages. Keep replies that help customers.

4. Audit your template categories. Meta assigns the category at approval, and a utility template with promotional wording gets recategorised to marketing. At 10,000 notifications a month, that mistake costs Rp2,296,800 extra under the rates assumed here. Read every approved template and strip the sales language from anything that is genuinely a receipt or a reminder.

5. Check entry-point exemptions. Click-to-WhatsApp ads or Facebook page buttons can qualify for pricing exceptions under specific conditions. Confirm eligibility, duration, and message categories in current account documentation rather than assuming the whole conversation is free.

6. Make one event send exactly one message. Give every outbound message a key derived from the event — booking ID plus event type, not a timestamp — and refuse to send twice on the same key. A retry storm used to waste bandwidth. Now it spends money.

7. Move the low-value traffic off WhatsApp. Not every notification deserves Rp395.88 with tax. Delivery-progress pings, marketing digests, and internal alerts belong in email, push, or your own dashboard. Keep WhatsApp for the messages people actually open within a minute — the reasoning behind that split is in WhatsApp automation for hotel operations.

What is not changing?

Marketing template rates stay where they are. The free WhatsApp Business app — the one you install from the Play Store and run on a phone — is untouched; this is a WhatsApp Business Platform change, so it only affects businesses sending through the API. Utility templates outside the window were already charged and stay charged. And the 24-hour window itself still works exactly as before.

One more: your provider's markup is separate from Meta's rate. Some resellers add a flat per-message fee on top of the rate card. Ask for the two numbers separately, then reconcile their invoice against your own message log for the first billing cycle. The month a pricing model changes is the month billing errors hide best.

The cost of a conversation is now a design decision

For two years the cheapest way to build a WhatsApp flow was also the chattiest one, and everyone built accordingly. That trade closed on 1 October. What replaced it rewards the thing good operators were doing anyway: answer completely, answer once, and stop sending messages nobody reads.

Count your replies this week. The next billing report will show the effect — the only question is whether you chose the number or inherited it.