From October 1, 2026, Meta starts charging for WhatsApp service messages —the free replies within the 24-hour window— per message, at rates equivalent to utility templates; and Meta AI agent messages are billed per token from August. For banks, every reply now has a cost and conversational efficiency becomes a financial metric. Five moves before October: measure cost per conversation, consolidate replies into a single message, use templates for the predictable, route by complexity and control where your AI runs.
Meta announced the most important change to WhatsApp API pricing since the per-conversation model: service messages are no longer free . Here's what you need to know and do before October 1. Meta is rolling out the change in three stages: July 1, 2026 — launch of the Meta Business Agent Platform, Meta's own AI agent platform. August 1, 2026 — messages generated by Meta AI agents start being billed per token consumed. October 1, 2026 — service messages (the free replies your team or bot sends within the 24-hour window) start being charged per message. For years the rule was simple: if a customer messaged you on WhatsApp, everything you replied within the next 24 hours was free. Banks built their support operations on that logic: the inbound conversation cost nothing, you only paid to start conversations with templates. That rule ends on October 1, 2026. What exactly changes Meta splits non-template replies into two categories, each with its own billing model: Message type Who generates it How it's billed Service messages Human agents or third-party AI (like your conversational platform) Per message, no volume discounts Meta Business Agent Meta's own AI Per token consumed The new service-message rates are equivalent to each market's utility and authentication template rates. ℹ️ What stays free: service messages within the 72-hour window of free entry points, i.e. conversations started from Click-to-WhatsApp ads or social-media CTAs. What it means for your operation The impact is direct: every reply your support team or bot sends now has a cost. An interaction that today generates 3 charges may generate 5 after October 1. For a banking operation handling tens of thousands of conversations a month, the difference is not marginal. To size it: in Brazil, processing 10,000 AI-generated replies can cost between USD 268 and USD 968 depending on query complexity. Multiply that by your real monthly volume and you have the new line in your channel budget. There's a deeper shift, more important than the number: conversational efficiency stops being an experience metric and becomes a financial one . An agent that resolves a query in 4 messages instead of 9 doesn't just deliver better service: it now also costs less than half. Five things you can do before October 1. Measure your cost per conversation today You can't optimize what you don't measure. You need to know how many outbound messages each type of query generates, which skills or flows are the most “talkative,” and what your projected bill will be under the new pricing. If your platform doesn't give you that visibility per conversation and per skill, that's the first problem to solve. 2. Consolidate replies instead of fragmenting them The classic chatbot pattern of sending 3 bubbles in a row (“Hi! 👋” + “Sure, I can help with that” + the actual answer) now costs 3 charges. Designing complete replies in a single message is the fastest optimization available. 3. Use templates for the predictable Repetitive, structured interactions (case status, confirmations, reminders) are best moved to utility templates, which have known and predictable rates, instead of resolving them with back-and-forth service messages. 4. Route by complexity Not every query needs generative AI. Simple, high-frequency queries can be resolved with deterministic automation (cheaper and faster), reserving AI agents for the cases where they truly add value. 5. Choose carefully where your AI runs With Meta charging per token on its own agent platform, architecture matters: controlling your AI stack (which model you use, how many tokens it consumes, how prompts are optimized) becomes a direct cost lever. How we solve it at Delto At Delto we've been designing for this scenario since before the announcement, because conversational efficiency has always been part of the banking business: Agents that resolve fast. They're designed to complete operations in as few turns as possible, with consolidated replies and flows that get to the point. Analytics with cost per conversation. You measure tokens, messages and cost per skill in real time, with dashboards and natural-language queries. You know what each type of interaction costs before Meta's bill arrives. Engage for smart outbound. Campaigns and notifications over utility templates with predictable rates, instead of improvising over service messages. Routing by complexity with Studio . Combine deterministic flows for the simple with generative AI for the complex: every query is resolved with the tool at the right cost. 💡 The question every bank should ask today: how much will my WhatsApp operation cost under October's pricing? If you don't have the number, it's time to calculate it. The takeaway WhatsApp is still the channel where your customers are — that doesn't change. What changes is that from October every message counts, literally . Banks that reach that date measuring their cost per conversation, with agents designed to resolve efficiently, will absorb the change without surprises. Those that don't will discover the impact on the bill. Want to project the impact on your operation? Write to us and we'll calculate it together with your real volumes.
What changes in WhatsApp pricing in October 2026? Service messages (the free replies your team or bot sends within the 24-hour window) are no longer free and will be charged per message, at rates equivalent to each market's utility and authentication templates.
Which messages stay free? Service messages within the 72-hour window of free entry points: conversations started from Click-to-WhatsApp ads or social-media CTAs.
How much can it cost? As a reference, in Brazil processing 10,000 AI-generated replies can cost between USD 268 and USD 968 depending on query complexity. Your real cost depends on your volume and how many messages each query generates.
How does a bank prepare? By measuring cost per conversation and per skill, consolidating replies into a single message, moving the predictable to utility templates, routing by complexity and controlling its AI stack (tokens and prompts).