Outbound calls with an AI agent: what is permitted and worthwhile for voice campaigns
Looking for a platform for outbound voice campaigns with an AI agent? What the law requires, which categories exist and which use cases pay off.
Opt-in, templates, the 24-hour window: the rules for proactive WhatsApp messages and what an AI agent makes of them. With five use cases for SMEs.

Appointment reminders, status updates, follow-ups on missing documents: in many businesses this still runs by phone and email, and a good part of it never reaches the customer. On WhatsApp, by contrast, messages almost always get read. The real question is therefore how a business is allowed to actively reach its customers there and what an AI agent for proactive customer outreach via WhatsApp actually takes on in the process.
An AI agent for proactive customer outreach via WhatsApp sends pre-approved messages to customers who have consented to the channel, understands their replies and triggers follow-up actions from them: it reschedules the appointment in the calendar, updates the order status in the system or creates a task for an employee. That sets it apart from a newsletter tool that only sends and never listens.
One thing up front: proactive does not mean spam. Without documented customer consent, nothing works on the WhatsApp Business Platform, and Meta enforces that technically. Anyone using the channel seriously sends occasion-based messages with clear value for the recipient, not mass advertising.
Three fixed rules from Meta apply to proactive messages over the WhatsApp Business Platform: recipient consent, approved message templates and the 24-hour window. You collect the consent (opt-in), for example when taking an order, in the contact form or directly at the counter, and you document it. Every message with which your business opens a conversation must be submitted to Meta as a template beforehand and approved, sorted into a category such as Marketing or Utility. Meta can reject templates or block them retroactively.
Once the customer replies, a 24-hour window opens in which you may write freely, without a template. Until 30 September 2026 these replies cost nothing extra; from 1 October 2026 Meta will charge for them individually (see the note below). It is precisely in this window that the AI agent works.
On costs: since July 2025, Meta charges per delivered template message, no longer per conversation. A marketing message to Germany costs roughly 11 to 14 cents according to Meta's price list (as of July 2026); utility messages such as appointment reminders or status updates are well below that, at 0.0456 EUR. Replies within the open 24-hour window are free until 30 September 2026; from 1 October 2026 Meta will charge for them per message as well. On top of that come the fees of the technical provider through which you use the platform. For a few hundred appointment reminders a month, that stays a two-digit euro amount.
Meta is changing how it bills: from 1 October 2026, all service messages and all utility messages you send to a customer as a reply within the open 24-hour window will cost money. Both are free today, service messages since 1 November 2024, replies within the window since 1 July 2025. That means the dialogue the AI agent conducts will cost money in future too, not just the template that opens it. Meta has not yet published the rate: the price list is announced for 1 September 2026 at the latest, and the column for service messages currently reads n/a. Only the mechanism is known so far: service messages cost the same as utility and authentication messages in the same market, currently 0.0456 EUR per message in Germany. All cost figures in this article apply until 30 September 2026.
In data protection terms, the platform runs through Meta as a processor; Meta provides a data processing agreement (DPA) for this. You need the customers' consent, a notice in your privacy policy, and you should keep sensitive content such as health data out of the channel. This is context, not legal advice: if in doubt, the topic belongs with your data protection officer or lawyer.
The strongest use cases are occasion-based messages the customer can respond to directly. Five examples from practice:
The AI part is not in the sending but in understanding the reply and triggering the right follow-up action. If the customer writes in response to an appointment reminder, “Thursday doesn't work for me, does Friday afternoon work too?”, the agent recognises the intent, checks the free slots in the calendar, proposes two appointments and enters the confirmed one directly. Without AI, the same reply lands in an inbox, and an employee has to step in after all.
Technically, the agent hooks into your existing systems via a workflow: calendar, CRM, order management. It reads the occasion for the message there and writes the result of the dialogue back. How such AI agents are built and where they pay off, we have described on our services page.
The approved template opens the conversation, the AI agent conducts it. It understands the customer's reply and completes the follow-up action directly in the calendar or CRM. Without this part, you are just sending text modules and the work stays undone.
Important for keeping control: a well-built agent knows its limits. If it does not understand a reply or the customer asks for something outside its remit, it hands over to a human and logs the conversation. You can see at any time what the agent has written and triggered.
The most sensible start is a single use case with a measurable effect, in most businesses the appointment reminder. The path there: set up a WhatsApp Business account through a technical provider, collect consents cleanly, have a utility template approved and connect the agent to the calendar via a workflow. Once this case runs stably, status updates and the waiting list follow. You save reactivation with an offer for last, because it needs the strictest opt-in.
WhatsApp does not replace the phone here, it relieves it. Anyone who also wants to catch the calls that come in anyway combines the chat channel with a voice agent. How that works is shown on our page about AI telephony.
No. For every conversation started by the company, Meta requires prior consent from the recipient, and the GDPR also requires a legal basis. Actively obtain the opt-in, for instance when taking an order or via a form, and document it.
Since July 2025, Meta charges per delivered template message, tiered by category and recipient country. To Germany, a marketing message costs roughly 11 to 14 cents according to Meta's price list (as of July 2026), utility messages much less at 0.0456 EUR. Replies within the 24-hour window are free until 30 September 2026; from 1 October 2026 Meta will charge for them individually as well.
From 1 October 2026, Meta will charge for two message types that were free until now: all service messages and all utility messages you send to a customer as a reply within the open 24-hour window. That affects exactly the dialogue the AI agent conducts, not just the template that opens it. Meta has not yet published the rate; it is announced for 1 September 2026 at the latest. Only the mechanism is certain so far: service messages are to cost the same as utility and authentication messages in the same market, currently 0.0456 EUR per message in Germany. If you are planning the channel now, it is best to calculate the dialogue from October at that rate.
For manual individual chats, yes; for automated proactive outreach, no. An AI agent needs the WhatsApp Business Platform (API), which is connected through a technical provider. Only through it can templates be sent automatically and replies processed by machine.
Via the Business Platform, data-protection-compliant use is feasible: Meta acts as a processor and provides a DPA, you need consents and a passage in your privacy policy. Sensitive data does not belong in the channel. To assess your specific case, ask your data protection officer; this article does not replace legal advice.
That depends on how many calls and no-shows the channel replaces. A business that reminds about appointments daily and answers status questions saves phone time from the first month on, while message costs stay in the two-digit euro range. The biggest lever is held by businesses with many appointments and a high no-show rate.
NordFlux builds digital employees for organisations: automations and AI agents that take over repetitive work. You stay in control.
Looking for a platform for outbound voice campaigns with an AI agent? What the law requires, which categories exist and which use cases pay off.
From October 1, 2026, Meta will charge for every reply within the 24-hour window. What this means for automated customer communication and what is still unclear.
Agentic Commerce means AI agents buy on behalf of your customers. What data and interfaces your website needs so an agent can complete a purchase with you.
In the free initial analysis we look at your appointments, calls and follow-ups and tell you honestly whether a WhatsApp agent pays off for you.