AI Agents on WhatsApp Business: Requirements, Costs and Use Cases

Running an AI agent on WhatsApp Business means connecting the WhatsApp Business Platform — Meta's official API, not the desktop app — to an agent that answers questions, takes orders, books appointments and chases payments, with access to your real systems and a clean handover to a human when needed. In Spain and much of Latin America and Southern Europe it is the highest-return channel decision an SME can make, for a boring reason: it is where the customers already are.

The part almost nobody explains properly is the commercial mechanics. Meta changed its billing model in July 2025 and now charges per message delivered rather than per 24-hour conversation, and template categories carry very different prices. Understanding that before you design the agent is the difference between a channel that pays for itself and a monthly invoice that surprises everyone. Here is what matters, with figures verified in August 2026.

What do you need to run an agent on WhatsApp Business?

Four things, and none of them is the agent. The agent is the easy part; what delays these projects is always Meta's paperwork, so start it on day one.

1. A verified business account with Meta. Business verification with real corporate documentation. This is the slow step: plan in days, not hours.
2. A WhatsApp Business Account and a phone number that is not registered on the consumer WhatsApp app. If the number you want is the shop line that someone currently runs from a mobile, plan the migration carefully.
3. API access, either directly through Meta's Cloud API or via a business solution provider (BSP). A BSP simplifies onboarding and adds a human agent inbox, in exchange for a per-message markup.
4. Approved templates for anything you initiate. Meta reviews and approves every template and assigns it a category, and the category sets the price.

That fourth point shapes the entire design. Outside an open conversation you do not write whatever you want — you send a pre-approved template. Inside the customer service window, by contrast, the agent converses freely. Designing a good WhatsApp agent is largely the art of opening that window for a legitimate reason and resolving everything inside it.

What does it cost to send WhatsApp messages in 2026?

Since 1 July 2025 Meta bills per delivered message rather than per 24-hour conversation. The price depends on the recipient's country and the template category, and free-form replies inside the customer service window are free. This is the structure in force as of August 2026:

For Spain, provider rate cards published in August 2026 put marketing messages at roughly €0.05 and utility and authentication in the €0.015-0.02 range, plus your BSP's markup if you use one. Treat those as budgeting orders of magnitude rather than official prices: Meta raised the Spanish marketing rate effective 1 July 2026 and has announced further rate card updates from 1 October 2026, so always check the current card before committing to a number.

The practical consequence of this pricing structure is blunt: the money is in the service window. An agent that answers well within 24 hours handles those conversations for free, while a company that only uses WhatsApp to blast campaigns pays for every message with no volume discount — because Meta priced it deliberately to discourage broadcasting.

Which use cases actually work?

Four deliver consistently, listed here by ease of implementation.

• Support and FAQs backed by real data. Not a button menu: the agent checks the order status, balance, appointment or availability in your system and answers with the actual figure. That is what separates this from a decorative chatbot, as we set out in our guide to automating customer service with AI agents.
• Appointment booking. Checking real availability, confirming, sending reminders and rescheduling in one message. WhatsApp reminders cut no-shows far more effectively than email, for the simple reason that they get read.
• Orders and repeat purchases. Especially in businesses with recurring orders: the customer writes "the usual" and the agent knows what that means, checks stock and places it.
• Payment reminders and collections. A utility template, a neutral tone, a payment link and an open conversation for disputes. It outperforms email for one reason: it gets read the same day.

The first two usually justify the project on their own. The fourth is the one that shows up fastest in cash.

Button chatbot or agent?

The difference is not sophistication, it is systems access. A button chatbot walks a decision tree somebody programmed, and when the customer steps outside the tree it stalls. An agent understands the request in natural language, queries the real system, decides the next step and executes it.

In practice the metric that separates them is the share of conversations resolved without a human. Well-built button trees typically stay below 40%; an agent connected to the systems goes considerably higher and — more importantly — escalates with full context instead of leaving the customer to repeat the problem from scratch. If you are choosing between the two, we compare cost and scope in how much a business chatbot costs.

What NOT to do on WhatsApp

WhatsApp is the most intimate channel your customer has. Anything that feels annoying by email feels invasive here, and the penalty is immediate: blocks, reports and a drop in your number's quality rating, which in turn lowers your messaging limits.

• Mass campaigns without explicit consent. Neither lawful under GDPR nor profitable: marketing is the most expensive category and the one that triggers the most blocks.
• Hiding that it is an agent. Introduce it as an assistant in the first message. Beyond being the right thing to do, EU AI rules require disclosing that the person is interacting with an AI system.
• Leaving customers with no way out. "Talk to a person" must always work, and work quickly during business hours.
• Automating serious complaints or negotiations. Detect and escalate instantly, yes; answer automatically, no.
• Sensitive data in the thread. Medical records, full banking details or sensitive documentation never belong in a chat thread. Link out to a secure environment.

How it gets built, step by step

1. Start Meta verification on day one. It is the project's critical path.
2. Decide the number. A new one, or a planned migration of the current one with customer notice.
3. Choose direct Cloud API or a BSP depending on whether you have technical capacity and whether you need a human agent inbox.
4. Design templates by category before writing the agent: decide which are utility and which are marketing, because that is where your monthly cost comes from.
5. Connect the agent to the real systems: orders, calendar, CRM, payments. Without this you do not have an agent, you have a chatbot.
6. Define human escalation with explicit criteria and coverage hours.
7. Measure at 4-6 weeks: conversations resolved without a human, first response time, cost per conversation and the number's quality rating.

A scoped deployment of this kind takes 4-8 weeks and sits in the usual custom agent range for an SME, €15,000-40,000 depending on integrations, plus 10-20% annual maintenance. We run our own businesses this way and the pattern repeats: the channel pays for itself through service, not through campaigns. If you would rather work it through with a team that builds these channels every week, that is what we do at our AI agency in Madrid; and if your focus is commercial rather than support, the fit is covered in AI agents in sales and CRM.

Frequently asked questions

Can we use the regular WhatsApp Business app instead of the API?

Not for an agent. The free app does not allow integration with your systems or real automation beyond quick replies and away messages. It is perfectly fine for a small business with one person answering, but the moment you want automatic lookups of stock, calendars or orders, you need the WhatsApp Business Platform.

How much will it cost per month?

Almost entirely down to the mix between service messages (free) and templates you initiate. A business that uses WhatsApp to answer and confirm usually lands at modest figures; one running recurring marketing campaigns can multiply that several times over. Do the maths with your real volume by category before you start, and revisit it after the rate changes announced for October 2026.

Do customers accept talking to an agent on WhatsApp?

Readily, as long as the agent resolves things. What generates resistance is not talking to an AI, it is talking to an AI that knows nothing about your order and returns you to the main menu. Transparency helps: identifying as an assistant and offering a human handover at any point raises satisfaction rather than lowering it.

What happens if our number gets blocked a lot?

Meta maintains a quality rating per number and tiered messaging limits. If quality falls due to blocks and reports, your sending limits drop and in extreme cases the account is restricted. It is another reason not to treat the channel as a promotional loudspeaker: the damage is operational, not just reputational.

Can it integrate with our CRM and ERP?

Yes, and that is exactly what makes the channel useful. The agent reads from and writes to the systems you already use through their APIs, so the WhatsApp conversation is logged against the customer record and the actions — order, appointment, payment — happen where they should. If your systems have no API, solving that access is phase one of the project.