You build a WhatsApp AI agent on the WhatsApp Business Platform (the API, not the regular Business app), connect it to an AI model that answers from your own information, give it tools to check your systems and take actions like booking or creating orders, and route conversations to your team through a shared inbox when a person is needed. The parts that trip people up aren't the AI. They're WhatsApp's own rules: opt-in, approved message templates, the 24-hour customer service window, and quality ratings that can limit how many people you can message.

This post covers the platform basics, how the agent handles sales, support and bookings, catalogs and payments, voice notes and regional languages, handoff to humans, and a sensible way to launch. It's especially relevant if you sell in India, Latin America, the Middle East, Africa or Southeast Asia, where WhatsApp is often the default way customers talk to businesses.

What's the difference between the WhatsApp Business app and the Business Platform?

Many small businesses start on the free WhatsApp Business app on a phone. It's fine for a person replying manually, with quick replies, labels and a basic catalog. It isn't designed for an AI agent: there's no proper API, it's tied to a device, and multiple staff can't easily share conversations at scale.

The WhatsApp Business Platform (often called the WhatsApp API, with the Cloud API hosted by Meta) is what you need for automation. You access it either directly through Meta or through a Business Solution Provider, companies such as Twilio, Gupshup, 360dialog, Interakt, WATI and many others, which add inboxes, template management and billing on top.

Setup involves:

  1. A Meta Business account, ideally verified.
  2. A phone number not currently active on the regular WhatsApp app (you can migrate an existing business number, with some steps).
  3. An approved display name.
  4. Message templates submitted for approval.
  5. A webhook endpoint where your agent receives incoming messages.

Meta's terms also matter for AI specifically. The platform is intended for businesses communicating with their customers, and Meta has restricted its use for offering general-purpose AI assistants. An agent that handles your sales, support and bookings is the intended use; check current policy before you build anything broader.

How do templates and the 24-hour window work?

This is the rule that shapes everything else:

  • When a customer messages you, a 24-hour customer service window opens. Within it, your agent can reply freely with any text, media or interactive message.
  • Outside that window, you can only start a conversation with a pre-approved template message, and only to people who have opted in to hear from you.
  • Templates are categorized, broadly as marketing, utility (order updates, appointment reminders, account alerts) and authentication (one-time passwords). Categories affect pricing and approval, and Meta reviews templates against its policies.

Pricing is set by Meta, varies by country and category, and has changed over time, so check the current rates with your provider rather than relying on old blog posts.

In practice, design your agent so that:

  1. Customer-initiated conversations are handled freely within the window.
  2. Follow-ups after 24 hours use a utility or marketing template with a clear reason, such as "your quote is ready" or "reminder of your appointment tomorrow".
  3. When a customer replies to the template, the window reopens and the agent continues the conversation.

Ignoring these rules leads to rejected messages, blocked numbers and a falling quality rating, which can restrict how many customers you can message per day.

What does a WhatsApp AI agent do for sales, support and bookings?

Here's how the main use cases break down:

Use caseTriggerWhat the agent doesSystems it touches
Lead capture and qualificationClick-to-WhatsApp ad, website button, QR codeGreets, asks qualifying questions, answers product questions, books a call or visitCRM, calendar
Product enquiries and ordersCustomer asks about a productShares catalog items, checks stock and price, builds a cart, sends a payment linkCatalog, inventory, e-commerce or ERP
Order status"Where is my order?"Looks up the order, shares tracking, handles address changes within rulesOrder management, courier APIs
Bookings"Can I come in Saturday?"Offers real slots, books, sends confirmation and a reminder template laterBooking or practice management system
SupportComplaint, question, photo of a problemAnswers from your knowledge base, collects details, creates a ticket or escalatesHelpdesk, knowledge base
Payment follow-upInvoice dueSends a utility template reminder with a payment link, records promises to payAccounting system, payment gateway

The AI lead qualification and AI customer support agent posts cover the logic of those workflows in depth; WhatsApp is the channel they run on.

How do catalogs and payments work on WhatsApp?

WhatsApp supports product catalogs connected through Meta's commerce tools, and interactive product messages that show one or several items with images and prices. Customers can add items to a cart and send it to you as an order message. Your agent then confirms availability, delivery and total, and handles payment.

Payment options depend on the country. In some markets, such as India and Brazil, payments can be completed inside WhatsApp through supported providers. Elsewhere, the usual approach is a secure payment link from your gateway, with the agent watching for the payment confirmation webhook and updating the order.

Two rules: the agent should never ask for card numbers, CVVs or banking passwords in the chat, and every price it quotes should come from your live system, not from the model's memory of a catalog that's since changed.

How does the agent handle voice notes, photos and regional languages?

WhatsApp conversations are not just text, and a good agent is built for that:

  • Voice notes. In many markets customers send voice messages far more than typed ones. The agent should transcribe them, respond in text (or voice, if that suits your customers), and flag low-confidence transcriptions.
  • Photos and documents. A photo of a damaged product, a prescription, a meter reading, an ID document or a handwritten list. Vision-capable models can read and classify these, but anything sensitive should be handled under strict data rules and checked by a person where it matters.
  • Location pins for delivery or service visits.
  • Languages. Customers write in Hindi, Tamil, Hinglish, Portuguese, Arabic, Bahasa or Spanish, often mixed with English and often in Roman script. Test with real conversations from your customers, not translated samples. Our guide to multilingual AI agents for Indian languages goes into the specifics.

Interactive elements help too. Reply buttons and list messages make choices quick and reduce typing errors, and WhatsApp Flows can collect structured information like booking details in a form-like screen without leaving the chat.

How should a WhatsApp AI agent hand off to your team?

Customers on WhatsApp expect a person to be reachable, and the handoff needs to happen inside the same chat. The pattern:

  1. Clear triggers: the customer asks for a person, the agent can't resolve the issue, the request is outside its limits, or the customer is upset.
  2. Assignment in a shared inbox, where staff see the full conversation, the AI's summary, and linked records from the CRM or order system.
  3. The agent stops replying once a human takes over, and only resumes if the human hands back.
  4. Honest expectations: "I've passed this to our team. They're available until 7 pm and will reply here."
  5. Window awareness: if the human can't reply within 24 hours of the customer's last message, they'll need a template to reopen the conversation.

Make sure your team's inbox works on mobile, because in many small businesses the person taking over is on the shop floor, not at a desk.

What can go wrong with a WhatsApp AI agent?

  • Over-messaging. Too many marketing templates lead to blocks and reports, which lower your quality rating and messaging limits. Keep outbound messages useful and expected.
  • No real opt-in. Messaging people who never agreed to hear from you on WhatsApp breaks platform policy and, in many places, privacy law.
  • Hallucinated prices or stock. Always look up live data.
  • Losing the human touch. In relationship-driven markets, regular customers may expect to talk to the owner or a known staff member. Let them.
  • Data handling. Conversations contain personal data, sometimes sensitive. Decide retention, access and where data is processed, and check obligations under laws like India's DPDP Act or GDPR. If you're an Indian SME weighing all this, our guide to AI for Indian SMEs covers the local context.

How do you launch a WhatsApp AI agent?

  1. Pick one high-volume use case, such as order status or booking, and count how many such conversations you handle now.
  2. Set up the Business Platform through Meta or a provider, verify the business and submit your first templates.
  3. Connect the systems the use case needs: catalog, orders, calendar or CRM. Our guide to AI agent integrations with CRMs and ERPs covers the patterns.
  4. Run with human approval on replies for the first week or two, then let the agent reply on its own for the intents it handles reliably.
  5. Expand to the next use case, and add outbound templates once inbound is stable.

Frequently Asked Questions

Can I use the free WhatsApp Business app for an AI agent?

Not properly. The app isn't designed for automated agents or multi-user access at scale. For an AI agent, use the WhatsApp Business Platform through Meta's Cloud API or a Business Solution Provider.

Can a WhatsApp AI agent send messages to customers first?

Only using pre-approved template messages, and only to customers who have opted in. Free-form replies are allowed within 24 hours of the customer's last message.

Can customers pay inside WhatsApp?

In some countries, including India and Brazil, in-chat payments are supported through approved providers. Elsewhere, the agent sends a secure payment link from your payment gateway and confirms the payment when it arrives.

Does a WhatsApp AI agent understand voice notes?

It can, if it's built to transcribe them. Voice note handling is important in markets where customers prefer speaking to typing, and quality should be tested in the languages and accents your customers use.

How long does it take to build a WhatsApp AI agent?

Platform setup and template approval can take days to a couple of weeks, depending on verification. A focused agent for one use case with one or two integrations typically takes a few weeks to build and test; multi-use-case agents with payments take longer.

Want your customers' favorite channel to work around the clock?

If your team is drowning in WhatsApp messages or you want to sell and support on the channel your customers already use, tell us what your WhatsApp conversations look like today. We'll map the platform rules to your use case and build an agent that hands off to your team cleanly. See our agentic AI development work.