Yes, you can create an AI agent for WhatsApp for your business — and you don't need to build a new app to do it. What you do need is a clear picture of three things: how the flow works, what the WhatsApp Business API actually requires, and the one rule that decides most of your cost and your user experience, the 24-hour window.
This article is for teams that already know they want an agent answering on WhatsApp and now need to understand the mechanics: how it is mounted, what it needs, what it can resolve and what it really costs.
A practical walkthrough of an AI agent on WhatsApp: the flow, the WhatsApp Business API requirements, the 24-hour window and templates, what it can solve and what it costs. Reading time: 7 minutes.
In this article:
01 How the flow works
An AI agent on WhatsApp is not a chatbot living inside the WhatsApp app. It is a small system that sits between the customer's phone and your business. Understanding the path a single message takes explains almost every other decision, from requirements to cost.
The sequence looks like this:
1. The customer writes. A message arrives on your WhatsApp Business number.
2. Meta delivers it. The WhatsApp Business Platform forwards the message to your server through a webhook.
3. Your agent interprets it. The message reaches the orchestration layer. This is where the agent decides what the person wants: an FAQ, an order status, a booking, a complaint, a sales enquiry.
4. It gathers context. The agent queries your systems — a CRM, an ERP, a product catalogue, a calendar — for the data it needs. Without this step, the agent can only answer generic questions.
5. The model composes a reply. The language model turns the intent and the retrieved data into a natural answer.
6. The response goes back. It travels through the same API and lands in the customer's chat, usually in seconds.
Two design choices in this flow matter more than the rest. The first is where the boundaries are: what the agent is allowed to answer on its own, what it must escalate to a person, and what it must never touch. The second is whether the agent has real context — whether it runs on a platform where the agent already knows your data. One with no access to your records is a polite FAQ page; one connected to your data can resolve the actual request.
That is the difference between a WhatsApp bot and the agent on WhatsApp — a system that acts on your business information, not just a script that replies. The orchestration and the data connection are also what later determine your bill.
02 WhatsApp Business API requirements
The single most common misconception is that you can automate the WhatsApp Business app. You cannot. To run an agent you need the WhatsApp Business Platform, commonly called the WhatsApp Business API. Before a single message is answered, two sets of things have to be in place.
On the WhatsApp side
- A Meta Business account with the business verified: legal entity, website and matching documents.
- Access to the WhatsApp Business Platform, either directly through Meta's Cloud API or through a Business Solution Provider.
- A dedicated phone number that is not already active on the regular WhatsApp app.
- An approved display name — Meta reviews it and rejects generic or misleading ones.
- A permanent access token and a webhook endpoint, where incoming messages and status updates are delivered.
- Opt-in and a published privacy policy, proving the customer agreed to be contacted on WhatsApp.
On the agent side
- A language model provider, chosen by the tasks and the budget.
- An integration layer that lets the agent read — and where allowed, write — data from your CRM, ERP, help desk or calendar.
- A set of tools and rules: what the agent may do, under which conditions, and what always needs a person.
- A human handoff, so a conversation can move to a person without the customer starting over.
None of this is exotic, but it is a project. Teams usually underestimate the Meta verification and the data connection, not the AI itself.
03 The 24-hour window and templates
This is the rule that surprises most businesses, and the one that shapes both your costs and how you write to customers.
Inside the 24-hour window — the customer wrote first, so the conversation is open. You can reply with free-form messages: normal, conversational text generated by the agent. This is where an AI agent feels natural.
Outside the window — the window closed, or you want to start the conversation. Free-form replies are no longer allowed. You must send a message template that Meta approved in advance, grouped by purpose (utility, marketing, authentication) and billed differently.
The timeline in practice:
- The customer sends the first message → the window opens.
- The agent replies free-form → cheap, natural, unlimited.
- 24 hours pass with no customer message → the window closes.
- You need to follow up → approved template, billed.
A flow that ignores this looks fine in a demo and becomes expensive in production. So design for it:
- Answer quickly, so every reply stays inside the open window.
- Invite the customer to write first: a button, a link, a QR code on a receipt.
- Collect opt-in properly — follow-ups and re-engagement depend on it.
- Budget templates as a real cost line, not an edge case.
04 What it can solve
With the API connected and the window understood, an agent can take over a large share of the repetitive conversation volume.
What it can do
- Answer frequently asked questions: hours, shipping, returns, pricing, availability.
- Pull order and delivery status live from your system.
- Handle booking, rescheduling and cancellations, synced to a calendar.
- Qualify leads — intent, budget and timing — before a human gets involved.
- Take simple orders and reorders from a catalogue.
- Triage support, gathering what a person would otherwise ask for.
- Send reminders and notifications through approved templates.
- Cover multiple languages, 24/7 — often the real reason teams start.
What it should not do
- Anything with legal, financial or health consequences.
- Any action that requires verifying someone's identity.
- Any irreversible change, without a person in the loop.
The strongest setups are explicit about this: the agent resolves the routine and hands off the sensitive with everything it has already gathered.
05 Cost
There is no single price for an AI agent on WhatsApp, because you pay several suppliers for several different things.
Component | How it is charged | Scales with |
WhatsApp messages | Per conversation / per template | Message volume |
Platform or BSP | Monthly fee or margin | Volume |
Language model | Per token | Conversations, not seats |
Integrations and data access | One-time build | Number of systems |
Human oversight | Ongoing staff time | Volume and complexity |
The two lines that grow fastest are message templates and model tokens. Integrations are usually the largest one-time cost. And because Meta's pricing model changes often, always confirm the current rate card before committing to a figure.
The honest summary: a small pilot can be run cheaply, but a production agent that answers in volume is a running cost with several components.
06
Frequently asked questions
Get the WhatsApp Business Platform in place first: a verified Meta Business account, a dedicated number, an approved display name, a token and a webhook. Then build the agent that consumes those messages: a model, a connection to your data, a set of rules and tools, and a human handoff. Plan the Meta verification and the integrations first — they take longer than the AI.
It depends on volume. You pay Meta for messages (templates always, service replies depending on the current model), your provider or BSP, the language model by tokens, the one-time cost of integrating your systems, and the ongoing human oversight. Message templates and model usage are the components that scale fastest with traffic.
Only at trivial scale. The Cloud API has a limited free tier and some service conversations may not be billed, but verification, integrations, model usage and oversight all cost something. Treat "free" as a pilot, not as a production plan.
For automation you need the API. The WhatsApp Business app is designed for a person typing, not for a system. The API is what allows your agent to receive messages, answer them and connect to your data.
See what an agent can do inside the tools you already use.
Start from your current stack, without replacing it. Find out what your setup already supports today and what it takes to add the rest.
KEEP REEDING
Where agentic AI is already running
The AI-powered ERP
How agents live natively inside the platform instead of being bolted on as add-ons.
Agents inside your Odoo
Already running Odoo? Add agents to the setup you have today, without migrating.
AI agents for CRM and sales
How an agent qualifies, follows up and updates the pipeline on its own.
Empiece a escribir aquí...
AI Bot on WhatsApp: How to Set It Up and What It Can Do for Your Business