WhatsApp AI Agents: When They Make Sense and How to Start

A WhatsApp AI agent makes sense when you get a lot of similar questions, your team can't answer them fast enough, and the answers already exist somewhere in writing: a catalog, a price list, your policies. It doesn't make sense when every conversation is a different negotiation, or when the knowledge lives only in one person's head. Here's how to tell which side you're on, what the agent has to be able to do, and how to start with a small pilot.

When it makes sense

• The questions repeat. Hours, availability, requirements, order status, directions. If your team types the same answer twenty times a day, that's a candidate.
• Messages arrive after hours. What comes in at 11 p.m. gets answered the next morning, and by then the customer has written to someone else.
• The information is already written down. A catalog, list prices, FAQs, an order system or a booking calendar. The agent answers from that; it doesn't make things up.
• There's a clear next step. Book an appointment, take an order, send a quote, or hand the customer to a salesperson with the full context.

When it doesn't (yet)

• Every sale is custom. If each customer needs a different proposal, the agent can collect and organize the request, but a person closes the deal.
• The knowledge isn't written down. Before the agent comes a document with the good answers your team already gives. That step alone clears up a lot.
• Volume is low. If you get ten messages a day and someone answers them well and quickly, an agent adds moving parts without solving anything.
• Mistakes are expensive. In health, legal or money matters, an agent can guide and book; it shouldn't decide.

What the agent must be able to do

Hand off to a person

This isn't a nice-to-have; it's a WhatsApp rule. The WhatsApp Business Policy allows automated replies but requires "prompt, clear, and direct escalation paths" to people. In practice: the customer can ask for a human at any point, the agent recognizes when it's no longer helping, and whoever picks up the chat sees the conversation so the customer doesn't have to start over.

Not make things up

An agent that doesn't know something should say so and hand off, not improvise a price. The way to get there is to have it answer only from your documents and data (an approach called RAG) and make every answer traceable to its source. That's the idea behind Tlamatini, a tax and legal advisor in our portfolio: every answer has to carry the Mexican statute or case law it comes from and pass an anti-hallucination check before it's shown.

Connect to your data

If the agent can't see your inventory, your calendar or your order status, all it can do is repeat the FAQ. Connecting it to your systems, through an API or an n8n workflow, is what lets it say "yes, it's in stock, want me to hold one?" instead of "let me check."

Keep a record

Every conversation, what it answered and when it handed off. Without that, there's no way to know whether it works or to fix it.

The WhatsApp rules worth knowing first

An agent connects through Meta's WhatsApp Business Platform (the API, not the WhatsApp Business phone app). These are the rules that shape the design most:

• The 24-hour window. When a customer messages you, a 24-hour customer service window opens and you can reply freely. Once it closes, you can only send pre-approved templates.
• Approved templates. Messages you start (reminders, updates, promotions) use templates that must be approved, and each one is categorized as marketing, utility or authentication.
• Opt-in. You can only message people who gave you their number and opted in to hear from your business, and you have to honor anyone who asks you to stop. Blasting numbers that never asked is the fastest way to lose the line.
• How Meta charges. Since July 1, 2025, Meta charges per delivered template message; non-template messages inside the 24-hour window are free. An agent that only answers inbound messages spends little on messaging; the real cost is building it, the AI model and upkeep.
• Your business, not a general chatbot. Meta's terms bar AI providers from offering general-purpose assistants as their main product on the platform. Meta has said this doesn't affect businesses using AI to serve their customers. In short: your agent talks about your business, not about everything.

And if the agent collects names, phone numbers or order details, make sure your privacy notice covers that use under whichever privacy law applies to you. If you serve customers in Mexico, that includes Mexico's federal data protection law for private parties, rewritten in March 2025.

How to start without risking much: a small pilot

1. One topic. Availability and hours, say, or order status. Not "answer everything" on day one.
2. The answers that already work. Collect real conversations where your team answered well. They're the source material and the yardstick.
3. The handoff rule. Which topics the agent never answers, when it escalates, and who receives the chat.
4. Internal testing. Your team messages it for a few days, trying to make it fail.
5. A limited launch. After hours first, or for one type of question only. If something goes wrong, you switch it off in a minute and normal service resumes.
6. A weekly review. What it couldn't answer, what it got wrong, what it escalated. That's how it improves.

Measure a few things: time to first response, conversations that ended in a booking or an order, and how many went to a person. That's enough to decide whether to expand.

One detail that's easy to miss: the WhatsApp Business account, the phone number, the Meta business portfolio and the templates should be in your company's name, not the name of whoever builds the agent.

What to ask whoever sells you one

• Where does the agent get its answers, and can I see the source for each one?
• What does it do when it doesn't know?
• How does it hand off to a person, and what does that person see?
• Whose name are the account, the number and the templates in?
• What do I pay each month beyond the build: templates, AI model, hosting?
• How do you test each change before it reaches my customers?
• Where are conversations stored, and who can access them?
• How do I switch it off if something goes wrong?

If the answers are vague, keep looking.

If you'd like a second opinion

If you have a use case in mind, we're happy to talk for half an hour and tell you honestly whether it calls for an agent or whether you should start elsewhere, like automating a process or fixing your website first. See how we work on AI Agents, or reach us through contact.