AI Agent on WhatsApp for Small Businesses: Getting Started

An AI agent on WhatsApp answers frequent questions, helps schedule and checks orders for you, and hands off to a person whatever it cannot solve. To start well: pick a single task, respect Meta's rules, leave an exit to a human and measure results every week.

This guide tells you what to automate first, which WhatsApp rules you need to know and how to avoid the most common mistakes.

What is an AI agent on WhatsApp and how is it different from a menu bot?

A menu bot gives you fixed options ("press 1 for opening hours"). An AI agent understands messages written in your customers' own words and answers based on the information you gave it: your prices, hours, policies and order status. That is why it can solve more, but it can also make mistakes if its limits are not well defined.

What should you automate first?

Start with repetitive tasks that have a clear answer:

• Frequent questions: hours, location, payment methods, delivery times.
• Bookings or appointments: collect the details and confirm availability.
• Order tracking: check the status and notify of changes.

And leave to a person the complaints, negotiations, money decisions and anything involving sensitive data. This is our recommendation.

Which WhatsApp rules must you respect?

These are the official Meta rules that most affect an AI agent:

1. 24-hour window. When a person messages you, a "customer service window" opens. Inside it you can reply with free-form messages, with no prior approval. After 24 hours, you can only send pre-approved message templates (Meta).
2. Prior consent. You can only message people who agreed to receive messages from you (same Meta documentation).
3. A way out to a human. WhatsApp Business's messaging policy allows automation but requires clear, direct escalation paths; one option is a transfer to a chat with a person (messaging policy, updated September 23, 2026).
4. General-purpose AI. According to ENTER.CO, Meta announced that from January 15, 2026 AI providers such as OpenAI or Perplexity cannot use the WhatsApp Business API when their AI is the main service. The restriction does not apply to companies that use AI for operations or customer service (ENTER.CO). Even so, check the current terms before hiring anyone.

And what does Colombian law say?

If the agent handles your customers' data, Law 1581 of 2012 applies: prior authorization, informed purpose and security measures. The SIC issued guidelines for AI in 2024. We summarize them in this guide to AI and customer data. Keep the agent away from sensitive data, such as health or biometrics, without legal advice.

How do you implement it in 6 steps?

1. Pick one task. Just one, with clear answers.
2. Write the approved answers. Prices, policies, hours: what the agent may state. Anything not written should go to a person.
3. Define escalation. When and how it hands off to a human, and who answers.
4. Ask for consent and disclose. Tell people the chat uses an automated assistant and what the data is used for.
5. Test with real cases before opening it to everyone: odd questions, typos, angry customers.
6. Measure every week. How many conversations it solves without human help, how long it takes to reply and which messages were badly handled. Use that to adjust the answers.

How do we do it at Mokev?

Our site has an AI assistant, Dimitri, and the provider keys live only on the server. And in digital cards with a catalog, the order button opens WhatsApp with the product message already written: a simple step before thinking about an agent.

If you want to evaluate an agent for your business, see our work in consulting and custom software, or write to us.