WhatsApp Chatbots for Customer Service: A Setup Guide for Businesses

In short: A WhatsApp chatbot that is connected to your systems and actually gets work done needs Meta's WhatsApp Business Platform (Cloud API), not the free WhatsApp Business app. Setup consists of verifying your Meta business portfolio, connecting a number, submitting message templates for approval and connecting the assistant to your own knowledge base. Since July 1, 2025, Meta has charged per message: within the 24-hour window that opens after a customer messages you, you can reply freely; outside it, you can only reach them with an approved template.

WhatsApp is one of the channels customers most often use to reach a business. Where's my order? When is my appointment? Is this product in stock? These questions arrive around the clock, and the longer an answer takes, the more likely a customer is to turn to a competitor.

An AI-powered WhatsApp assistant takes on that load: it answers repetitive questions instantly, pulls information from your systems and hands conversations it can't resolve to your team. In this guide we walk through how a WhatsApp chatbot is set up, from choosing the right infrastructure to going live and reading the costs correctly.

WhatsApp Business app or WhatsApp Business Platform (API)?

The first decision is the infrastructure. The two options look similar at first glance but serve very different needs:

For a chatbot to actually get work done, such as checking orders, creating records and handing off to an agent, you need the WhatsApp Business Platform. Today you connect to it through Meta's Cloud API, either directly or through a solution partner.

A WhatsApp chatbot isn't an auto-reply machine; it's an assistant connected to your systems, working on the channel your customers use most.

What can an AI assistant do on WhatsApp?

• Answer frequently asked questions: opening hours, delivery terms, the returns process.
• Share order and shipping status: pulls live information from your e-commerce platform.
• Book appointments: checks your calendar, suggests a slot and creates the booking.
• Recommend products: understands what the customer needs and points them to the right product.
• Capture leads: saves quote requests and contact details to your CRM.
• Hand off to an agent: passes conversations it can't resolve to your team, with full history.

All of this runs on a RAG architecture that relies on your own knowledge base rather than the model's general knowledge. The assistant doesn't make things up; it says when it doesn't know and hands the conversation over.

There is an important boundary here: Meta's WhatsApp Business Platform terms prohibit providers whose primary purpose is offering a general-purpose AI assistant from using the platform for that purpose. Customer service assistants, where AI is incidental and supports your business's own service, are not the target of this ban. In practice this means: design your assistant not as "a ChatGPT you can ask anything", but as a customer service tool limited to your business's topics. That is the right design both for compliance and for trust.

Setting up a WhatsApp chatbot, step by step

1. Define the scenario and the goal. Which requests will the assistant handle, and how will you measure success? Starting with a single, clear scenario delivers results fastest.
2. Prepare and verify your Meta business portfolio. Open your Meta Business account and start business verification. According to Meta's messaging limits documentation, new business portfolios can message 250 unique people per 24 hours outside a customer service window; business verification is one of the ways to raise that limit to 2,000. If you're planning campaigns or bulk notifications, put verification first.
3. Decide on the number. Will the number connected to the API be a new one or your existing business number? If the existing number is in use on the WhatsApp Business app, plan the migration in advance; settling this early avoids problems later.
4. Gather your knowledge base. FAQs, product information, campaigns and delivery terms should be up to date and in one place. The assistant is only as good as this source.
5. Submit message templates for approval. Business-initiated messages (order notifications, appointment reminders, campaigns) are sent using approved templates. According to Meta's template guidelines, each template is categorized as marketing, utility or authentication, is reviewed automatically on creation, and review can take up to 24 hours. Because the category also determines the price, create each template in the correct category.
6. Connect your systems. Integrate your e-commerce platform, CRM and calendar so the assistant can get things done, not just answer questions.
7. Test and design the hand-off. Try it with real customer questions. Decide when, how and to whom the assistant hands a conversation over.
8. Go live and measure. Track the automatic resolution rate, response time, hand-off rate and template quality rating, and add unanswered questions to the knowledge base.

Rules to know before you start

The 24-hour customer service window

When a customer messages or calls you, a 24-hour customer service window opens; if the customer writes again within that time, the timer resets to 24 hours. While the window is open, you can reply with free-form text without a template. Once it closes, you can only reach the customer with an approved template. Your assistant's flows, including promises such as "we'll get back to you tomorrow", need to be designed around this rule.

Pricing: per message

According to Meta's pricing documentation, since July 1, 2025 charges are made per delivered message rather than per conversation. The price depends on the country (market) and the message category; messages customers send to you are not charged.

There is one more exception: if a customer messages you from a Click to WhatsApp ad on Meta's platforms using the mobile app and you reply within the window, a free entry point window starts that can stay open for up to 7 days, and messages sent in it are not charged. Meta updates its rates several times a year; on October 1, 2026, for example, both the free service message allowance and the in-window utility charge changed. So plan your budget against Meta's current rate card, not a figure in this article.

Consent, İYS and data protection

Under Meta's opt-in rules, you must obtain a person's permission before messaging them. The opt-in must clearly state which business the person agrees to receive messages from and must comply with applicable law; it can be collected through different channels such as a website, SMS, a phone IVR flow or in person. In Turkey, Law No. 6563 states that commercial electronic messages may only be sent with the recipient's prior consent, and records in the İleti Yönetim Sistemi (İYS), the national system where consent for commercial electronic messages is managed, must also be taken into account. If the assistant processes personal data, the privacy notice should be shared in the first message and only the data that's actually needed should be collected. We covered this in detail in our article on responsible AI and data protection compliance.

Pre-launch checklist

• Business portfolio verified, number connected to the API.
• Templates you'll use created in the right category and approved.
• Opt-in collection points (form, checkout page, first message) ready; opt-in records are stored.
• The privacy notice is shared in the first message.
• The assistant's scope is written down: what it answers and what it doesn't.
• Hand-off to an agent tested both during and outside business hours.
• A template-based follow-up flow exists for conversations outside the 24-hour window.
• The monthly messaging budget is calculated against Meta's current rate card.

3 common mistakes

• Giving the assistant unlimited scope: a bot with no clear boundaries loses trust by trying to answer everything, and turning it into a general-purpose assistant also conflicts with the platform terms.
• Forgetting the human hand-off: a bot with no way out drives customers away from the channel entirely.
• Turning WhatsApp into a bulk messaging channel: frequent, unsolicited marketing messages lead to blocks and a lower quality rating for your number, and the quality rating also determines whether your messaging limit can grow.

If you're wondering what drives the budget for a WhatsApp assistant, take a look at our article on AI chatbot costs, and for the broader picture of AI in customer service, see our article on AI-powered customer service.

A well-built WhatsApp assistant turns every minute a customer would have spent waiting into sales and satisfaction. At Kodakod, we build your WhatsApp assistant end to end, from Business Platform setup and knowledge base preparation to CRM and e-commerce integration and the agent hand-off flow. For details, visit our AI integration and chatbot development page.