How to Automate WhatsApp Customer Replies (Step-by-Step Guide)

WhatsApp is where many customers want to talk to businesses, especially in Nigeria, the UK, much of Europe and Latin America, and among many communities in the US and Canada. It’s also where messages pile up fastest. This guide shows how to automate replies properly, from what you can do today for free to a full AI assistant.

Option 1: The free WhatsApp Business app

If you’re not using these yet, start here. It takes about 30 minutes:

• Greeting message. Sent automatically to new contacts or after 14 days of no contact. Use it to set expectations: “Thanks for messaging [Business]. For prices, see [link]. A team member will reply within [time].”
• Away message. Sent outside your business hours. Include the one link that answers most questions.
• Quick replies. Saved answers you trigger with a shortcut (for example /prices). They cut typing, but a person still has to send each one.
• Labels and catalogue. Labels organise chats (“new lead”, “paid”, “follow up”). The catalogue lets customers browse products without asking.

The limit: none of this understands questions, takes actions or connects to your other tools. Someone still reads every message.

Option 2: An AI assistant on the WhatsApp Business Platform

The WhatsApp Business Platform (often called the WhatsApp Business API, or Cloud API when hosted by Meta) lets software send and receive messages on your business number. That’s what makes real automation possible.

What you need

1. A Meta Business account (Business Manager). Business verification is recommended. It’s required to raise messaging limits and display your business name.
2. A phone number registered on the platform, either directly with Meta’s Cloud API or through a provider (such as Twilio, 360dialog or other Meta business solution providers).
3. Message templates approved by Meta, for any message you send outside the 24-hour service window (reminders, order updates, follow-ups).
4. An assistant connected to the number, with a knowledge base and rules.
5. A shared inbox or handover route so staff can take over conversations.

How the costs work

There are three parts, and they scale with volume:

• Meta’s messaging fees. Meta charges for certain template messages (marketing, utility and authentication categories, priced by country). Replies inside the 24-hour customer service window are free-form, and Meta’s pricing has made many of them free. Check Meta’s current pricing page for your markets, because rates and rules change.
• AI usage. Each AI reply has a small model cost, typically a fraction of a cent to a few cents per conversation depending on the model and length.
• Hosting and provider fees. These depend on whether you use a provider’s platform or a custom build.

We estimate all three for your actual volume before any project starts.

Step by step: building it properly

Step 1: Collect your most common questions

Scroll through your last few hundred conversations, or export them, and group the questions. For most businesses, a handful of topics (prices, availability, location and hours, order status, how to book or pay) make up most of the volume. Write down each topic and how often it appears.

Step 2: Write approved answers

For each topic, write one clear, current answer with the next step. For example:

Delivery times: Orders placed before 2pm ship the same day. Lagos delivery takes 1–2 days. Elsewhere in Nigeria, 3–5 days. Track your order here: [link].

These answers become the assistant’s knowledge base. If you wouldn’t be happy with a staff member sending it, don’t give it to the AI.

Step 3: Decide what the assistant must never do

Write explicit rules, such as: never confirm refunds, never give medical or legal advice, never invent prices or delivery dates, hand over complaints immediately. Clear “never” rules prevent most chatbot embarrassments.

Step 4: Set up the platform and templates

Register your number, verify your business, and submit templates for the messages you’ll send proactively (order confirmations, appointment reminders, payment links). Template approval is usually quick if the message is clearly transactional and not misleading.

Step 5: Connect the assistant and the actions

Connect the assistant to the number and give it the tools it needs, and nothing more. That might be reading order status from your store, checking calendar availability, or creating a lead in your CRM. Each action should be narrow and logged.

Step 6: Design the handover

When the assistant can’t help, or the customer asks for a person, it should:

1. Say it’s passing them to the team, and when to expect a reply.
2. Send the full conversation to a person (shared inbox, email or the app via coexistence).
3. Stop replying in that conversation until the person hands it back.

Step 7: Test with real questions, then launch

Before launch, run 30–50 real past questions through the assistant and check every answer. Fix the knowledge base, not just the prompt. Then launch, and read a sample of conversations every week for the first month.

Common mistakes

• Trying to answer everything. Narrow assistants are more accurate and more trusted.
• Hiding that it’s a bot. Say it’s an AI assistant. Customers forgive a bot that’s honest and helpful.
• No opt-in for proactive messages. You need consent before sending business-initiated messages. Collect it at checkout or booking.
• Set-and-forget. Prices, policies and stock change. An assistant with outdated information is worse than none.

What results to measure

Track first-response time, the share of conversations resolved without a person, handover reasons, bookings or orders started in chat, and customer complaints about the bot. Compare against the month before launch.

If you want to see how a grounded assistant behaves, try ours. It follows the same rules described here. Or take the free AI audit to see whether WhatsApp automation should be your first project.