Automating WhatsApp can mean very different things. For a small shop, it is the message that tells people your opening hours. For someone selling through campaigns, it is sending an alert to people who signed up. For an agency with dozens of clients, it is making sure nobody is left waiting. Each goal needs a different tool, and picking the wrong one costs money or the number itself.
This guide sorts the options into four levels, from what comes free in the app to an AI agent that writes the replies. For each one you get what it solves, what it costs and the main thing to watch.
Before you choose, answer three questions
- Who starts the conversation? Replying to people who wrote to you is very different from sending the first message to a list. The second one requires the official API and templates approved by Meta.
- How many people answer? One person on a phone handles the app fine. Three agents on the same number already need a tool with a queue and assignment.
- How much of the answer is the same every time? If 80% of questions are "what are your hours" and "how much is it", a menu does the job. If every client asks something different about their own project, a menu does not help, and you need someone, or something, that reads the history.
With those answers, the levels below are easy to pick from.
Level 1: the free WhatsApp Business features
The free WhatsApp Business app already includes three automations under Business tools.
- Greeting message. Sent automatically when someone writes to you for the first time, or after a stretch with no conversation.
- Away message. Sent when you are outside the hours you set, or always, if you prefer.
- Quick replies. Shortcuts you trigger by typing "/" that fill in saved text, like your payment details.
That covers a small business well. The limit shows up fast: the message is the same for everyone, it does not use the client's name and it does not understand what they asked. Meta documents each feature in the WhatsApp Business help center.
Cost: zero. Risk: none, it is the official app.
Level 2: a menu chatbot
A menu chatbot is the options tree ("press 1 for billing, 2 for support"). Tools such as Typebot, Botpress and ManyChat build that flow with visual blocks, and some have free plans. The WhatsApp AI agent platforms comparison covers the paid support platforms in this space.
It works well when the conversation has a few predictable paths, like booking a slot, sending a copy of an invoice or qualifying a lead before handing them to sales. It works badly when the client writes a paragraph that fits none of the options, and in B2B that is almost always.
Cost: from zero to a few hundred dollars a month. Risk: it depends on how the bot connects to WhatsApp. Through the official API, none. Through an unofficial connection, the same as level 4.
Level 3: the official WhatsApp API
The WhatsApp Business Platform is Meta's API. You need it to send the first message at scale, to have several agents on one number safely and to integrate WhatsApp into your own system.
Since July 2025, Meta charges per delivered template message, in categories such as marketing, utility and authentication. Free-form replies inside the customer service window, which opens when the client writes to you, are not charged, according to Meta's pricing documentation. You can sign up directly or through a provider (Twilio, Gupshup, 360dialog, Infobip and others), who adds their own fee on top.
Cost: per message, plus the provider's subscription. Risk: low, as long as you follow the opt-in and quality policies. The best WhatsApp APIs roundup and the Twilio WhatsApp alternatives page compare the providers.
Level 4: an unofficial connection and an AI agent
There is a fourth route that developers use a lot: connecting to WhatsApp as one more device, the same thing WhatsApp Web does. Libraries such as Baileys and whatsmeow, and servers such as Evolution API, WAHA and Z-API, work this way. The upside is that you keep your usual number, with groups, with no template approval and no per-message fee.
The price is risk. WhatsApp's terms do not authorize this kind of client, and bulk sending through it is the shortest path to a ban. The guide to automating without getting banned covers the rules, and the open-source WhatsApp APIs page compares the projects.
This is the level where AI agents come in, the ones that read the history and write the reply. They handle the support that does not fit a menu: the client who asks about the project deadline, sends an error screenshot and a three-minute voice note in the same minute.
Cost: hosting and the AI model's API. Risk: the number's, which you control by keeping volume low and only replying to people who wrote to you.
Set up your automation in five steps
- List the ten most common questions from the last thirty days. Open the chats and copy the questions, not the answers.
- Split the repetitive from the specific. Hours, address and payment methods go into an automatic message or a quick reply. Deadlines, quotes and technical problems stay with people, or with an agent that reads the context.
- Set the order of service. WhatsApp sorts by the latest message, which hides whoever has waited the longest. A queue by wait time fixes that, and managing client WhatsApp groups shows how to build one.
- Decide what never goes out without review. Prices, deadlines, discounts and any promise. An automation that makes promises in your name creates problems no tool can undo.
- Measure one thing at first: time to first response. If it drops and complaints do not go up, the automation is helping.
The mistakes that show up most
- A greeting that is too long. Nobody reads four paragraphs before asking their question.
- A menu for B2B clients. Someone paying a monthly fee wants to talk to a person, not press 3.
- Sending to a bought list. Besides the ban, it burns the number's reputation.
- AI without review. Generic text with a customer-service-manual sign-off gives the bot away in the first line and cools the relationship. The fixes are in WhatsApp AI automation that sounds human.
Where WACrew fits
WACrew is an open-source level 4 project, built for people who serve clients in groups. It merges groups and chats into a queue by wait time, transcribes voice notes, writes replies in your own style and runs every text through a review before sending. You choose per conversation whether it only suggests or answers on its own. It connects as one more device, not through the official API, so it is meant for answering people who already talk to you, never for bulk sends. The code is on GitHub.



