How to automate WhatsApp in a business: what to automate first and what not to

Most businesses that want to automate WhatsApp start with the tool. They open an account, connect the number, write a welcome message and a list of FAQs, and two weeks later the team is still answering the same things as before, only now with a menu in front.

The problem is not the tool. It is that nobody decided what was going to be automated. This guide is that order, in six steps.

If you already know what you want to automate and you are looking for who does it, the AI agent for WhatsApp page sums up what a Met handles on the channel and which systems it connects to. If you are choosing a platform, the guide on how to choose a platform to automate WhatsApp says what to ask for in a demo.

1. Count one week of conversations before configuring anything

Take a normal week of WhatsApp and note three things for each conversation: what the customer asked for, what had to be checked to answer and who ended up solving it. You do not need a tool; a spreadsheet with those three columns is enough.

At the end of the week you will see two groups. The first is the repeated questions with a clear source: “do you have this size?”, “what times are open on Thursday?”, “has my order shipped?”, “how much do I owe?”. The second is what needed judgement: a complaint, a pricing exception, a question nobody has written down. The first group is what gets automated first. The second stays with your team, and the automation’s job is to hand it over in order.

2. Separate the three kinds of automation

“Automating WhatsApp” mixes three things that are configured differently, and confusing them is the most common reason a project stalls halfway:

Kind	When it starts	Example	What it needs
Answering what comes in	The customer writes	Checking a slot and booking it	An agent with access to the system that holds the answer
Messaging first when something happens	A date or an event	The day-before reminder, an instalment falling due	A Meta-approved template if more than 24 hours have passed
Notifying many at once	You decide	An event day, a change of hours	A list with consent, an approved template and a sending cap

In Meteor each one has its place. The Met assigned to the channel answers what comes in without you building a flow. Automations handle what starts without anyone writing. And notices to a segment go out from Broadcasts, with an approved template, a segment built from tags or contact fields and a messages-per-minute cap that protects the line.

3. Move the line to the official API

The WhatsApp Business phone app has greeting messages, away messages and quick replies. For anything to answer by checking your systems, the line has to be connected to the official WhatsApp API, which is the one Meta opens to platforms. With coexistence it can be the same number you already use in the app, if Meta enables it for yours.

Avoid connectors that read a WhatsApp Web QR code without going through Meta: they are not a supported route: they work until Meta detects them, and then it suspends the number. The step by step (business verification, number registration, templates) is in the guide on how to implement a Met on WhatsApp Business.

4. Connect the source of each answer

An agent that checks nothing can only repeat what you wrote in its instructions, and that goes stale the day a price changes. For each question in the first group, note which system the answer comes from and connect that system:

• The CRM first. Every contact that comes in is recorded with its history, tags, notes and owner. In Meteor the CRM is native and needs nothing external connected.
• Then, the source of the most frequent question. Inventory if you sell products, a calendar if you sell time, receivables if you collect instalments. A complete example is booking appointments on WhatsApp: first the services with their length, then the open slots and finally the booking.
• Payments, if you close sales in the chat. A payment link moves no money on its own: the payment counts as received when the transaction is checked, not when the customer says they paid.

Add one system at a time. If the first one does not work well, the second only adds another source of errors.

5. Decide what it does alone and what goes to a person

Before opening the channel to real customers, write two lists. The first is what the agent resolves on its own: checking, booking, recording the contact, sending a payment link. The second is what it never resolves on its own: a refund, a discount outside policy, a complaint, any clinical or legal question.

That second list is configured, not left to the model’s judgement. In Meteor you decide which tools each Met has, so an action you do not want it to take alone is simply not among its tools. When a case from the second list comes up, the Met leaves an internal note with the summary and the reason, pauses itself and assigns the conversation to a person. The pause matters: without it, the agent and the advisor reply over each other.

6. Measure the first week with the same sheet from step 1

Go back to the sheet from the first step and compare. The three questions that tell you whether the automation works are not how many messages the agent sent:

1. Of the conversations in the first group, how many were resolved without a handover?
2. Of the ones that went to a person, how many arrived with full context? If your team has to ask again, the handover is misconfigured.
3. What did it get wrong? Each mistake is a source that still needs connecting or a rule that still needs writing. Fix it before adding the next process.

Once the first process works, repeat the cycle with the next one on the list. Automating WhatsApp is not a project you finish: it is a list of conversations that keeps getting shorter.