Can AI Run Your Business on WhatsApp?

Open WhatsApp right now. Somewhere in those unread chats is a customer who's about to stop being a customer, and on the screen they look exactly like the supplier saying good morning.

An AI assistant can read all of it in a second. That's the easy part, and it's the part everyone is selling.

The honest answer is that AI can run parts of your business on WhatsApp today: the sorting, the drafting and the remembering. It can't run the parts that depend on knowing your business, because a chat thread isn't a record of your business. It's a record of things people said.

Here's where that line sits, and the four things that move it.

The new AI assistants are built for the wrong conversation

The agents launching right now are built for the conversation your team has with itself.

In two days in September, Salesforce shipped a productivity agent that lives in Slack and a service agent that answers customer requests. Both were pitched as a digital teammate. Both are genuinely good. Both assume the work is already written down somewhere an assistant can read: tickets, records, documents that somebody filled in first.

Plenty of companies here run Slack internally, and that's fine. But that's the staff channel. The conversation that carries the money is the one with the customer, and in this part of the world that conversation is on WhatsApp, every time.

That's the conversation nothing is being built for. It's also the one with no record behind it. Your shop wrote everything down: it's all in WhatsApp, in order, with timestamps. And almost none of it is anywhere you can use.

What a chat thread can't tell an assistant

Three things, and they're the three that matter most.

Who this person is. You have a saved contact called "Ahmed tires" and 400 lines of conversation. You don't have a customer. There's no record that says this is the same Ahmed who came in twice last year, because the only thing linking those visits is your memory.

What they actually bought. Scroll back far enough and you'll find "yes send it please". Send what? The thread knows what was said. It doesn't know what was agreed, and an assistant reading it can't tell the difference between a quote, a maybe, and a sale.

What you promised. This is the expensive one. Somewhere in there you said you'd call back on Thursday. Thursday came and went. The thread has no idea the job is still open, so neither does anything reading it.

You could work all three out by reading carefully for ten minutes. That's exactly why you don't. And it's why an assistant given the same ten minutes hands back a confident summary that's quietly wrong.

The three jobs worth handing over first

Start with sorting, then drafting, then remembering. In that order, because that's the order of how much damage a mistake does.

1. Sorting. This is the safest and the most useful. Forty unread chats is not forty problems. It's usually two that need you today, ten that need a one-line reply, and the rest that need nothing. Having something read the pile and tell you which two is worth more than any clever feature, and if it gets one wrong you've lost a minute.

2. Drafting. Not writing on your behalf. Writing the reply you were going to write anyway and putting it in front of you. Prices, opening hours, "yes we have it in black", the same eleven answers you've typed nine hundred times. You still press send, so your voice and your judgement stay in the loop.

3. Remembering. The follow-up you'd have forgotten. Not a campaign, just the one message to the one customer who said "next month" five weeks ago. We wrote a whole post on the four moments that win repeat business, and the reason those four are hard isn't that they're complicated. It's that nobody remembers them on a busy Thursday.

What stays with you is deciding and promising. An assistant that quotes a price, commits to a delivery date, or tells a customer their order is ready has made a promise in your name that you now have to keep.

That's not an AI problem. You wouldn't let a new hire do it in week one either.

Why "just put a chatbot on WhatsApp" goes wrong

Because a chatbot answers the easy half of your messages and hands you the hard half at the worst possible moment.

Picture a phone repair shop in Jeddah. A customer asks whether they fix a particular model. The bot handles it. They ask the price. The bot handles it. They ask whether it can be done today because they fly tomorrow morning, and the bot hands the conversation to a human. Which means the customer explains all of it again, to the owner, while annoyed.

The easy questions were never costing you anything. Answering them was three seconds of your day. The customer with the flight was the whole reason you were watching the phone.

There's a version of this with broadcast lists too. Owners build one, send an offer, and get a response rate that makes no sense given how many people are on it. The reason is mechanical: a WhatsApp broadcast only reaches contacts who have your number saved in their phone. Most one-time customers never save it. So the list quietly filters itself down to the regulars, which means you're paying attention to a channel that's mostly talking to people who were coming back anyway.

HubSpot published a good piece this month on where small-business email tools hit their ceiling. The same shape applies here. The tool didn't break. It just stopped being the thing that was limiting you, and nobody noticed.

What you need in place before any of this works

One thing: a customer record that lives outside the conversation.

Not a database. Not a project. Four fields, and you already know all four:

• A phone number attached to a name, not a nickname you'll forget
• What they bought, and when
• What you promised, and whether it's done
• When you last spoke to them

That's it. With those four, an assistant stops guessing. "Who hasn't been back in 90 days" becomes a question with an answer instead of a reading exercise. "Did we ever deliver the Al Olaya order" becomes a field instead of an argument.

Without them, every AI tool you try is reading the same fog you are, just faster.

This is the same wall a spreadsheet hits, for the same reason. AI made everything you wrote down far more useful, and did nothing at all for everything you didn't.

The order this actually happens in

For most shops it goes like this, and it's worth knowing so you don't skip a step and blame the tool:

1. Write down the four fields for new customers only. Not the back catalogue. Starting from today is enough, and trying to reconstruct three years of WhatsApp is how people give up in week two.
2. Let something sort the inbox for a month. Don't automate replies yet. Just find out which two chats a day actually needed you.
3. Turn on drafting for the eleven questions you answer constantly. Read every draft before it sends, for at least a few weeks.
4. Add the follow-up last. By then you have enough of a record for the follow-up to be specific, and a specific follow-up is the only kind worth sending.

Skipping to step four is the common mistake. It's the most exciting one and it's the one that needs the most groundwork.

Where Swarlet comes in

Everything above works without us, and if you go do it in a notebook this post did its job.

What Swarlet does is remove the part where you have to remember to write the four fields down. The customer record has a phone number field whether or not you thought about it. The order knows whether it was delivered. The follow-up knows when you last spoke, because it's the same system that recorded the conversation. WhatsApp stays where your customers are, and the record stops living inside it.

That's the whole idea: not a smarter chat, just a business that's written down well enough for something to help you run it.

We're not open yet. If any of this sounds like your phone, join the waitlist and we'll bring you in early.

And if you're running more than one branch, everything on this page is already twice as true, because there are now two WhatsApp inboxes and you can only look at one of them.

Frequently asked questions

Can AI actually run my business over WhatsApp?

It can run parts of it. An assistant reads every chat, sorts them, drafts replies and reminds you what's open. What it can't do is decide anything that depends on knowing your business, because a chat thread doesn't hold that.

Why are all the new AI assistants built for Slack and email?

Because those are internal channels, and internal channels already have records sitting behind them. Plenty of companies here use Slack with their staff, and that's fine. The customer conversation is on WhatsApp, and that's the one nobody is building for.

What can't an AI assistant learn from reading my WhatsApp?

Three things it needs most: who the person is beyond a saved phone number, what they actually bought and when, and what you promised them that's still open. None of that is reliably in the thread.

What should I hand over to an assistant first?

Sorting, then drafting, then remembering. Let it triage the inbox, write the reply you'd have written anyway, and chase the follow-up you'd have forgotten. Keep deciding and promising for yourself.

Isn't a WhatsApp chatbot the same thing?

No. A chatbot answers the easy half of your messages and hands you the hard half at the worst moment, usually after the customer has already repeated themselves twice and is annoyed.

Why do my WhatsApp broadcasts reach so few people?

A broadcast only lands with contacts who have your number saved in their phone. Most one-time customers never do, so the list quietly shrinks to the people who were coming back anyway.

What do I need in place before any of this is worth doing?

A customer record that lives outside the chat, with four things in it: a phone number attached to a name, what they bought and when, what you promised and whether it's done, and when you last spoke.

Do I need to stop using WhatsApp to fix this?

No, and you shouldn't. WhatsApp is where your customers are and that's worth keeping. The change is that the thread stops being the only place the information lives.