How Magalu's Lu became a WhatsApp salesperson that converts 3x the app
Magalu’s Lu sells on WhatsApp and converts about three times better than the company’s own app. The system behind it is simpler than the number suggests. And more expensive than it looks, for a reason that isn’t in the code.
Lu isn’t a new chatbot. She’s a persona Magazine Luiza has been building since 2003, first as the website’s virtual assistant, later as one of the largest digital influencers in Brazil, with tens of millions of followers across Instagram, TikTok, and YouTube.
What changed was the channel. The company moved that persona into WhatsApp, where the Brazilian customer already is, and let her do in the chat what a store clerk does at the counter: ask what you want, suggest, answer, close.
What the system does for the operation
In practice, Lu on WhatsApp handles three moments of the sale inside one conversation.
Discovery: the customer says what they want in text or audio, no menu to navigate. Recommendation: Lu returns a few options, not the whole catalog. Checkout: payment happens without leaving the thread.
It’s the same buying path as the app, without the app. The customer downloads nothing, creates no account, forgets no password. They open a conversation that was already open.
What the public record says
The numbers come from press coverage, not a technical paper. SBVC, StartSe, and CNN Brasil (2026) report WhatsApp conversion at around three times the app.
Nobody outside has seen the architecture from the inside, and we won’t pretend we have. What you can state safely is the shape: conversational commerce, on top of the WhatsApp Business API, wired into Magalu’s catalog, stock, and payment.
The AI is in there, reading the request and shaping the recommendation. But it’s the smaller piece of the story.
The part the case hides
Here’s where it gets honest. Lu doesn’t convert because of AI. She converts because of brand, and because she’s in the right channel.
Think about what the 3x is comparing. On one side, people opening the app on impulse and browsing. On the other, people messaging a brand they’ve known for twenty years, intent already formed. Those two groups aren’t the same. Part of the jump is selection, not persuasion.
And the trust doesn’t come from the model. It comes from Lu being built over two decades into a face the customer accepts on WhatsApp without flinching. That doesn’t ship with a chatbot license.
There’s a ceiling, too. Chat is linear; comparing eight products side by side is worse in a thread than in a grid. WhatsApp shines on reorders and quick questions, not long research. Copy the format expecting it to handle every kind of purchase and you find the limit fast.
The real cost, and where it sits
Look at the conversational layer alone and it seems cheap. That’s not where the money is.
The expensive part is the integration. The assistant has to read real stock, in real time, or it promises what isn’t there and burns trust on the first missed delivery. It has to talk to the catalog, the payment method, the order system. That wiring is weeks of engineering, not a weekend.
And the most expensive item, the brand, Magalu paid for over twenty years. That one has no invoice. It has a calendar.
The smaller version that fits your chain
You don’t have a Lu. The good news is you don’t need one.
The version at 5 to 10 percent of the scale doesn’t try to build an influencer. It takes the channel your customer already uses, the same WhatsApp, and makes the assistant do one job well: availability, reorder, store pickup. One scope, not the whole catalog.
The rule that separates what works from what embarrasses you is a single one: the assistant only promises what stock confirms. The version that works has fewer parts. It integrates deep into one thing instead of pretending to do everything.
The brand can be your own name, no persona, no costume. The customer doesn’t need to love a character. They need the answer to be right.
Now open your company’s WhatsApp number and send a message as if you were a customer. What happens? A human in forty minutes, a form, or silence? That gap between the message and a useful answer is the project. It’s the one costing you the sale Magalu captures.
Speccing this alone is where most teams stall. The decision crosses operation, stock, payment, and service, and without a written design the project either grows wild or never starts. We’ve written about build vs buy in this kind of decision and about why we work from inside the operation. It’s the kind of thing we deliver.
Tell me the result you want, the specific number: channel conversion, response time, reorders per customer. In one hour I’ll send back a sketch of the architecture sized for your operation, with a build-or-buy marker on each piece: what to buy ready-made, what to build, what to leave for later. If the shape confirms, the two-week Diagnóstico turns the sketch into a signed spec: the paths, the costs, the dependencies, and the acceptance criterion for each component. That document becomes the contract for the Implementação.