From Formula to Feed: Inside L’Oréal’s Beauty-Tech AI — and the Three-Engine Playbook Every Operator Can Steal

A few weeks ago I wrote about LVMH’s MAIA — one shared AI brain wired across seventy-five luxury Maisons. It’s a beautiful piece of architecture. But it’s only half the luxury-AI story. Walk a few blocks across Paris and you find the other half, built on a completely different logic. Where LVMH centralised, L’Oréal multiplied.

L’Oréal — the world’s largest beauty company, with more than 90,000 employees and thirty-seven brands from Lancôme to Maybelline — didn’t build one assistant and roll it out. It built three separate AI engines, each aimed at a different audience: one for staff, one for content, one for customers. And the way those three engines stack is, I’d argue, more copyable for the average operator than anything LVMH did.

The Case Study: Three Engines, Not One Brain

If MAIA is a supertanker, L’Oréal runs a small fleet — three purpose-built vessels that each do one job extremely well. Here’s the architecture worth studying:

• Engine 1 — the staff copilot. L’Oréal built an internal assistant (its “L’Oréal GPT”, running on Azure OpenAI) for everyday knowledge work. Just as important: in 2024 alone, 42,000 staff went through a “GenAI for All” course covering how the tools work, what they can do and where they’re risky. They didn’t just ship a tool — they shipped the literacy to use it.
• Engine 2 — the content factory. CreAItech, L’Oréal’s in-house generative-AI “Beauty Content Lab,” lets thousands of marketers spin up on-brand images, 3D product renders and video. Internal tooling now produces up to 50,000 images and 500+ videos a month. That’s an industrial creative line, not a novelty.
• Engine 3 — the customer agent. Beauty Genius, an agentic AI beauty advisor living on WhatsApp and the web, has handled over 1.1 million conversations in the US alone — routine, but high-volume: diagnosing skin, recommending products, telling people where to buy.

Notice the shape. L’Oréal didn’t ask “what’s the one AI platform?” It asked “who are we serving, and what does each of them actually need?” Three audiences, three engines, three very different success metrics.

One Brain vs Three Engines — Which Is Right?

It’s tempting to declare a winner. Don’t. LVMH’s single-platform approach and L’Oréal’s three-engine approach are answers to different questions — and the gap between them is the most useful thing in this whole comparison.

• One brain wins on coherence. A shared platform means one governance model, one set of guardrails, one place to improve. If your biggest risk is fragmentation — dozens of teams buying dozens of tools — centralise like LVMH.
• Three engines win on fit. A staff copilot, a content factory and a customer agent have almost nothing in common technically. Forcing them into one platform can slow all three. If your audiences are genuinely different, specialise like L’Oréal.

The honest answer for most organisations: you’ll end up somewhere in between — a shared foundation (identity, security, model access) underneath a few specialised engines on top. That’s the bit nobody puts in a keynote, and it’s exactly the bit that matters.

Why Beauty Is the Other Canary

In the MAIA piece I argued fashion-tech is the enterprise canary. Beauty is the louder one, and here’s why it moves faster than your industry will:

• The feedback loop is brutal and immediate. A skincare recommendation either works on your face or it doesn’t. That forces real evaluation — not vibes — into the product, which is precisely the discipline regulated industries keep skipping.
• Content is the bottleneck, and AI hits it directly. A global beauty brand needs thousands of localised assets a month. 50,000 AI images isn’t about replacing creatives — it’s about removing the production drudgery so the humans can do the taste-making.
• The customer already lives in the channel. Beauty Genius works because it’s on WhatsApp, where people already are — not behind a login on a website nobody opens. The lesson: meet your audience in their existing harbour, don’t make them sail to yours.
• Literacy was treated as infrastructure. 42,000 trained staff is the quietest, most important number here. Tools don’t create adoption. Trained people do. (We say this so often it’s nearly a tattoo — the tool is the easy part; adoption is the work.)

What to Take to Your Own Ship

You don’t have 90,000 employees or a CreAItech lab. Good — you don’t need them. Three questions to ask your team this quarter:

• Who are our three audiences? Staff, content, customers is L’Oréal’s split — yours may differ. But naming the distinct audiences first stops you from building one mushy tool that serves none of them well.
• Where’s our 42,000? Not “did we buy licences” — did we teach people to actually use them? If you’ve deployed a tool without deploying the literacy, you’ve bought a gym membership and never gone.
• What’s our content factory? Every business has a high-volume creative chore — proposals, social posts, reports, decks. That’s your CreAItech in miniature. Industrialise one of them and you’ll feel the lift within a week.

The Takeaway

LVMH and L’Oréal landed on opposite architectures and both are winning — which tells you there is no single right answer, only a right answer for your shape. One centralised because coherence was the risk. The other specialised because fit was the prize. Both, crucially, invested in the boring foundations first: governance, content pipelines, and people who actually know how to use the tools.

That’s the part we obsess over at ANCHR AI Labs. We help non-technical leaders — founders, operators, brand and marketing teams — figure out whether they need one brain or three engines, and then build the literacy to make either one stick. You don’t need an engineering org. You need a clear map and a trained crew.

AI for non-techies ⚓