Every week a new tool lands in my inbox promising to revolutionise the store with artificial intelligence. Some are useful. Many are a marketing layer over something that already existed. And quite a few are pure hype with a logo. After more than 25 years arranging stores, my rule for AI in visual merchandising is simple: it’s worth it when it helps decide better which product goes where and when, and when that decision can be measured. If it only produces pretty images, it’s decoration. Expensive decoration.
Here are real cases from brands already using it, a table separating what works from what doesn’t, the questions to ask before paying a vendor, and how to start without burning money, whether you have one store or two hundred.
What works: real cases, not promises
Real-time product data: Zara and RFID
Inditex has been tagging its garments with RFID for years: each one has a unique identifier that shows in real time what is in every store and online. In March 2025 it announced a further step, piloted in four Zara stores (Compostela in A Coruña, Gran Vía in Bilbao, Plaza de España in Madrid and Paseo de Gracia in Barcelona): staff read the garments with a handheld device, deactivate the security tag and take payment anywhere in the store, without going to the till.
That isn’t AI in itself. But it’s the foundation for everything else: without reliable stock and sales data per item and per zone, no algorithm decides well. If your store doesn’t know what sells on the entrance table versus at the back, the first step isn’t AI. It’s measuring.
Simulating before moving fixtures: Lowe’s digital twin
Lowe’s, the US home improvement chain with more than 2,000 stores, presented digital twins of two stores (Mill Creek, Washington, and Charlotte, North Carolina) in 2022 with NVIDIA Omniverse and Magic Leap. With them, associates use AR headsets to compare how a shelf should look with how it actually looks, leave notes for central planners about planogram improvements, and see 3D heatmaps of items bought together but placed far apart. Above all, they simulate different layouts with historical order data before moving a single fixture.
For a chain, this saves installs that don’t work. For a small store, the affordable version is a floor plan, before-and-after photos of every change and the sales next to them. The logic is the same: test on a copy, not on the customer.
Computer vision on the shelf: the planogram that polices itself
In April 2026, US convenience chain QuikTrip and SymphonyAI won a Convenience Store News award for planogram excellence. Their system designs planograms by store cluster and uses computer vision to compare shelf photos with the approved planogram: it detects stock-outs, misplaced products and execution gaps. QuikTrip grew from around 700 to more than 1,200 stores while staying consistent. Note: they published no sales percentages, and that is information too.
What matters for visual merchandising isn’t the technology, it’s what it frees up: the team stops checking shelves by hand and spends its time fixing what actually fails. A performance planogram like the one below, with each item coloured by how it sells, is exactly the kind of data an algorithm can read and a good VM knows how to interpret.
Help with repetitive tasks
Writing product copy, translating a visual merchandising guide into three languages, preparing sign variations, summarising weekly sales by zone or suggesting ten concept ideas for a sale window. Here AI saves real hours, as long as someone with judgement checks the result. It’s a very fast intern who doesn’t know your brand: useful if you correct it, dangerous if you leave it alone.
What is hype (for now)
The “AI-designed” window with nobody behind it
Generated images are a starting point, not a buildable project. They have no measurements, materials, weight, real light or your real product. Nobody can fabricate a backdrop from an impossible render. They’re good for opening ideas in a meeting; the window is still designed by someone who knows how to hang a mannequin.
Screens with automatic content nobody watches
A screen isn’t phygital by itself. If it isn’t connected to the product and the customer journey, it’s expensive noise. The two photos below explain it: in the first, the screens are the whole window and become a sculpture; in the second, the screen shows the detail of the product in front of you. Both have an idea behind them. An auto-generated looping video doesn’t.
Promises of “+X% sales” with no comparable case
When a vendor promises a percentage, ask for three things: the exact figure, the period and what it’s compared with. Same store, same month last year, same range? No answer, no case. Notice that even the award-winning cases above talk about availability, consistency and time saved, not magic sales figures.
