Physical retail’s AI case rests on a single premise: that a store can solve problems a website cannot. Puma’s new multilingual concierge in Las Vegas and Home Depot’s aisle-level Magic Apron both test that premise — and in doing so, they map the difference between AI that performs intelligence and AI that removes a real obstacle. One is technically impressive; the other is operationally essential.

In late March, Puma activated its first AI Store Concierge at its Las Vegas flagship — a large digital display built on Nvidia’s Nemotron model, Google Cloud infrastructure, and LiveX.AI’s agent platform. The system auto-detects the shopper’s spoken language and switches among English, Spanish, Mandarin, French, and Arabic without prompting. It answers questions about fit, performance, and real-time stock availability across Puma’s running range, handing off complex requests to a human associate. No personal data is retained.

The multilingual capability is the deployment’s strongest feature. Las Vegas sees substantial international tourist traffic; a Mandarin or Arabic speaker who cannot find a staff member who shares their language faces a concrete obstacle. Solving that is friction reduction in the clearest sense. The general product-question functionality sits on weaker ground: the same shopper could search Puma.com or pull up a result in the same time. The concierge’s advantage over those alternatives is physical presence — it meets the shopper where they are, which is the one thing a website cannot replicate.

Rui Pedro Silva, Puma’s VP of DTC Technology, described the launch as “making product exploration faster, more intuitive and more accessible.” That framing fits the multilingual use case precisely. It sits less comfortably on a display answering standard product questions for a shopper who has already decided to visit, already navigated to the running section, and already has a phone in their pocket. The friction Puma is targeting is real — but selective.

Home Depot’s AI deployments, presented at Shoptalk Spring 2026, operate from a different friction calculus. EVP Jordan Broggi argued that “the store has never been more relevant than it is today” — defensible for a company where, as reported by Digital Commerce 360, e-commerce accounts for $25 billion in annual revenue and the online and physical networks are deeply fused. The question AI has to answer is how the physical leg gets faster.

Magic Apron’s in-store capability, being rolled out nationally over coming months, doesn’t answer general questions — it guides shoppers to the right location for a specific item and delivers technical guidance there. The Home Depot website tells a customer whether an item is in stock; it doesn’t help when they’re standing in front of 60 feet of fasteners on a time-sensitive job. That locational precision closes the gap. The AI earns its place precisely where the shopper’s alternative — finding an associate — is slow and uncertain.

The distinction between these deployments comes down to where friction sits relative to the shopping decision. Puma’s Las Vegas visitor has already navigated discovery, shortlisting, and store selection; the heavy friction is behind them. Home Depot’s pro customer arrives with a project agenda and time pressure; the friction is immediate, continuous, and high-stakes throughout the visit.

The counter-argument for Puma is brand surface area. A tourist who receives fluent Arabic guidance about NITRO foam cushioning has an experience they probably won’t replicate at a competitor on the same block, and that impression has value independent of whether the concierge closes the sale. Physical retail has always sold experience alongside product. What Puma hasn’t yet answered publicly — the company says it will assess shopper feedback before deciding on expansion — is whether a display-based concierge is the most efficient way to create that experience across its full shopper mix.

That feedback decision will be the revealing one. If the multilingual feature shows measurable conversion among non-English-speaking visitors — a cohort consistently underserved in U.S. brick-and-mortar — the deployment earns its place on friction-reduction terms. If engagement is high and conversion is flat, Puma has built an impressive demonstration. Demonstrations don’t compound.

The retailers who extract the most from in-store AI will build from a different logic: locating the specific obstacles that physical environments create, then targeting them precisely. Broggi’s Shoptalk framing — that the store has never been more relevant — only holds if the store is solving a problem the alternative cannot. A concierge that speaks five languages is powerful where the language gap is the actual obstacle. Where it isn’t, it’s a screen.