What Fits actually does
Foot Locker's new AI closet builder, called Fits, asks customers to upload a photo of an item they already own, either from their camera roll or taken live in the app. The tool then suggests complete outfits built around that item and renders them on a virtual model so the customer can see the full look rather than a flat product grid. From that model view, customers can add any of the suggested pieces directly to their cart, which means the entire discovery-to-purchase loop happens inside a single screen.
Michael Cera, Foot Locker's interim head of product management and design, described the internal challenge as resisting the urge to build an AI feature for its own sake. In a LinkedIn post announcing the launch, he said the team engaged with customers and really listened before settling on starting with something a customer already owns rather than something the app is trying to sell them. He credited the build as a collaborative effort spanning Experience Design, Product Management, Ecommerce, Brand, and Stylitics, the styling technology vendor Foot Locker partnered with to build it.
Why starting from ownership changes the trust equation
Most generative AI shopping tools ask a customer to trust a recommendation engine's judgment about what they should buy next, a cold-start problem that requires the AI to infer taste, fit, and intent from limited signal. Fits sidesteps a meaningful piece of that problem by anchoring every suggestion to an item the customer has already chosen and presumably likes. The AI's job narrows from predicting taste to completing a look, which is a more constrained and more checkable task, and one where a bad suggestion is more obviously a styling miss than a fundamental misread of the customer.
That narrower framing also makes failure cheaper and more legible. If Fits suggests a pairing that does not work, the customer can see exactly why, because they are looking at a specific combination against a specific base item rather than an opaque ranked list of unrelated recommendations. That is a meaningfully different failure mode than a recommendation engine quietly serving irrelevant products, and it is likely to generate more forgiving customer reactions even when the AI gets something wrong, because the miss is visible and specific rather than vague.
The cart-from-model-view detail is the real commerce decision
The detail worth lingering on is not the generative styling itself but the decision to let customers add items directly from the virtual model view. That collapses what would otherwise be a multi-step journey, discover a look, find each individual product, check size and availability, add to cart separately, into a single action taken from the point of maximum visual persuasion. Any retailer building a similar feature should treat that cart integration as the actual commerce innovation here, with the AI styling layer functioning as the traffic driver into it.
It also means Foot Locker's product, ecommerce, and brand teams had to agree on inventory and sizing logic sitting directly behind a generative AI output, since a suggested item that is out of stock or mis-sized at the moment of highest customer intent is a worse experience than no suggestion at all. That cross-functional alignment, which Cera's post explicitly credits, is the unglamorous infrastructure work that determines whether a feature like this converts or just entertains.
What this says about Foot Locker's broader positioning
Foot Locker has spent recent years rebuilding its digital and in-store experience under real competitive pressure from Nike's direct-to-consumer push and from resale and streetwear platforms that have eaten into its traditional role as a sneaker destination. A styling feature built around items customers already own is a product bet on community and self-expression rather than pure transaction volume, which fits a brand trying to re-establish cultural relevance with a younger shopper who already has opinions about their existing wardrobe.
It is also a lower-risk AI investment than a full virtual try-on or AI stylist chatbot would have been, both technically and in terms of brand risk. Generative outfit suggestions anchored to a real photo are easier to constrain and quality-check than open-ended styling advice, and a visible miss is a styling disagreement rather than a hallucinated product claim. For a retailer managing a turnaround, that is a sensible place to spend an AI budget: visible, differentiated, and contained if it underperforms.
The lesson for other retail product teams
The broader lesson for retail CTOs evaluating generative AI feature requests is that narrowing the AI's job to a specific, checkable task tends to produce a more defensible product than giving it an open-ended mandate to recommend or advise. Fits works because it is not trying to be a general shopping assistant; it is trying to do one thing, complete an outfit around a known anchor item, and do it well enough that customers trust the suggestion enough to tap through to checkout.
That is a useful filter to apply to any AI feature proposal currently sitting in a retail product backlog: does the AI have a narrow, visually checkable job, or is it being asked to make an open-ended judgment call about customer taste or intent that is much harder to validate before it ships. Foot Locker's team explicitly chose the former after listening to customers, and the resulting product is a reasonable template for what a defensible, cross-functional AI shipping process looks like inside a large retailer.
