The most persuasive sentence your AI stylist can produce is: you already own something that works, skip this one.
Nike, Zalando, and Daydream are all building toward AI styling. None of them start from the customer's existing closet. They start from the catalog, which means every answer is a purchase, which means every answer is discounted by the customer before they read it. Starting from what someone owns is the open lane.
The build is less exotic than it sounds. A customer uploads photos of what they own, or you seed the closet from their order history, which you already hold. A vision model tags each piece by category, color, formality, and season. The stylist then finds the single item that completes the most outfits from what is already there, and matches it against live catalog and stock. Sometimes the recommendation is to wear the jacket they bought in March. That answer is the product.
The closet is the moat and the liability at once. You are asking for photographs taken inside someone's home. Consent language, storage, retention limits, and a working delete path all need to exist before the first upload, and privacy counsel should review it rather than your PM. Treat those images as sensitive data, because they are.
The commercial catch is sharper. A stylist that says no will suppress short-term conversion on the sessions it touches. If your merchandising team is measured weekly on revenue per session, this program dies in month 2 regardless of how well it works. Fund it against repeat rate and 12 month value, or do not fund it.
Personalization has always meant a brand knowing what you clicked. This version means knowing what is hanging in your closet. One of those feels like being watched. The other feels like having a friend with taste.
