TLDR;
Fanatics has rolled out an OpenAI-powered Shop the Look feature across its sports stores, recommending coordinated, in-stock products that match the shopper's team. Fanatics says a test on roughly 30,000 MLB products lifted engagement by 10%, but it has not published conversion or revenue results. For retailers, the lesson is to combine clean catalogue and inventory data with hard merchandising rules, then test commercial outcomes rather than just clicks.
What happened
Fanatics launched the new Shop the Look experience on 26 October across Fanatics.com and many official NFL, NBA, MLB and NHL online stores. A shopper viewing a jersey can see matching hats, socks or accessories on the product page. The system draws on enriched product data, team information and current inventory, while Fanatics' own merchandising rules keep combinations compatible with the selected team and intended style.
OpenAI's case study says the looks refresh daily as the catalogue changes. Codex helped the Fanatics and OpenAI teams build tools that enrich catalogue data, identify compatible items, evaluate recommendations and generate a look for each product. The teams tested the first 50 products in less than a day, iterated for several more days and then expanded the approach across thousands of products.
Fanatics reported that a test covering about 30,000 MLB products achieved 10% higher engagement than its previous Shop the Look experience. CNBC's reporting adds the scale behind the problem: Fanatics manages more than 1.5 million SKUs across hundreds of sites and introduces at least 1,000 products in a typical week. That makes manual styling and photo production too slow for fast-moving launches such as championship merchandise.
The result needs a careful label. Fanatics and OpenAI supplied the engagement figure, and neither published the test duration, assignment method, definition of engagement or effect on conversion, order value, returns or margin. Independent coverage also noted that the companies did not disclose whether the higher recommendation views translated into more sales.
Why it matters
This is a stronger enterprise example than a generic shopping chatbot because it operates inside a live retailer experience and respects operational constraints. Team allegiance is not a soft preference. Recommending a rival team's product is an obvious failure, while showing an out-of-stock accessory wastes the shopper's attention. The system therefore combines model-led matching with deterministic data and rules.
The daily refresh matters as much as the model. A recommendation engine that sees stale stock, missing variants or delayed launches can create attractive but unbuyable combinations. Retailers need reliable feeds for product relationships, availability, price, imagery and eligibility, plus a way to suppress items that legal, licensing or merchandising teams do not want paired.
Quartz reported that Fanatics views the tool as a way to narrow a very large catalogue while keeping human curation relevant. That is the right division of labour. AI can generate and score more candidate combinations than a visual merchandising team can photograph, but people still define taste, brand rules, exclusions and the quality bar.
The claimed 10% engagement gain is a signal, not a business case. More recommendation views could lead to larger baskets, or simply more browsing. Teams should connect exposure to add-to-cart rate, conversion, average order value, gross margin, returns and customer complaints before deciding that the system has created incremental value.
How your brand can benefit / be affected
Start with one category where compatibility rules are clear and inventory changes frequently. Define the intended outcome before launch, such as incremental attachment rate or margin per session. Keep a stable control group, document which products were eligible and measure downstream behaviour rather than using recommendation clicks as the only success metric.
Treat catalogue enrichment as product infrastructure. Add explicit attributes for compatibility, style, audience, team or licence, and make stock and price freshness observable. Give merchandisers hard controls for exclusions, required pairings and seasonal priorities. Log the inputs and rules behind every displayed recommendation so teams can explain bad matches and reproduce incidents.
Build a review loop that samples recommendations by high-value category, new launch and edge case. A team mismatch is easy to spot, but other failures are subtler: incompatible sizes, duplicate products, low-margin bundles or combinations that increase return risk. Human review should target these commercial constraints, not merely judge whether an outfit looks plausible.
Benzinga's coverage correctly separates the 10% engagement claim from sales evidence. Use that discipline in procurement and internal reporting. Ask vendors for test design, latency, catalogue freshness, model fallback behaviour and total operating cost. Fanatics shows that owned-site AI discovery can reach production scale, but each brand still has to prove that better matching produces profitable purchases.
News date: 26 October 2026. Editorial review: 27 October 2026.