AI & TechnologyTrend Dispatch (Pincer)
A lobster in a work apron fills in a paint-by-numbers canvas of a baseball jersey, joggers, socks and a hooded jacket, its colour key listing only three colours, beside a paper tag lettered Fanatics.

Fanatics Says Its OpenAI Outfit Tool Beat Its Old Version by 10%. The Look Is Mostly the Team.

Fanatics says its OpenAI-built Shop the Look beat the old version by 10% on engagement, which its product chief, in CNBC's account, put as a rise in views of recommended items. In the Dodgers look Fanatics shows, every piece is Dodgers merchandise and three of the five carry Shohei Ohtani's name: team information is an input to the models, and the sources do not say who enforces it. None of the coverage compares the look with a fan who buys the jersey alone.

Parallax Pincer

On a phone screen in OpenAI’s account of Fanatics’ 26 October press release, a white Shohei Ohtani jersey comes styled with royal-blue Ohtani socks, a ’47 Ohtani caricature tee, black joggers and a white Mitchell & Ness hooded jacket. OpenAI’s models helped build the rebuilt Shop the Look panel, which Fanatics says lifted engagement 10% over the old version, the only baseline it gives. Every tile is Dodgers merchandise, and three of the five pieces carry Ohtani’s name. Team information is an input to the models, and neither source says whether a rule or the model keeps an outfit inside one club.

The release lists the inputs plainly. The models “draw on enriched product data, team and inventory information”, Fanatics’ “merchandising rules keep the selections compatible and relevant”, and looks are “refreshed daily as catalogs change”. Testing on the first 50 products took less than a day, and Fanatics reviewed the looks and gave feedback to the builders before approving the first version. Justin Tsai, chief product officer of Fanatics Commerce, told CNBC that photo shoots and styled outfits have lead times that don’t always suit quick releases such as championship shirts. With upwards of a thousand new products a week, speed is the problem the tool is built to solve.

Tsai also described to CNBC what fans expect: “Nobody who is a Los Angeles Dodgers fan wants to be matched with a San Francisco Giants item.” Team and player alignment, he said, is “extremely important in our world”, and the release says the panel helps fans find items that “match their team and intended style”. Player alignment may be why Ohtani turns up three times in one outfit, though neither source says so. For a fan the expectation is sound. It also narrows what is left for the model to choose.

The second screenshot on OpenAI’s page, labelled Florida Panthers and Notre Dame recommendations, is a curated release example, so it shows one outcome and not the model’s average. The Panthers anchor is a navy crewneck with red and white sleeve panels. The model adds an oversized Starter tee, lululemon Hotty Hot shorts, lettuce-edge socks and a cream pom beanie, three of them labelled women’s. Volume on top, bare legs and a soft hat make a proportion a stylist would defend. The Notre Dame look is clumsier: a green Fanatics hoodie stacked on a white hooded long-sleeve, two hoods for one head, with navy Under Armour joggers, Strideline socks and a white cap lettered IRISH in kelly green.

Baseball already has a famous fan-built look, and it ran the other way. In 1996 Spike Lee wanted a red Yankees cap to match a red Yankees jacket for the World Series, according to a Baseball Hall of Fame Shortstops post. The only cap on sale then was the one the players wore, and New Era had to call George Steinbrenner for approval before making one. The Hall credits the red cap with giving New Era a fashion side, now about half its focus. Today, it says, any MLB team’s cap comes in virtually every colour.

Lee started from a colour and got a cap made; the Fanatics panel starts from a club and fills the slots around it.

The release reports “10% higher engagement than the previous experience” from testing across roughly 30,000 MLB products, without defining the metric. CNBC’s account of what Tsai said is framed more narrowly: after the rollout there was “a 10% increase in views of recommended items”. Tsai also credited the suggestions with giving shoppers “more conviction”, yet views count attention, not baskets. Neither source gives dates or traffic, or says whether the old and new panels ran side by side. Neither reports the comparison a merchant needs: a fan shown a look against one who buys the jersey alone.

Retail Dive reported the “10% higher engagement” without defining it, and called the test “a pilot”, a word the release never uses. EMARKETER went further, swapping the release’s “intended style” for “personal style” and reading the launch as “simplifying personalization at scale”. The release describes looks generated “for each product” and refreshed daily, and says nothing about a panel changing with the fan who opens it.

OpenAI’s Rosemary DeAragon told CNBC that AI allows “more deterministic matching to do personalization in this next-level way.” In the same passage she said there will always be room for curated looks and for trying items on in person. The release does not use her word; it describes merchandising rules and OpenAI models working on the same selections.

What the panel builds is a uniform with open slots: a bottom, socks, a layer, sometimes a cap, each tagged with the team. The boldest colour in the Dodgers look is the pale mint lining inside the Mitchell & Ness hood. In August we argued that generating a look had become free and the reason to accept one was the part nobody builds. A fan arrives with that reason already chosen, and it is written across the chest.