This article on the “Clueless closet” is part of our Vogue Business membership package. To enjoy unlimited access to our weekly Technology Edit, which contains Member-only reporting and analysis and our NFT Tracker, sign up for membership here.
In the canons of pop-culture fashion, the 1995 movie Clueless maintains an indelible hold. In the past year, designers Marc Jacobs, Christian Siriano and digital fashion’s House of Blueberry have all remade lead character Cher’s looks, with her yellow plaid skirt set even immortalised at the Super Bowl and in the metaverse. But, in the nearly 30 years since the film was released, the fictional computer programme that Cher uses to put together her outfits has been much harder to recreate.
“We can put a man on the moon but we haven’t come out with the ‘Clueless closet’,” says Erin Flynn, co-founder and CEO at smart closet app Cladwell, where people can store their wardrobes and plan looks. To truly recreate the scene, the technology needs to ingest a person’s existing style and wardrobe, suggest and modify fully styled outfits, offer a digital preview of the full look on the body and even suggest new potential purchases. As if.
It turns out that training a computer to understand personal style is even more complex than teaching it to identify a potential love match or source a qualified job candidate. Predicting the compatibility of an entire outfit is more complicated because it requires combining so many separate pieces, says Jenny Wang, founder and CEO of yet-to-launch styling service Alta. “So many people have tried to build the Clueless closet over the years, but this is not a solved problem.”
Still, Cladwell and Alta are among the tech startups throwing their hats in the ring, chasing the dream of helping people decide what to wear. And, new advances in artificial intelligence are bringing it closer to reality. Cladwell recently added an ‘Ask Cladwell’ styling service using generative AI tool ChatGPT. Alta, currently in beta-testing mode and set to launch in the coming months, will use generative AI to recommend what people should wear from their closet based on details such as weather, personal style and what’s in the calendar. Outfit planning app Capsule is planning to add an AI stylist that will recommend outfits and let users shuffle recommendations.
Similarly, online shoppers will soon be able to visualise and style highly personalised full looks, thanks to new tools built with the understanding that personalised and styled outfits are more compelling than images of individual pieces. Zelig, a new company from the founders of luxury distributors Madaluxe Group, will enable online shoppers to photorealistically style layered looks on their own images, while switching out various pieces and viewing different ways of styling each item. Meanwhile, Stylitics, the company used by Macy’s, Revolve and Bloomingdale’s to suggest outfits on e-commerce product pages, is adding a virtual closet tool that shows styled outfit and product recommendations based on items the shopper has already purchased.
Research suggests significant consumer appetite for style help from AI, even with the caveat that they might have to ultimately share personal details including style preferences, purchase history and full-body images. Of the online US consumers willing to use an AI tool, the top desired category is apparel, says Molly Burke, senior retail analyst at software intermediary Capterra, part of consultancy Gartner, citing a July survey on the use of AI in retail. Almost half of all respondents do not mind sharing personal info with an AI tool if they get a more personalised experience, including style preferences and images of their faces and bodies.
“There is a little bit of dissonance in our results,” she notes. “Consumers are very concerned about how generative AI will use their data, but from the rest of their insights, they are just shrugging their shoulders, saying, ‘We are concerned — but we will still use this thing.’” Burke says based on the research, she feels it’s not a question of “if”, but of “when” a fully operational Clueless closet tool will exist.
“I love this question. We talk about it all the time. It is one of the great North Stars of technical AI work that pop culture was able to create for us and that people still have in our heads,” says Jeff Cooper, director of data science at personal styling tech company Stitch Fix. Stitch Fix recently partnered with Clueless costume designer Mona May on a style-finder called ‘Main Character Closets’, designed to take advantage of the popularity of nostalgic movies and TV shows (including Clueless) to help women define their own style. “The reason why we see the fashion from Clueless continue to stick around shows how powerful the storytelling is. Clueless was so unique and really innovative for the time that it has become a staple in pop culture,” May says.
Ingesting closet content
While there is no tool that successfully replicates Cher’s mythical programme yet, these new tools provide significant upgrades to key elements of that equation. “We’re doing crawl, walk, run here,” says Stylitics founder and CEO Rohan Deuskar. “A simple idea, but difficult to execute”. One of the most challenging aspects is knowing what people have in their closet; it’s a tall order to require customers to manually upload each piece.
This is one reason that Stylitics, founded in 2011 as a consumer-facing outfit management app, has since pivoted to a business-to-business model of AI-powered digital merchandising, meaning that it shows how individual products might be styled into full looks when people are shopping retailer websites. Now, it’s expanding its capabilities to take advantage of its knowledge of past purchases, with a new option for retailers to show outfit recommendations based on previous purchases and searches.
Revolve is one of the first retailers to sign on. “It is the richest form of personalisation — to have a closet full of items merchandised and recommended just for each customer,” says Revolve co-founder and co-CEO Mike Karanikolas. “As our customer base widens, being able to personalise outfit recommendations in a scalable and relevant way will be increasingly important and valuable.”
The long-term vision, Stylitics’s Deuskar concedes, is to link a shopper’s profile across retailer websites. This could mean that someone shopping Revolve might see outfit suggestions that are styled with a past purchase from Macy’s. For now, the company is focused on getting retailers comfortable with this capability on their own sites first. Already — without the new tool — Stylitics increases average order values by 17 per cent, according to the company. “It was our fantasy where we were a digital closet platform to connect to a thousand catalogues, understand what people are buying across hundreds of millions of people in real time so we can personalise it, and having all of the product information, attribution and images — and we actually have all of that stuff now,” he says.
Stitch Fix, which uses a combination of data science and human stylists to recommend items to customers, has recently added the ability to recommend full outfits to its customers, including styling items that it knows they have already purchased from the company. Cooper calls it “outfitting”, and says that Stitch Fix is actively investing in how to bring that capability more to clients. He also notes that because the definition of “generative AI” is a bit amorphous, Stitch Fix considers automated, guided “outfitting” as generative in that it generates entirely new outfits. “We really think that genAI is a great term to describe something that is bringing something new into the world that is a creative endeavour.”
Cladwell doesn’t inherently have access to past purchases that way an e-commerce platform does, so it has found a novel approach to encouraging people to upload their wardrobes to the app. The smart closet app determined that when people are seeking outfit inspiration based on what they have in their closets, it’s not as important that the items are the exact make and model of an item; rather, it is easier to prefill similar-enough items. “We learned that most people think of a brand when they go to buy, but they don’t when they put it in their closet,” Flynn says, enabling people to upload a majority of their closet in minutes. They can also manually add pictures of niche items and import URLs from recent purchases. “If you ask people to take a picture of every single item in their closet, it is just not going to happen. It’s fundamentally why this type of business has failed over the years.”
With one million app downloads, Cladwell’s database now has 13 million clothing items and the average user has 125 items in their closet. It linked with ChatGPT in July to enable people to give it a prompt — such as “Put together some outfits inspired by Clueless character Cher Horowitz” — that will return suggestions and styled outfits from their closet, including the images. It’s facilitated “thousands” of AI conversations since (meaning multiple messages per user). (People can ask up to five messages or questions for free before needing a paid subscription.)
Some apps are also leaning on social sharing to gain traction. Capsule is a two-year-old closet management app whose Gen Z-centred audience helped it become popular on TikTok; its 37,000 active users can see each other’s wardrobes to get inspiration. A forthcoming AI stylist will also recommend a daily outfit and let users “shuffle” outfits. (It’s worth noting that the oldest Gen Z was born two years after Clueless came out.)
Alta, co-founded by Wang with AI researcher Shreya Shankar, is planning a more high-touch offering. The service will begin with a personal, human concierge visiting a client’s house to assess and catalogue a closet before the AI-powered personal styling takes off. Similar to Stitch Fix, people complete a style quiz then provide ongoing feedback, including sharing what they actually wore, to train the AI.
This is compared to the neural networks of the past, which were slower and required more manual training on style. Now, generative AI used by Alta uses transformer models (the “T” in GPT) that enable multimodal data (meaning images and text) and can learn by itself, combined with traditional machine learning algorithms to set specific rules related to style. It uses social media trends and user preferences to understand style.
The human element
The last stage of Cher’s outfit programme automatically overlays the clothes on her own image. While a number of companies have developed automated virtual try-on technology for still images and videos, the quality is often not photorealistic unless the items are manually digitally fitted. That’s because details like fit, lighting, texture and layering are incredibly difficult for a computer to inherently understand — and that’s just on still images. Zelig aims to solve that.
Launching this month, Zelig developed technology that will enable people to upload a photo (or select from a model library of more than 40 people) to not only see how an individual item will look on them, but to mix and match various items from a retailer’s site and save them to a personal closet. The startup is co-founded by CEO Sandy Sholl (with co-founder Adam Freede); Sholl is also the co-founder and executive chair at Madaluxe group, which distributes luxury fashion. Rather than an automation tool like the one recently introduced by Google Shopping, Zelig works with the retailers to onboard the clothing items through a mobile photo studio; retailers also go through an onboarding process that takes from two weeks to a couple months, depending on their tech stack.
Zelig doesn’t have strict requirements on the uploaded image other than a relatively straightforward pose with hair tied back; it can “recreate” skin colour by matching the colour of the shopper’s face, so the submitted image doesn’t have to be in their underwear. Sholl says that details like this make the tool more likely to succeed in fashion. “Fashion is emotional. You cannot have technology doing random try-ons. It really needs a gold standard attention to fashion,” says Sholl, who recruited executives from Net-a-Porter and discount plugin Honey (acquired by Paypal in 2020) to develop Zelig.
The company has already filed multiple patents. Like Stylitics, it is considering enabling shoppers to carry their profile with them as they shop various retailers. Sholl cites McKinsey research that generative AI could add up to $275 billion to apparel, fashion and luxury operating profits in the next three to five years. In the immediate term, Sholl sees a major opportunity to reduce returns, which in fashion e-commerce can range from an average 35 to up to 60 per cent of sales, she says. Notably, Zelig will also be positioned as a tool for human sales associates meaning that they will be able to show clients how fully styled looks will look on the body, in place of flat-lay product collages.
This idea of a live human touch might seem antithetical to the promise of an artificially intelligent service, but experts say that this can be critical to the overall success — both technically and culturally — to the future of AI in fashion.
Research shows that 60 per cent of online shoppers are more likely to use an AI tool if a human was evaluating the tool’s responses, says Gartner’s Burke. When Stitch Fix was launched, it popularised the concept of “human in the loop”, she says. Although Stitch Fix’s data science is what led to its success, its marketing has consistently reiterated that clients work with a human stylist to curate the final items that are ultimately sent in their Fixes. That’s why Alta begins its client relationships with a “closet concierge”, and Stylitics employs 40 human stylists to provide “quality assurance” and to curate some experiences — although it would need to hire 40,000, Deuskar estimates, if most of the processes weren’t automated. And no, AI won’t replace stylists, Cladwell’s Flynn says. But it can help increase access.
Stitch Fix’s Cooper says that one of the reasons that Stitch Fix partnered with a well-known wardrobe stylist like May is because the company is emphasising people. Clients, it turns out, aren't looking for total automation. So while generative AI isn’t totally clueless on recommending outfits, it often needs a human touch to get it right, Cooper says. “But I promise — we have watched that scene a lot.”
Correction: Stylitics employs 40 stylists, not four, as an earlier version stated. 10 October, 2023.
To receive the Vogue Business newsletter, sign up here.
Comments, questions or feedback? Email us at feedback@voguebusiness.com.
