Stitch Fix’s radical data-driven way to sell clothes–$1.2 billion last year–is reinventing retail

CEO Katrina Lake’s e-commerce retailer uses data science to find you the best-fitting clothes you’ve ever worn

Stitch Fix’s radical data-driven way to sell clothes–$1.2 billion last year–is reinventing retail
“It’s a totally radical way to sell clothes,” says Stitch Fix CEO Katrina Lake [Photo: Jason Madara; Hair and makeup: Erika Taniguchi at Kern Represents]

While working on a PhD in astrophysics, Chris Moody used supercomputers to simulate how galaxies crash into each other. For his first nonacademic job, he joined Square as a data scientist in 2013. About a year later, he started talking with some data-scientist friends who were employed at a startup called Stitch Fix, an upstart e-commerce service that delivered boxes of women’s fashion, known as “Fixes,” using a mix of algorithmic and human curation.


Moody was mystified. “What on earth are you guys doing at a clothing company?” he recalls asking, admitting that his sartorial taste at the time hewed to “what costs less than ramen?” Their response, though, sent his brain firing. How do you mail customers clothes they’ll love, and that fit them perfectly, without the client ever getting measured or viewing the inventory? Soon he was pushing for a job. “When I was interviewing, I was like, Ooh, this is a place where I’m going to be continuously thinking about this stuff in the shower, going to bed, waking up in the morning.”

He joined in January 2015, and he’s still obsessed. Frustrated that the company only received feedback from customers on the five items mailed in each box, he designed a feature in 2017 called Style Shuffle, which allows customers to rate a set of clothing images each day. A sort of Tinder for clothes, it became available on Stitch Fix’s iOS app in March and has proven to be stickily addictive: It not only trains the company’s algorithm to understand holistically a client’s personal style, but it also draws customers back to the app and interests them in Stitch Fix’s inventory. More than 75% of Stitch Fix’s 2.9 million customers have used it, providing the company with more than a billion ratings. Style Shuffle has vastly improved the company’s ability to personalize its offerings and has boosted Fix requests. “The problems here are extremely interesting,” Moody tells me while wearing a decidedly unfrumpy Nehru-collared shirt.

Stitch Fix is using its data prowess across every aspect of its business to reinvent the $334 billion U.S. apparel industry. For consumers, it’s solving the discovery problem exacerbated by the endless sea of product online, where more than a quarter of clothes are now sold. “Here are all these beautiful things,” says Stitch Fix CEO Katrina Lake, sitting in a fishbowl conference room at her San Francisco headquarters last fall, “but the reality is only a subset of things are right for me.” By soliciting millions of customers’ feedback and precisely measuring every aspect of the clothes it sells, from more than 1,000 brands plus its own in-house labels, Stitch Fix can offer personal styling at scale, widening the market from the very rich to the average consumer, who currently pays the company an average of $55 per item to avoid the headache of shopping.

Stitch Fix is also bringing innovation to suppliers. The brands it works with—including Kate Spade, Karl Lagerfeld Paris, Sam Edelman, John Varvatos, Toms, and Rebecca Minkoff—receive not only a vehicle for mainlining merch directly into the hands of new customers but also a trove of unprecedented marketplace insights, such as whether plus-size women find their pants too long or men over 40 consider a shirt “too Brooklyn.” Stitch Fix offers brands an alternative to declining department stores, fading specialty chains, or Faustian bargains with Amazon, while increasingly becoming the place to make their styles better, no matter where they’re sold.

When Lake took Stitch Fix public in November 2017, at age 34, she was not only the youngest female founder to ever lead an IPO but she also stood at the Nasdaq podium while holding her toddler on her hip. The company has expanded from exclusively selling women’s wear into men’s, kids’, plus size, and basics like underwear, and will launch in the U.K. this year. Stitch Fix has been profitable since late 2014. It generated $1.2 billion in its fiscal 2018 with earnings of $45 million and took in $366 million in its first quarter of 2019. The company also recorded its highest-ever rate of purchased items per Fix among female customers in its most recent earnings, indicating that all that data science and personalization is paying off.

Despite all this steady momentum, Stitch Fix’s stock has been volatile, peaking in mid-September when investors valued the company at $5.1 billion before dropping two-thirds, to its IPO price, in three months, then trending upwards in January. Investors and analysts don’t seem to appreciate what makes the company run. While data has developed a pernicious image in the hands of Google and Facebook, in the context of Stitch Fix, the bigger issue is that “data” has been leeched of any meaning coming from a tech company. “There is so much buzzword overuse in Silicon Valley that I think when a lot of people hear Stitch Fix say ‘data science’ or ‘AI’ or ‘machine learning,’ they think it’s just being said for the sake of it,” says Bill Gurley, a Stitch Fix board member and one of the venture capitalists who bet early on the company. “Some may think we’re just putting stuff in a bag and sending it to you.”


Stitch Fix is not just putting stuff in a bag, as my time among its executives, data scientists, and stylists revealed. But I also have a fairly personal reason to believe in the power of a thoroughly data-driven enterprise: the dark-rinse Liverpool jeans that arrived in my second Fix. I’m a veteran e-commerce shopper for the top half of my body, but when it comes to legs, I have nursed a nearly manlike loyalty to the same Urban Outfitters black skinny pants for the past eight years in order to avoid the emotional turmoil of the dressing room and the crapshoot of the internet. (When a pair inevitably gets holes or turns saggy, I order a replacement online.) So I requested some jeans from my Stitch Fix stylist in my test-drive of the service, signing up and paying like any other customer.

When I opened the Stitch Fix box and pulled the jeans on, I felt that modern amalgam of elation and disquiet when totally nailed by an algorithm, like when Spotify pushes a perfectly pleasing new blues tune into my curated mix. The jeans alone convinced me to overlook the duds in my box—and there were some—and consider that what I thought was my highly personal, hard-earned taste in clothes, honed by more than 30 years of victories and misses, might really be just a complicated math problem of the sort Moody could solve.

Algorithms drive Stitch Fix’s every move. There’s one to anticipate buying and repurchasing needs, letting the company know, for example, that it’s going to need an inventory refresh on sizes 12 and 14 of a particular style of jeans. Another algorithm assigns each Fix to one of five warehouses, and one sets the most efficient path for a warehouse worker to walk through the rows, assembling multiple clients’ boxes at a time. Stitch Fix also demands efficiency of its customers, who have to decide what to buy within three days. The quick turnaround time, combined with its algorithmic buying, lets Stitch Fix turn over its inventory six times a year, instead of a department store’s four.

The merchandise team takes precise, detailed measurements of each article of clothing and tags its texture and aesthetic—similar to how Pandora tags a song’s attributes. A proprietary platform then delivers potential matches to Stitch Fix stylists serving customers. They understand requests too subtle for AI, such as “I need something to wear to my ex-girlfriend’s wedding.” Some of the selections are of the “clients who bought this also bought that” type used by Netflix, but most match the weighted attributes of a person’s taste to the same attributes in clothes.

To make these matches, Stitch Fix asks customers for a lot of data up front. They must fill out a lengthy style profile, the details of which are entered into an algorithm. They’re also urged to create a Pinterest board of fashion likes, which software will scan for shape and style to match to the company’s inventory. Then there’s the feedback users deliver after receiving their first box (e.g., “Ruffles make me look like Big Bird”).

“[We] almost should be paying the customers to do this for us,” says Eric Colson, a former VP of data science and engineering at Netflix, who joined Stitch Fix in 2012 and is its chief algorithms officer. Products like Style Shuffle are designed both to expedite this process and discover new insights. A customer could request a button-down shirt, but with Style Shuffle, Stitch Fix can discern that she tends to like dressy ones with prints over casual denim. Customers may not articulate such nuances themselves, yet they communicate it with two clicks in the shuffle.


How Stitch Fix puts data to work

The 3,900 stylists who translate these signals into clothing choices are mostly part-time employees who work from home, clustered in smaller cities like Sacramento and Dallas, so they can join periodic in-person meetings. As they put together Fixes, they scroll through algorithm-generated clothing suggestions, each with a “match score” (the percentage chance a particular client will buy the item), and glance at what’s been sent in the past to avoid anything too similar.

Stitch Fix has found that the most important attribute isn’t style or price, but fit. “We ‘re better off sending you a dress that’s $68—that you’re going to love, that fits you great—than something that’s discounted because it’s not working,” Lake tells me. “There’s no price for a bad dress.” That is to say, if a stylist recommends a Calvin Klein shift dress that you never would have considered, but it fits uncannily well, you may just shell out $88 to keep it, as I did.

Still, using algorithms as the North Star to identify people’s style and find them the right clothes to match is difficult. The data science is getting better, but it’s far from perfect. “A client who’s 60 years old, who lives outside of Minneapolis,” Lake says, “is going to get a totally different Fix than a client who is 24 years old and lives in Manhattan.” But I didn’t entirely see that play out in my own Fixes, nor have some of my Bay Area friends, who feel they keep receiving iterations of the same cardigans and basic layering shirts—i.e., suburban-mom wear. Stitch Fix is ideal for people who hate shopping, don’t have a great idea of what looks good on them, and for whom clothes from the fashion median are just fine. Which is a lot of women, and arguably even more men. It’s the place to find a leather jacket, a floral maxi dress, a button-down shirt, and jeans. (Unless you’re willing to pay for Stitch Fix’s premium designer labels, with prices in the $100 to $600 range.)

Although Stitch Fix works with more than 1,000 brands, it also designs its own clothing in-house to fill out holes in the marketplace and ensure consistent inventory. The company’s data and merchandise teams combine popular attributes from past items to create pieces that will have predictably high profit margins and buy rates. Stitch Fix does not disclose the percentage of its own labels that make up any given Fix, but in 2017 revealed that it was about 20%. Lake insists that Stitch Fix is not trying to become a vertically integrated fashion house, making all its own clothes. “There are so many great vendors out there,” she says.

House brands let the company quickly iterate on the changes customers are asking for. When men were returning the company’s dress shirts in size XXL, for example, the fit team realized that it needed more material in front to cover their bellies, a larger arm opening, and a smidge more room in the chest. After adjusting those measurements, Stitch Fix ordered up new shirts as a test. The buy rate for XXLs soared by 40%.

Stitch Fix also passes along fit information to its partner brands. From department stores, “you really only get, ‘Did it sell or not?'” says Eric Fleet, CEO of Threads 4 Thought, a sustainable fashion brand. “With Stitch Fix, there’s still interpretation as far as why somebody may like something or not, but at least you have a lot of tangible feedback to make those determinations.” Threads 4 Thought’s XXL button-down shirts were also described as too small by men who categorized themselves as “husky.” By enlarging its dimensions, Fleet was able to improve the buy rate not only on his Stitch Fix–destined inventory but also across all his sales channels.


Lake made her first Stitch Fix shipment to 29 customers from her Cambridge, Massachusetts, apartment while still a student at Harvard Business School. Back then, in 2011, she and her small team were still eyeballing items: “This maxi dress? Never going to work for you.” Then she sought out Colson at Netflix for advice. He saw a world of possibility for algorithms to intervene but also less room for error. While Netflix viewers pay the same subscription fee whether they like a show or not, Stitch Fix’s profit model is the same as any retailer: It purchases clothes at wholesale, sells at retail, and only makes real money if someone buys the clothing. (A $20 styling fee is charged for each Fix, but is credited toward any item a client chooses to keep. In December 2017, Stitch Fix began to roll out a $49 annual Style Pass, offering unlimited Fixes. The company doesn’t release numbers but said Style Pass grew 60% quarter over quarter in 2018.)

Wall Street, though, has seemingly always wanted Stitch Fix to be something it’s not. Lake says she doesn’t get worried about the fluctuations in her stock price, given the company’s healthy financials. “We did what we said we were going to do for [2018]: We committed to growth rates, revenue, and profitability, and we delivered on all of that. As a public company, there’s a lot that’s outside of your control—and to some extent the stock price is one of those things in the short term—but it’s very much in our control long term. An important part of my job is to be able to help people focus on that.”

As much as Lake ultimately wants to define the future of retail, she’s also committed to modeling the future of Silicon Valley. Lake was once annoyed by always being cast as a female founder, but she has come around to serving as the example she lacked as an economics undergrad at Stanford. Back then, Lake believed founders were engineering dudes based out of garages—not driven, yet risk-averse business students like her. After graduation, she worked as an associate in venture capital firms for several years, hoping to find an early startup to join. Lake’s proximity to founders demystified them to her, and she realized there was no reason she couldn’t join their ranks.

Now a successful founder herself—Lake, after all, raised just $43 million before going public and growing Stitch Fix into a profitable, billion-dollar public company—she keeps pushing for a tech industry amenable to more Katrina Lakes. She insists she didn’t engineer the toddler-on-her-hip IPO moment to create a zeitgeisty photo op—an update of Diane Keaton in Baby Boom, but with the protagonist always working for herself. Lake says she just happened to be holding her son when it came time to approach the podium.

She’s joined the board of the female-led makeup startup Glossier, and played a prominent role in tech’s #MeToo reckoning. In 2017, reports revealed that she had asked for venture capitalist Justin Caldbeck to be removed as a Stitch Fix board observer for inappropriate behavior toward her, lending credence to the female entrepreneurs accusing him of sexual harassment (Lake signed a non-disparagement agreement preventing her from saying anything about it). She also urged Gurley to make changes to improve the culture at Uber in 2017, after the company came under fire for sexual harassment and other bad behavior.

Stitch Fix’s board is more than 60% female, and its tech staff is 35%—still not gender parity, but far better than the industry average and without hiring quotas. Interviewees are informed that the company values “bright” people over the purely book smart and “kind” people over nice. During the application process, instead of having to solve a technical problem alone, candidates are paired with a nontechnical staffer from styling or merchandising to collaborate, which quickly surfaces inventive applicants. Lake has also been adamant about fostering work-life balance. Most notably, she provides 16 weeks of parental leave to all full-time employees who are primary caretakers—whether they work in data science or a warehouse. It never occurred to Lake to create the kind of caste system of disparate benefits for different types of workers that’s prevalent at many tech companies.


She took the full 16 weeks herself this winter after the birth of her second child in November. “There might’ve been times, years ago, where I would’ve felt a little bit more uncomfortable taking the leave,” Lake says during her last full week in the office in the fall. “We have lots of women at Stitch Fix who are growing their families and also doing a great job here. Being able to take a leave is the right thing for your family. It’s also the right thing for your work so that you can come back and be focused and be excited.”

Mike Smith, Stitch Fix’s COO, led the December earnings call. Lake Instagrammed photos of her baby throughout the month, and a shot of her 36th-birthday cake on Christmas Eve. “The reality is my team is strong and everything’s going to be just fine,” she tells me in a breezy tone that sounds like she really means it. Because this isn’t just about her or Stitch Fix. “I’m not going to be the last one who is going to be pregnant and be a public company CEO.”