How Stitch Fix Is Using Algorithmic Design To Become The Netflix Of Fashion


Data science is baked into the online fashion brand.

[Photo: Stitch Fix]
Hybrid A [Photo: Stitch Fix]
A Parent 1 [Photo: Stitch Fix]
A Parent 2 [Photo: Stitch Fix]
A Parent 3 [Photo: Stitch Fix]
Hybrid B [Photo: Stitch Fix]
B Parent 1 [Photo: Stitch Fix]
B Parent 2 [Photo: Stitch Fix]
Summer Hybrid 1 [Photo: Stitch Fix]
Summer Hybrid 2 [Photo: Stitch Fix]
Summer Hybrid 3 [Photo: Stitch Fix]
Summer Hybrid 4 [Photo: Stitch Fix]

Algorithms control many elements of our lives: the music we hear, the shows we watch, the stories we read, and even how we vote. They’ve been used most commonly to uncover consumers’ preferences and make recommendations, but now brands are starting to integrate them into the products themselves.

Stitch Fix, an online apparel retailer, thinks algorithms are the future of designing garments and has begun to bring these items to market. (In her 2017 report, trend forecaster Mary Meeker of the venture capital company Kleiner Perkins mentioned Stitch Fix’s algorithmic design approach as an e-commerce “aha.”) But will this strategy yield better things, or will it just lead to the normcorization of design?

“[Our customers] come to us because they don’t want to go shopping,” Eric Colson, Stitch Fix’s chief algorithm officer and head of the company’s 75-person data team, says. “Five years from now, people will say, ‘Remember when we had to wander malls and find our own things? That’s crazy!’”

Since its founding in 2011, Stitch Fix has slowly been gaining steam and has grown to over 5,700 employees, has been profitable since 2014, and generated $730 million in revenue last year.

Here’s how Stitch Fix normally works: Customers join the site, fill out a detailed questionnaire about their size, how they like their clothing to fit, what their style is like, what colors they love and loathe, and how often they dress for certain occasions (like work, events, dates, etc.). Last year, the company introduced a Pinterest integration that lets the company learn more about what customers want. Users can create boards of images they like–which can come from any source–and algorithms analyze the assortment and feed that information into a customer’s profile. Using all that data, an algorithm then mines Stitch Fix’s inventory to find pieces that match their profile.

After the algorithms do their work, human stylists take over. Someone looks over the final selection and offers styling suggestions–like how to accessorize or wear the pieces–before the items are boxed up and shipped to the customer. When she finally gets the items, she can choose to keep or return them. Stitch Fix tracks what the person likes and doesn’t like and uses those data points to inform the next batch of items the customer receives.

Last year, Stitch Fix began using algorithms not only to pick clothes for its customers but to actually design new pieces. So far, the initiative, called Hybrid Design, has machine learning to develop over 30 pieces using this methodology.

[Photos: Stitch Fix]

“We noticed gaps in the market and an opportunity to produce something that doesn’t exist, but should,” Colson, who was formerly the head of engineering and data science at Netflix, says of the Hybrid Design initiative.

Stitch Fix meticulously catalogs all of its inventory and breaks down each garment into anywhere between 30 and 80 granular characteristics, such as color, length, how many buttons it has, what shape its hem has, what fabric it’s made from, what type of pattern it has, sleeve opening, collar type, and so on. Because so many theoretical combinations can exist, an algorithm is needed to analyze them.

The program can then discover which traits are most popular with consumers–based on their profiles–and see if they overlap on a specific article of clothing. If they don’t, that tells Stitch Fix there’s a gap in the market. The thought process is, if these details do well on their own, would they do even better if they were on the same garment? It’s the equivalent of trying to design a supergroup or all-star team.

Stitch Fix still relies on human designers to come up with the actual garment using traits that the algorithm serves them. “It’s machine learning with expert human judgment,” Colson says.

The big question with algorithmic design is if the products are any good. Stitch Fix says its Hybrid Design garments–which make up less than 1% of the company’s total inventory–have been performing strongly. That said, none of them are particularly adventurous pieces or fashion-forward: flowy sleeveless tops, non-offensive polka dot patterns, flattering drapey tunics, and lacy shirts with shoulder cut outs. They’re perfectly wearable pieces that the no-hassle shopper Stitch Fix attracts would probably enjoy.

I asked Colson if he’s concerned that creating garments based on the most popular characteristics would yield too much similarity. “We have this model about why people choose clothes,” he says. “The reality is two reasons. One is the innate desire to fit in–you do want to look like everyone else. That’s how trends happen. People dress the same and want to fit in. We also have this innate desire for individualism. Our algorithms are geared to both.”

For example, because Stitch Fix’s model is based around highly personalized data as opposed to marketing profiles commonly used in product development–like the working mom, the young professional, the single man, etc.–Colson argues that there isn’t a one-size-fits-all design solution. What their success hinges upon is having a deep enough product assortment and the exact right garment for a specific customer. Since their Hybrid Design algorithm lets them analyze what customers want against what current inventory is like, they’re better equipped to fill the need.

“I don’t think the consumer cares, as long as it works,” Colson says about Stitch Fix’s use of machine learning to design clothes. To him and his team, it’s just another way to better serve people, offer more variety, and fill gaps in the market. Algorithmic design is one avenue to get there. “Our two goals are to satisfy existing need, and fulfill more clients’ needs,” he says. “Personalization is having a deep enough product assortment for everyone.”

This article first appeared in

Seeking to build and grow your brand using the force of consumer insight, strategic foresight, creative disruption and technology prowess? Talk to us at +9714 3867728 or mail: or visit

About Author

Diana Budds

Diana Budds is a New York–based writer covering design and the built environment. She's a staunch defender of Brutalism.

Comments are closed.