Personal Style Gets AI Assist at Stitch Fix

Algorithms are designed into the clothing company’s business model and culture

By Thomas Davenport

Stitch Fix is one of the more interesting and faster-growing retailers of the last decade. Founded in 2011, its 2020 revenue — though hurt by the sluggish COVID-19 economy— was $1.7 billion, and it had 3.5 million active clients. Retail personalization services are becoming standard practices in the industry, but at Stich Fix, the online personal styling service relies on AI algorithms alongside human stylists to make client recommendations of clothing, shoes, and accessories.

The goal of the combined human-machine sources of intelligence is to provide clients with a “Fix” — a mailed box of five personalized clothes choices — that are a close fit to their style, size, and price preferences. Customers can keep the selections or send them back at no cost. Once on board, the clients can either continue to receive mailed boxes, or they can directly order recommended items from the website.

Stitch Fix now offers men’s and children’s clothing as well as women’s, and serves U.S. and U.K. customers. There are now 5,000 stylists across the U.S. and almost 150 data scientists at the company. Data science and algorithms were at the company’s core from the beginning, and Stitch Fix was perhaps the first company to have a Chief Algorithms Officer (Eric Colson, now CAO Emeritus). Various approaches to AI are in use, but statistical machine learning is the primary tool. Machine learning models are used to inform styling, marketing, supply chain, customer service, and many other aspects of the company’s operations.

Data Science is Fashionable

The styling algorithms team is about a third of the data science team. Data scientists collect and use as much data as possible, including an initial style quiz for each client when they sign up. They also get considerable client preference data from a “Style Shuffle,” a Tinder-like online quiz in which clients are encouraged to give their quick reactions to a series of clothing items. All client responses, and particularly rejections of mailed items, are considered carefully and incorporated into styling algorithms.

Tatsiana Maskalevich is Director of Data Science at Stitch Fix. She also styles clients — a skill Stitch Fix teaches every full time employee, whether a data scientist, a software engineer, or an accountant. It’s an unusual combination, but it helps her personally understand how algorithms and styling advice interact with each other.

Algorithms make recommendations to stylists, and stylists can choose whether to accept or modify them based on their knowledge of the client and the context.

Maskalevich explained that stylists both supply data to Stitch Fix’s algorithms and use them to help make styling decisions. “The whole process of styling is a balance between data and human judgment and relationships. Stylists and customers make certain choices and not others. All of their actions are captured and used to refine recommendations. The stylist role is to understand the nuance of clients’ personal style, create a connection with clients, and build a long-lasting relationship. It takes extensive interviewing and onboarding to find the right kind of person,” she said.

The styling algorithms team works closely with stylists, Maskalevich said. Sometimes they offer “algorithm schools” to the stylists to give them a high-level idea how the algorithms work. They explain that the models are based on many different features — size, expressed client preferences, reactions to previous Fixes, choices in the Style Shuffle, etc. Maskalevich said that it’s important for stylists to know that the recommendations are based on many different features that they would not be able to keep in their heads. Both automated and human coaching is offered to stylists that lets them know how they are doing in keeping clients happy with their choices. The human coaches can discuss with the stylists their mix of art vs. science, and how it is working out in terms of satisfied clients.

Stylists are always able to override the recommendations of algorithms. The primary advantage that human stylists have over algorithms is that they know the context for the clothing.

Stylist Request Notes (which clients can fill out when they order), are the primary vehicle for context. General statements such as “I don’t like pink shirts,” or “I want a flowered dress,” can be interpreted by a natural language processing (NLP) algorithm and acted upon, e.g., by adding a flowered dress to the initial recommendation or purposefully removing pink shirts.

However, client statements on request notes like, “My husband is returning from being stationed overseas for 12 months,” or “I’m going to a wedding that my ex will also be at,” or “I’m just about to start a new job and need to dress to impress,” are so nuanced and infrequent that NLP algorithms have neither the sample size nor the compassion to adequately address them.

A human stylist is able to truly understand the importance of these contextual comments that may lead them to override the recommendation of the algorithms. When the stylist and the algorithm disagree, Maskalevich explained, “We capture that data point.”

“Algorithms are great at taking all the data into account. People are great at taking the context into account and making subjective judgments. If a stylist is rejecting the algorithms and still doing well, that’s great, but in general we think stylists who use the data are more successful. They will see their level of success in the metrics and the outcomes, and adjust their choices.

Overall, the algorithms make their jobs easier, and overriding the decisions makes the job harder.”

Stylists with long tenures, she says, typically become very good at their jobs, but the algorithms improve over time as well. When she styles clients, Maskalevich says she is always trying to get better at the role, but perfection is difficult to reach. “Everything is changing all the time,” she noted. “There are different style trends, people have sudden shifts in preferences. I could not have anticipated, for example, the ‘sweat-pandemic’ we have been going through.” She added that being comfortable with change is an important aspect of the stylist mindset.

Even for a data scientist, Maskalevich says the stylist job can be quite intellectually demanding; “You can’t get in a rut,” she said. The most important part of the job is establishing and maintaining the connection with each individual client. She has been working with one client for five years; her client has shared vacation photos and they have shared life moments over Fix request notes and styling notes. Maskalevich gives considerable thought to what items she’s going to send the client next. “That really makes the job engaging,” she commented.

A Blend of Art and Science

Caitlin Yacopetti is a Styling Supervisor at Stitch Fix with over six years of experience supporting a team of stylists to deliver great experiences for clients. She said that the styling at the company is a blend of art and science. The science, of course, is the algorithm-based recommendations that help them make informed decisions on what they send to the client, whether recommending items based on their style preferences or making suggestions for clothes that will fit them best. Yacopetti said that stylists also get a really good understanding of clients — where they live, their lifestyle, and fit challenges — without ever meeting them.

She also described the art component of the job: “What the algorithms can’t do is interpret the nuanced preferences or feedback a client might have, or interact with clients in a personal way like people can. That’s the art of it. Especially during these times, people are craving that human connection to provide a bright spot in the day. That’s something an algorithm can’t offer!”

Algorithms “help to remove our own personal style biases so that we don’t get distracted by the item that might match our own personal taste, and instead focus on the items that are right for the client. The algorithms also provide the efficiencies that enable personalization at scale.”

Human stylists are an integral part of the Stitch Fix business model, and it seems unlikely that the need for both client/stylist relationships and interpretation of unusual client comments will be satisfied by AI alone. For now, Stitch Fix will further optimize its blend of art and science in the service of its clients buying and wearing stylish clothes.

Originally published at



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
MIT IDE Paula Klein, Editor

MIT IDE Paula Klein, Editor

Addressing one of the most critical issues of our time: the impact of digital technology on businesses, the economy, and society.