Multi-Threaded

“The approach we’ve taken is much more like being a gardener. You just want to create circumstances where people can do good work and, occasionally, you need to trim a branch back or make room for a new sapling. Generally, you’re just trying to get the conditions right to then get out of the way.”

^{Brad Klingenberg, Chief Algorithms Officer, Stitch Fix}

Millennials young and old will remember the opening scene of the 1995 film “Clueless”, with its cheerful protagonist Cher. The valley girl opens an application on her microwave-sized computer to sort through a digital conveyor of outfit ideas for the day. The “Clueless Closet” idea may have sprung from the mind of filmmaker Amy Heckerling, but today you can actually buy custom styled outfits just like Cher, thanks to the data team at Stitch Fix.

There Goes Your Social Life

Stitch Fix Founder Katrina Lake launched the online apparel company in 2011, seeking to make well-styled fashion less time-consuming. Lake operated the styling and shopping service at cost out of her apartment in Boston while she attended Harvard Business School. She scheduled her classes to accommodate frequent trips to Silicon Valley. Eventually, Lake earned a small investment from Steve Anderson, an early Instagram investor, to help scale her new business, and she moved to San Francisco.

In a bold move for a young fashion company, Stitch Fix decided to focus on data science. As Jesse Anderson writes in Data Teams, “Stitch Fix made the integrity and veracity of data an early priority . . . this high-quality data meant that the data scientists could focus on value creation instead of trying to decide if they could trust the data.” Chief Algorithms Officer Eric Colson made sure of that. Colson joined from online movie rental pioneer Netflix in 2012. He was responsible for standing up the first data team at Stitch Fix. His first three hires: data scientists.

By 2017, the company had gained more than 2M customers. Lake became the youngest female CEO (not to mention one of Korean descent) to take a company public. That year Colson released the “Algorithms Tour” website which boosted the company’s pedigree as a data science innovator and earned praise in tech publications like Wired. Hilary Parker, one of the company’s early and most well-known data scientists, admitted, “I never realized how important PR departments are to do this type of thing . . . I was a tool for this strategy.”

Driving in Platforms

Lake, Colson and their successors have deliberately aimed to differentiate Stitch Fix through data science. That meant creating an executive leadership role. As Chief Algorithms Officer, Colson divided his department into two groups: the Data Science team and the Algorithms Platform team. Both reported to him. “You find it often buried in marketing or engineering or finance or awkward other places, and we felt like it had to have its own thing, because data science has its own tooling, workflows, ethos, all this stuff that has to be unique to it,” Colson explained in an interview with Venture Beat.

The Data Science team, with more than 100 members, handles analysis and model building. Within the team, smaller groups with hands-on managers specialize in key functions like styling (i.e., the recommendation engine), merchandising or inventory management, client operations, and customer service. “Having the support of my manager and the skip-level was huge,” Parker mused on one of her podcasts. The data scientists had the ability to build their own data pipelines and models, and even deploy them to production themselves. The data teams often collaborated closely with business teams to ensure their models made an impact. Brad Klingenberg, who succeeded Colson in 2019, made “impact” one of his signature achievements. “Brad’s department is accountable for real impact to various metrics, whether it’s increasing revenue or optimizing to reduce costs.” Colson stated in Data Teams, “In addition to their own impact, they also partner with nearly every department in the company - marketing, merchandising, operations, styling, etc. - to provide algorithmic capabilities embedded in those functions. That is pretty unique to Stitch Fix: that real sense of accountability.”

Meanwhile, the Algorithms Platform team established and maintained the infrastructure and tools for data science. Its members worked to remove the complexities of computer processing, error monitoring, and other aspects of infrastructure while minimizing “hand-off” to researchers. They managed the migrations of the company’s original Ruby web development stack on a PostGRES database to its now all cloud-based architecture as well as the migration from Tableau to Flask/React. The team even developed an open-source python library for data visualization. Eric Magnusson, current VP of Platforms once wrote on his blog, “At Stitch Fix, we strive to be Best in the World at the algorithms and analytics we produce. We strive to lead the business with our output rather than to inform it . . . We are not optimizing the organization for efficiency; we are optimizing for autonomy . . clear ownership of ideas and accountability . . . These are roles that are very attractive to folks who embrace an entrepreneurial mindset.” A lofty vision for Infrastructure.

Rollin’ with the [Data Science] Homies

Within the data science teams though, the work was more fluid. According to Hilary Parker, “it’s not traditional data science work . . . the problem was too big for there not to be a hand-off . . .” In her case, she partnered with a machine learning modeler while she supported data collection and preparation. “I was looking at the org and [saw] that we’re over-indexed on people who like to write models and under-indexed on people thinking holistically about fashion.” For example, she partnered with the studio to collect clothing attributes from images. Then her partner on the data science team would incorporate that data into the model, improving its accuracy, thereby improving the likelihood of a consumer to buy. As Colson admits again in the Data Teams book, “it’s inherent that these types of algorithms can’t be designed upfront; they have to be learned as you go.” That is why the company encouraged what they call low-cost exploration. Data scientists have the latitude to find problems to solve through statistics and machine learning.

Stitch Fix’s distinctive roles for its data teams contributed to the company’s reputation as a data science leader, a boon for other aspects of the business. From the company's algorithms tour, its “Multi-Threaded” blog, and strong public relations, Stitch Fix became a darling of both Wall Street and Silicon Valley around the time of its IPO. That kind of attention has helped Stitch Fix garner recognition as a customer service leader in e-commerce broadly, not just apparel. Good PR also helps with with attracting talent. Practitioners with backgrounds as wide-ranging as astrophysics and retail, from companies like Google, Disney, and Pandora have woven into the fabric of Stitch Fix’s data teams. Even a cursory glance at LinkedIn reveals over half of the employees today at Stitch Fix use “data'' in their profile.

Totally Buggin’

By 2019, Stitch Fix’s model began to wear. Even the best data science and public relations will not cure all business ills. Stories about profitability concerns at the company began to appear in the media, pointing to the high labor costs for staffing the data teams along with that of stylists. Indeed, human-in-the-loop machine learning, the type of recommendation system used by Stitch Fix, requires the stylist’s involvement. “The way we think about our AI and ML when it comes to recommendations is that it’s really in service of our stylists,” one of the company’s recent Chief Technology Officers explained.

“At the end of the day no one cares what model you use,” former Stitch Fix data team lead Tatsiana Maskalevech has said. Some have even accused the company of “science-washing.” Still, Stitch Fix churns out new data products. The company’s web app Style Shuffle, like Tinder for fashion, has become more useful in collecting consumer preference data than an actual sales mechanism for sales in Parker’s perspective. The company’s Freestyle application diverged even further from the signature subscription box model by leaning into its “AI stylist.” Meanwhile, executive leadership offers clashing messages, promoting automation but also personalization by professional stylists. Today, sales at the company have shrunk and it has exited several geographic and product markets. Its future remains uncertain.

But in the words of Cher, “Ugh, as IF.” Stitch Fix still represents a great case study on how to market your data team. No other company since has had such precise messaging around its analytical products. The strategy is simple: a good PR strategy begets good press. That helps with sales, which generates more data for research. Stitch Fix has had a deliberate data approach with its C-suite role, functional teams, and freedom for data scientists. In that way, Stitch Fix has differentiated itself in the crowded marketplace of fashion retail. Most importantly though, Stitch Fix has proven that you and your data team’s success hinges on how well you sell yourself.