The Science of Style at Stitch FixEric Colson / Chief Analytics Officer at StitchFix
Stitch Fix is a popular styling service for women. The company’s personal styling application is one of the main reasons for its popularity and the focus of the Technology team, including its data scientists. The application blends the science of machine learning algorithms with the art of expert-human judgment in exactly the right ways to delight Stitch Fix customers
The Great Divide
Eric Colson is Chief Algorithms Officer for Stitch Fix. Blending people and systems effectively is his focus. The issue I had in all previous companies was the divide between algorithms – or data science – and analytics – or business intelligence,” says Colson. “The problem with having two teams is that you fragment context. Each team does their job in a silo even though they’re working with the same business partners.”
Before Shiny, data scientists put data into files, which business users loaded into Excel. With Shiny, a few more lines of code delivers what the business user needs directly from R.
To blend these disparate functions into one, Stitch Fix challenged the data science team to do more.
“We need people with full context and diverse skills. Our data scientists need the ability to both develop algorithms and provide visibility to the business in the form of dashboards and apps.” says Colson. “So, we absorbed the analytics role into the data scientists and said, ‘You’re going to do both.’ It’s extra work but the greater context makes them better.”
Next, he considered how to equip the team to deliver on this unified approach.
“The data scientists were already fluent in R and Python,” continues Colson. “We needed tools that leverage these technologies for building analytic applications.”
For R development, Stitch Fix chose the RStudio integrated development environment, the open source package Shiny, a web application framework for R, and Shiny Server Pro , the commercial server for Shiny applications.
Enabling Curated Analysis
Stitch Fix chose RStudio products because they enable business users to access the analyses done by the data scientists.
“The data scientist are almost developing dashboards as a by-product,” says Colson. “As they analyze data they produce visualizations and metrics that are extremely useful to their business partners. With just a few more lines of code the data scientists can turn that work into a web app that makes it available to the entire organization.”
“Shiny lets us deliver curated analysis to the business,” Eric explains. “The apps we create provide flexibility but they are not overly ad hoc. They are visual and elegant, but also austere, providing just enough interaction. We chose Shiny Server Pro because it gave us the ability to authenticate Shiny app users and came with support from RStudio. Long before I ever bought, we were using Shiny. It was apparent that it came from credible people who were very passionate; and it has a great community rallied around it.”
A Data-Driven Culture
“It’s hard to properly associate dashboards to the business value they provide,” concludes Colson. “They raise the knowledge of the organization, but that can’t be A/B tested. Yet, each day there are new insights from them that support better decisions. It’s become so second nature to our data-driven culture that we don’t often acknowledge the tools that enable it.”
So far this year Stitch Fix has deployed a ‘couple dozen’ Shiny apps to their Algorithms and Analytics portal. About half of the company’s employees touch the applications at least once a week. The data science team also uses Shiny applications to monitor how well their algorithms are performing.
As Colson says, “We proudly eat our own dog food.”
Curated analysis contributes to a better personal shopping service. Stitch Fix is the first fashion retailer to blend expert styling, proprietary technology and unique product to deliver a shopping experience that is truly personalized for you.
RStudio provides open source and enterprise-ready professional software for data science teams using R and Python.