Stitch Fix: Using Machine Learning to Help The Grinch

Stitch Fix uses machine learning to help you answer that age old question of what to wear

In the Jim Carrey version of The Grinch Who Stole Christmas, he despairs, “but what will I wear?!” when fronted with attending Christmas festivities. While she may have not had the green Dr. Seuss character in mind when creating Stitch Fix, founder Katrina Lake was essentially trying to solve that problem—helping women figure out what to wear without breaking the bank. In other words, how could you make apparel personal stylists more scalable and affordable for consumers?

According to Forrester, online sales will account for 17% of all US retail sales by 2022[1]. As some major retailers are filing for bankruptcy (e.g., Sears, Toys R US), others are scrambling to figure out how technology, big data, and digital efforts can help them compete in a space that’s increasingly online. The beauty of Stitch Fix is that Lake saw an opportunity to take a data-driven approach to solving some of the pain points seen in both brick and mortar and traditional online apparel shopping. In other words, a proactive approach to serving customers based on industry-wide pain points, rather than a traditional retailer taking a reactive approach to navigating an evolving landscape.

In an interview, Lake describes early tests with ~20 individuals, trying to figure out whether it was even feasible to make personalized recommendations without meeting in person.[2] Initially, she started out making the recommendations to customers herself and including forms for users to share feedback. Even though they have been data-driven from the start, Lake needed to take a more rigorous approach to data science after amassing more users. Machine learning has been key to scalable recommendations for clients, and they have been able to implement a data feedback loop. They have up to 100 measurement points per garment, which is married up with the feedback they get from their 2+ million person user base to further enhance the recommendations they are able to make[3]. This ever-increasing pool of data from different sources is critical for not only enabling the business model, but also for improving over time since “results improve as the amount of training data they’re given increases.”[4]

While Stitch Fix was only founded in 2011, a longer term view of their trajectory shows they have since used machine learning to expand into additional offerings (Mens[5], Plus Size[6], Kids[7]).They are now also using machine learning for more back-end functions like inventory management—something their data scientists developed by their own volition[8]. Interestingly, inventory management and inventory holding was part of the business model that investors saw as a risk for Lake early on.

Moving forward, I would be concerned with how Stitch Fix maintains its competitive edge. If they aren’t already, I would try to incorporate unstructured data sources to supplement their current data. Leveraging unstructured data such as social media (e.g., Instagram comments, blog posts) could be beneficial for improving offerings and customer loyalty[9], which is increasingly important with the rise of competitors such as Trunk Club. Stitch Fix has called out some objective comparisons with competitors on their website which shows they are actively thinking through their comparative value proposition.

Figures from Stitch Fix website[10]

Two final questions I have are 1) is there transferability of their algorithms to non-apparel purchasing?, and 2) should they think about monetizing some of their technology for other retailers? One example for the first is Sephora claiming they have “trained a program that can measure [facial features] in real time” [11] to enable recommendations and virtual try-on. While the appeal of machine learning for beauty recommendations mirrors that of retail, Stitch Fix’s model hinges on the ability to send clothes back (which you couldn’t resell in cosmetics). This begs the question of whether there are other subsets of retail that are conducive to this model. And finally, I wonder if monetizing their inventory management machine learning algorithm would be beneficial or whether that point of differentiation will keep competitors from succeeding.

 

Word Count: 797

Footnotes:

[1] “Online Retail Forecast, 2017 to 2022 (US)”, Forrester (August 2017)

[2] “Stitch Fix: Katrina Lake,” How I Built This with Guy Raz (April 2018)

[3] “Stitch Fix’s CEO on Selling Personal Style to the Mass Market,” Katrina Lake, Harvard Business Review (May-June 2018)

[4] “What’s Driving the Machine Learning Explosion?”, Erik Brynjolfsson & Andrew McAfee, Harvard Business Review (July 2017)

[5] “Stitch Fix launches Stitch Fix Men”, PR Newswire (September 2016)

[6] “Stitch Fix adds to its portfolio by launching Stitch Fix Plus”, PR Newswire (February 2017)

[7] “Stitch Fix launches Stitch Fix Kids”, PR Newswire (July 2018)

[8] “Stitch Fix’s CEO on Selling Personal Style to the Mass Market,” Katrina Lake, Harvard Business Review (May-June 2018)

[9] “The benefits of machine learning for retail”, Cape Town: SyndiGate Media Inc. (May 2017)

[10] https://www.stitchfix.com/stitch-fix-vs-competitors

[11] “How Sephora is leveraging AR and AI to transform retail and help customers buy cosmetics,” Alison Rayome, Tech Republic: https://www.techrepublic.com/article/how-sephora-is-leveraging-ar-and-ai-to-transform-retail-and-help-customers-buy-cosmetics/

Previous:

The Free Hand of Mickey Mouse & The Hobbyist Printer

Next:

India’s InMobi is Disrupting Online Advertising by Making 15 Billion Decisions Per Day

4 thoughts on “Stitch Fix: Using Machine Learning to Help The Grinch

  1. I think there is definitely transferability with Stitch Fix’s algorithm to other industries. As you mentioned, Lake’s team created different algorithms on their own volition. With the right data scientists, I believe machine learning can be applied to many industries. I actually just watched an interview of Katrina Lake and she mentioned that she could see this business model being applied to the travel industry very well, something she may consider down the line, but thinks it would be transferable across many industries. I definitely agree – everything these days is data-based and when you have a large data-pool you should be able to utilize machine learning to improve.

    1. I also agree with the above comment that Stitch Fix’s algorithm is transferrable to other industries. I would argue that it may be more relevant for certain other sectors. For example, within the beauty space, we see high repeat purchases – consumers have very particular preferences and makeup intrinsically runs out, both resulting in high predictability of purchases. As consumers are shifting purchases from retail to the DTC channel, I can see machine learning as a differentiating factor and a way for brands to disrupt the traditional beauty industry.

  2. Your question regarding monetizing their inventory management machine learning algorithm is fascinating, because, in some ways, it would move Stitch Fix from a retailer to a service provider. In a sub-sector of retail that’s quite competitive, you’re able to add an additional revenue stream and potentially gain access to more data to improve your offering as well. I do worry about the overall sustainability of this business as barriers to entry are so low and customer service and quality of recommendations matter a lot. Stitch Fix needs to find way to beat out its competitors and became the “Bandaid” of the subscription service and personal shopping market.

  3. “I wonder if monetizing their inventory management machine learning algorithm would be beneficial or whether that point of differentiation will keep competitors from succeeding.”

    I am such a fan of this company. 1) The CEO is a Harvard alumna and 2) she is super young! What an amazing example of leadership for all current HBS students and women entrepreneurs. With that being said, I the threat of competition in the apparel industry, especially concerning the use of machine learning to predict one’s wardrobe, is a real and urgent one. I thought the Stitch Fix model was unique, until I noticed that Amazon came out with a similar product. I don’t believe that Amazon uses machine learning to pick your wardrobe, but it does allow customers to pick items, try them, and then return them out. I would not be surprised if Amazon develops its own algorithm that will recommend different styles to customers. If that happens, Stitch Fix will have to determine a new competitive advantage.

Leave a comment