Stitch Fix: Your Ideal Fashions are IN the Computer (yes, that’s a Zoolander joke)

Do you ever wonder whether you’ll actually use any of the lessons we’re learning at HBS? Katrina Lake’s usage of data and deployment of machine learning at Stitch Fix, proves that paying attention in class pays off.


Founded in 2011, Stitch Fix is a fashion e-commerce company that takes the shopping out of fashion without sacrificing style. Users sign up for this on-demand or subscription service in which they receive personally curated collections of clothing and accessories based on a combination of data-driven predictions and human stylist oversight. Based upon an initial survey, Stitch Fix uses machine learning algorithms developed by over 75 data scientists to predict which items will most delight the shopper.[1] Human stylists then use the algorithm output to select five perfect items for the user. The shopper simply pays for the items she likes and returns the misses, only paying a $20 “styling fee” to Stitch Fix for its services.[2] Despite only raising $42 million, Stitch Fix reportedly reached $375 million in revenue in 2016 and has achieved profitability.[3] 

Value Creation

Stitch Fix allows shoppers to find a customized wardrobe with minimal time and energy. Additionally, Stitch Fix secures the items at a wholesale price point from brands and offers shoppers a 25% discount for keeping an entire shipment.[4] Finally, shoppers who use Stitch Fix are relieved of irrational decision anxiety (how many times have you had a wish list item sitting in a shopping cart in your browser for weeks? I’m certainly guilty of this) and are simply pushed items that are a good fit for them.

How can the company do this? After the input of preference data via an initial survey of style, fit, and price preferences and optional Pinterest board or Instagram handle sharing (a marginal inconvenience – it took me less than 10 minutes), nothing is required of shoppers besides either buying or returning their Stitch Fix shipments. This process provides Stitch Fix with feedback data that makes the company smarter about fashion trends and shopper preferences.[5] The company’s ability to digest this structured data (surveys/returns) and marry it with the stylists’ interpretation and categorization of unstructured data (Pinterest selections/notes about upcoming special occasions) allows it to create immense value for shoppers who don’t have the time to shop for themselves.[6]

Stitch Fix is also using creative data analysis to improve the success of recommendations, and thus the value for customers. The company has done extensive testing in determining what information to share with stylists to prevent bias and promote the most successful recommendations.[7]  Through validation systems using A/B testing, Stitch Fix can better understand whether sharing a client’s location, photo, or other attribute with a randomly selected group of stylists results in less returns (details of this experiment can be found here).[8]

Additionally, Stitch Fix has developed advanced machine learning capabilities that allow the company to create proprietary fashion items that “fill in the gaps between what’s commercially available and what customers tell Stitch Fix they want.” [9] Stitch Fix’s computers analyze purchases and fashion trends with countless attributes to create “hybrid designs” that incorporate the most popular features (see infographic from WSJ to the right). While these designs currently account for a minuscule percentage of Stitch Fix’s business, this capability allows the company to better customize offerings to shoppers whether or not other fashion brands have created the necessary styles.[10]

Value Capture

Stitch Fix’s revenue model is very straightforward. It charges a $20 styling fee regardless of whether the shopper keeps the items. However, this amount is credited against the customer’s purchase if they keep any of the items. If the company continues to successfully create and sell hybrid designs via machine learning, it has the opportunity to supplement this revenue stream with additional income from clothing sales.

Looking Forward

While Stitch Fix seems to be performing well at present and is reportedly considering an IPO, it should be aware of some challenges and opportunities on the horizon.[11]

  • Presently, Stitch Fix does a tremendous job of combining algorithmic analysis with human judgment. As the company continues to scale, it would be better if current stylists became better and faster as opposed to human capital growing in lockstep with users. Even more extreme, will computational creativity make big enough strides such that human stylists are not even necessary?[12]
  • After recently expanding into both menswear and women’s plus-size fashions, Stitch Fix will certainly attract more attention from larger competitors. There exists a risk that tech-enabled e-commerce players like Amazon attempt to replicate the Stitch Fix model.














Freebird: Using data to get you out of a jam and onto a plane


Big Data for Music Festivals

Student comments on Stitch Fix: Your Ideal Fashions are IN the Computer (yes, that’s a Zoolander joke)

  1. Thank you for the great post, Christy. Seems that Stitch Fix is doing an excellent job figuring out and predicting what styles people will be interested in wearing. You mention that they are currently growing their human capital of stylists as the user base grows – do you think that is necessary? Interestingly, one of their competitors Trunk Club is coming exactly with that value proposition of offering every user “their personal professional stylist”, relying on the premise that people do not trust a computer algorithm recommending them what to wear. As we talked in class today, it would be pretty difficult to figure out a perfect algorithm that is able to take so many variables and constraints into account.

    What about Le Tote’s model of suggesting you clothes, shipping them to you, and letting you wear them – it’s a bit like Rent the Runway where you also have the option to buy and keep what you like. I prefer this model as a user – I can wear the cool items once or twice and then get new ones selected for me. Do you think letting people wear the shipped clothes increases or decreases the chance they keep them?

    1. thanks for the comment, Lidiya. No, I don’t think they should continue to grow human capital at the pace of new users. I think they idea is that their algorithms improve enough such that less stylists are needed to refine the results. On your second question, I think letting people wear the clothes definitely increases the chance that they keep them but I’m not certain how their branded partners would feel about this. Since Stitch Fix doesn’t own the inventory it would be a complicated task for them to get collective agreement on this policy.

Leave a comment