Stitch Fix Story: Algorithm vs. Human
How interested are you in fashion?
The word ‘fashion’ as we use it these days does not only mean ‘clothes’, but in a broader sense, it also refers to styles that are fashionable at a particular time, including automobiles, architecture, and housing. However, microscopically, it is enough to understand fashion as ‘a certain form of clothing or hairstyle that is popular at a specific time’.
Fashion has always been an important issue in the history of mankind, and the way people dress has also been important in a social sense. The luxurious clothes and ornaments found in Egyptian murals show that interest in appearance and clothing goes back thousands of years, and that fashion was not only for self-gratification but also an important means of revealing the social status agreed upon among many people.
In modern times, fashion has also taken a solid position as an ‘industry’. According to Kantar, a global consumer panel marketing research company, the size of the global fashion market is more than 300 billion USD in 2020, and it is expected to grow by 4% every year thereafter and expand to about 360 billion USD in 2025. Fashion United also analyzes that the total value of the fashion industry is 2.5 trillion USD, accounting for 2% of global GDP.
While the growth of the fashion industry is undoubtedly driven by SPA, about half of the growth will be achieved by leading global companies such as Fast Retailing (Uniqlo), Inditex (Zara), H&M Henness & Moritz (H&M), and TJX Group.
In the case of fashion distribution, due to the growth of a new core consumer group represented as the ‘MZ generation — millennials and the generation Z’ and rapid online movement in the aftermath of the corona pandemic, new platforms and startups are rapidly growing with unique offers enabling convenience and customization based on the combination of data and new technologies. Representative examples include France’s Heuritech, which predicts fashion trends by analyzing social media images; Korean online fashion platform Musinsa, which distributes more than 3,500 global and domestic brands along with fashion content and trend information; American startup Entrupy, which uses artificial intelligence to determine the ‘authenticity’ of luxury products within 3 seconds; and Stitch Fix which we are going to talk about today.
Start of Stitch Fix
Stitch Fix is a ‘personalized clothing subscription service’ founded in 2011 by Katrina Lake, a then-28-year-old Harvard University MBA student.
After graduating from Stanford University, Katrina has worked for Polyvore and The Parthenon Group (a US strategic consulting firm, acquired by Ernst Young in 2014, now called EY-Parthenon).
Polyvore was a ‘community-based social commerce service’. With a function called ‘Virtual Mood Board’, community members collected clothing, miscellaneous goods, interior products, etc., and created an image collage with the name of ‘Set’ on the board. Polyvore users communicated by sharing, commenting, comparing each other’s collages, competing, and more. Also, manufacturers including fashion brands utilized APIs to connect to this board, providing product information and link to product sales. Those of you who have not heard of Polyvore may have heard of SSENSE; SSENSE ended Polyvore’s service when it acquired Polyvore in 2018.
Based on her (fashion) online commerce experience and career at a consulting firm, Katrina seems to have developed Stitch Fix model that ‘combines the algorithm and the human stylist to provide a collection of clothing that users want in a subscription form’. You can also see the similarity between Stitch Fix’s term ‘Fix’ — a term for clothes shipped to subscribers in a box — and Polyvore’s term ‘Set’.
In the early days of its establishment, she used SurveyMonkey service to collect data that could be used to gauge the style of her initial customers, selected and delivered the clothes that satisfy customers’ desire, and received a monthly subscription fee directly from the customer — it is said that she didn’t set up the Internet payment function back then. In 2012, after a long persuasion process, she had Eric Colson — former VP of Data Science & Engineering at Netflix — on board as Stitch Fix’s Chief Algorithms Officer, and a full-fledged business system building and growth started.
How does the ‘Fix’ service work?
When signing up for Stitch Fix, the customer first goes through an online survey of their basic body figure, occupation, preferred fashion style and taste, price range, etc. The ‘Fix’ (a set of clothing) recommendation is generated by the algorithm and then is passed on to the human stylist, so he or she can select 5 clothing items for delivery. Additional information such as customer’s social media information and images can be used during the selection process.
Customers can set the delivery cycle as they wish: e.g., every two weeks, once a month, every two months. Shipping and returns are free, but a styling fee of USD 20 is charged for each Fix delivery.
On receiving Fix, the customer has 3 days to decide whether to buy the clothes as is or return some or all of them. If the customer purchases even one piece of a clothing set, the styling fee is deducted from the clothing price, and if you purchase all of the clothing in the Fix, an additional 25% discount is given to the total price.
Customers can leave a note not only about the Fix they have received, but also their request for a future Fix (eg, “I have an important social event next month, please recommend a dress”), and the stylist reflects the request on the next Fix and update the customer’s information. Through this feedback loop, the collaboration between the algorithm and the stylist creates a virtuous cycle to understand customers better.
Under the strong leadership of Katrina Lake and Eric Colson, Stitch Fix has made it a very important task to organically ‘weave’ the innovative technologies into all areas of its business while continuously expanding its data science, engineering, and algorithmic practices. Stitch Fix’s sub-site ‘Multithreaded’ explains in detail how Stitch Fix applies innovation and technology not only for personal styling, but also for warehouse, merchandising, and client experience. In particular, in the ‘Cultivating Algos’ section of ‘Multithreaded’, you can take a high-level look at how Stitch Fix’s organization, roles, and work processes are structured differently from typical companies to rapidly and continuously promote data-driven innovation.
However, more important is the philosophy behind this startup’s founder, Katrina. She said “shopping is fundamentally a very personal and human activity, so proper collaboration with stylists who can sometimes change or reject the results recommended by algorithms at their discretion is of paramount importance”. And in an interview with Harvard Business Review magazine in 2018, she also said: “It’s simple: a good person plus a good algorithm is far superior to the best person or the best algorithm alone.”
Growth, IPO, and the Pandemic Crisis
Based on this unique business model, Stitch Fix grew rapidly — realized operating profit in 2014, and recorded approximately 2.2 million active users and 1 billion USD in sales in 2017, and was successfully listed on NASDAQ in November 2017.
However, Stitch Fix’s business performance did not show the same level of high growth after the IPO. Stitch Fix’s FY20 investor presentation is below:
Active clients (those who have made at least one purchase within a year) and sales are showing stable growth, especially considering that other giant apparel brands’ sales have declined by more than double digits during the coronavirus pandemic situation. However, the growth rate of active clients fell by more than half from 18% in 2019 to less than 9% in 2020; and both overall sales volume and sales per active client fell compared to the previous year.
In a nutshell, investors are faced with a situation not to consider Stitch Fix as an attractive investment target. (Stitch Fix was also dubbed a ‘Broken IPO’ as its stock price fell in the 10s of USD in mid-2020 — of course, it should be taken into account that we were amid the impact of the coronavirus pandemic)
The other side of the crisis: inherent risks
Behind this crisis Stitch Fix faced, there were some inherent risks from my perspective.
Chase of competitors with innovative technologies and services
At the time of its founding, Stitch Fix’s business model may have been quite differentiated and unique. But after the pandemic upended everyone’s lifestyle, traditional retailers have also quickly expanded their business capabilities and scale online services, similar to Stitch Fix in some cases. Examples include Macy’s clothing subscription service, Nordstrom’s Trunk Club, and Amazon’s Prime Wardrobe.
According to Recurly Research’s benchmark, the ‘Box of the Month’ type of subscription service shows the churn rate of 10.5%, twice the average of various types of subscription services including media/entertainment, educational contents & services, consumer goods, and groceries.
In the end, as traditional retailers launch a lot of ‘similar’ services, the cost of new customer acquisition increases, and downward pressure on business performance naturally follows.
Scalability of Stitch Fix Service Model
The second issue lies in Stitch Fix’s service model — which combines algorithm and human stylist. This combination, when carefully implemented, seems ‘ideal’ at first glance, but it has potential issues in scalability in the context of personalization styling and distribution business.
In the US market, the return rate of online commerce is around 25–30%, three times bigger than that of offline commerce. F&B (and anything close to ‘commodity’) are on the lower rate side, and fashion/clothing is on the higher rate side. I expect a similar pattern in subscription services, even higher than typical online commerce in the case of the fashion/clothing business. Of course, a high return rate is not necessarily just a bad thing because retailers can leverage ‘returns’ as a strategic link to deepen relationships with their customers, but in principle, a high return rate leads to the ‘bad inventory problem’ especially considering the sensitive nature of fashion to the fast-changing trends and personal choice.
In addition to the ‘bad inventory problem’, all the elements of human stylist involvement for style selection (when the average purchase price of 30 to 50 USD), client communication, and feedback to algorithms inevitably help the operation hit the limit of scalability after a certain level of transaction size.
Limitation of ‘Subscription’ Model Penetration
The penetration of subscription services in the overall apparel market is still very small, accounting for only 1% of the total size in the US market. Stitch Fix occupies about half (~0.5%) of this, which makes it the undisputed leader in the clothing subscription service market. But, do you expect the majority of people will change their clothing shopping behavior towards a ‘subscription’ service rather than having the freedom and flexibility to choose their fashion styles themselves?
Stitch Fix’s ‘Freestyle’: New (but not new) BM
In the end, Stitch Fix seeks to transform its business model to overcome the crisis and bring about a reversal in the market atmosphere.
Stitch Fix has continuously tried to change its service model little by little in the process of growth since its foundation — expansion into men/kids segment, penetration into the UK market, the introduction of ‘Styling Pass’ (which charges only 49 USD as the annual styling fee), the introduction of ‘Extras’ (which allows customers to add desired props to the Fix recommended by the stylist), to name a few.
However, the biggest change is the addition of a business model that allows customers to choose their clothes from Stitch Fix inventories. Originally released as ‘Direct Buy’ in 2019, this new model allows existing customers of Stitch Fix to purchase clothing recommended based on the purchase history and personal style/preferences. It takes the form of a typical online/mobile commerce that allows you to buy — without going through a Fix sent by a stylist. ‘Direct Buy’ has continued to grow stably through the corona pandemic crisis and has established itself as a major business strategy from the leadership’s perspective.
The ‘Freestyle’ service can be used even if you are not a subscriber to the existing Fix service, and of course, there is no styling fee, no charges for returns or exchanges, and no minimum order requirement.
The ‘Freestyle’ service provides a variety of features to shop Stitch Fix’s products:
- ‘Trending for You’ : Personalized style recommendations based on the latest trends
- ‘Complete Your Looks’ : Apparel recommendations based on what customers have purchased from Stitch Fix
- ‘Featured Brands’ : A collection of customers’ preferred brands
- ‘Categories’ : Customized product selection according to customer’s size
- ‘Fresh Finds’ : Customized personal shop organized by trends and season
- ‘Expert Picks’ (coming soon): A collection curated by experts, from the unique styles of influencers, celebrities, stylists, or other customers
During the period of this new business model adoption from 2020 to 2021, Katrina Lake, the founder & CEO of Stitch Fix, stepped down from the CEO position to become the executive chair. Elizabeth Spaulding, a former partner from the consulting firm Bain & Co., was newly appointed as CEO.
It is noteworthy that the corporate color of Stitch Fix — which emphasizes the importance of the stylist as to the core element of the personalized style choice and communication — is fading out while ‘algorithm-centered operation’ gets more and more focus under the leadership of Elizabeth Spaulding.
Face-off : Algorithm vs. Human
In the process of Stitch Fix finding a new growth engine and changing its business model, the focus on ‘algorithm and stylist collaboration’ blurred, and the organization is shifting to ‘algorithm-centered operation’. As a result, the human stylists in Stitch Fix became the subject of restructuring.
In 2020, about 1,400 stylists in California were laid off for ‘cost savings’, and instead Stitch Fix hired workers in Dallas, Minneapolis, Austin, and Texas where labor costs were relatively cheap. In August 2021 after the launch of ‘Freestyle’, Elizabeth Spaulding, the new CEO, communicated a policy stating ‘Stylists who cannot work at least 20 hours per week (between 8 am and 8 pm, local time) will not be able to work for Stitch Fix going forward. Those who cannot accept this term within two weeks will have the option to resign with 1,000 USD voluntary exit payment.’ You can read much more details here.
This announcement made headlines soon after and about a third of the stylists quit. And more are preparing to, according to the news.
‘Pressure for cost-saving and revenue growth to meet market’s expectations’ definitely is the main driver of this incident, but the confrontation between ‘algorithm’ and ‘human’ works as the medium.
From the company’s point of view, the scale issue in ‘combining algorithm and human stylist’ must be solved. ‘Freestyle’ service can be viewed as a way to distribute the role of ‘stylists’ to ‘customers’ and ‘algorithm’, to make it possible to achieve scale further. Maybe Stitch Fix might think the ‘algorithm-centric operation’ for both Fix and Freestyle is possible with the data and technologies polished over the past 10 years.
On the other hand, from the stylist’s point of view, it is clear that the company gives more weight to the role of ‘algorithms’ and gives power to the Freestyle service. And what’s more important, the role of stylists gets reduced from ‘giving styling advice to and building relationships with customers’ to ‘training styling algorithms’ — which seems to replace their jobs.
Many former Stitch Fix stylists are saying, ‘even for Fix service, we were already given the guidance to work mostly on writing notes for customers rather than making choices for style recommended by algorithms”. Elizabeth also highlights the stylist’s role as ‘training machine learning algorithms’ in several interviews.
Future of Stitch Fix: Phase of Meaningful Experimentations
Neither Elizabeth nor Stitch Fix’s executives would consider all of the current structures to be final. In addition to the role to train the algorithm used in Fix and Freestyle, various attempts are expected to be made to see if there is room for stylists to perform higher-value, differentiated tasks in the future. An upcoming ‘Expert Picks’ feature within Freestyle would be one of them from my viewpoint.
The mode and tools with which individual stylists execute and manage interactions with both customers and algorithms will have to change given the current business structure of the Fix service, driving a higher level of scale and efficiency. It would be interesting to see how ‘Expert Picks’ will impact the role of stylists and eventually help boost the business performance.
The new shift, if successful, will change the stylist’s working environment, requirements, and money-making structure, so proper education and retraining to help stylists equip with new mindsets and capabilities are critical.
The next two years will be very important. ‘Freestyle’ will expand Stitch Fix’s addressable market beyond that of ‘Fix’ service, but that market is another one with (probably) even stronger competition. How, if possible, can Stitch Fix provide differentiated services from other bigger fashion commerce companies with the data and algorithms polished over the past 10 years? Personally, it is not only about continuously training the algorithm by collecting and verifying customer responses in both Fix and Freestyle, but also how to create a virtuous cycle structure that scales the interaction between customers and stylists. Isn’t it the identity of Stitch Fix, ‘a personalized style service with the human touch’.
The market seems to show mixed reactions to Stitch Fix’s response for growth. Although Freestyle is still in its infancy, it is still disappointing that Stitch Fix acquired only 58,000 new active clients in the past 3 months. Some analysts comment they don’t expect dramatic changes for the rest of FY2022.
Stitch Fix’s forecast to grow YoY sales more than 15%, while it’s not a bad number at all, is not very impressive compared to the 20% YoY sales growth Stitch Fix has achieved in the past.
I look forward to seeing if Stitch Fix will figure out a new recipe of ‘algorithm-human collaboration’ with the two pillars of Fix and Freestyle, and successfully drive new momentum after its relatively sluggish performance after IPO.
