Sign up to receive the Vogue Business newsletter for the latest luxury news and insights, plus exclusive membership discounts.
Personal styling service Stitch Fix has finessed how to use artificial intelligence to understand personal style. Now, it’s building on that technology with a new tool that helps to “predict trends” and inform its inventory decisions. During the recent SXSW conference in Austin, Texas, Stitch Fix executives Loretta Choy, chief merchandising and client services officer, and director of merchant algorithms Sophie Searcy spoke with Vogue Business senior innovation editor Maghan McDowell about how the company blends creative work with data science to predict trends and spoke for the first time about this new technology.
Choy began by outlining how Stitch Fix’s business model aims to solve the paradox of choice in e-commerce. “Clients today are so overwhelmed by choice. When you’re looking online, you’re shopping through probably 52 or more pages until you find potentially one or two items that you like — and then you start site-hopping or looking on social media, where you’re just scrolling, scrolling, scrolling. It feels endless,” she said. Stitch Fix’s algorithms curate the options that are served to personal stylists from its hundreds of thousands of products; stylists then go on to use a client’s purchase history, personal feedback and preferences to manually narrow down recommendations even more. It also uses natural language processing to summarise data that is then integrated into the most recent client request note for the stylist. Stitch Fix can get a further read on client preferences and larger trends through a gamified tool called “Style Shuffle”, in which clients rate potential items and outfits.
Personalisation algorithms don’t remove the opportunity to respond to external cultural micro-trends, Choy added. Instead, stylists can help mitigate the decision fatigue that comes with these short-lived trends — think “cottagecore” or “mob wife” — by filtering the trends through the lens of a client’s own preferences and requests. In advance of the Barbie movie, for example, Stitch Fix saw a 300 per cent increase in requests for “Barbiecore”; it has also seen a 1,000 per cent increase in requests for “quiet luxury”. “It’s just such a dichotomy of people wanting to participate in all of these activities, but at the same time they’re like, ‘My brain is so tired right now. How do I make sense of this?’” Choy said.
Because of its atypical business model, which heavily relies on a “feedback loop” from clients, Stitch Fix can better track what Searcy calls “incrementality”. In other words, they can measure if sales of a top-selling product are taking the place of another potential product and selling to existing loyal clients (which would ultimately be a net neutral in terms of sales) or if a top-selling product is actually filling a new need or attracting a new customer (which would mean ultimately increasing sales). The goal, of course, is to increase incremental sales.
This is an advantage over traditional retailers, Searcy said. “When you’re looking at performance and something’s been flying off the shelves in a conventional retail environment, you don’t get insight into if that’s doing well with your core clients, or if this is actually doing well with a new client segment. It’s the difference between something that is a ‘nice to have’, versus something that’s truly incrementally positive.”
Stitch Fix has built on this knowledge by sharing this information with partner brands via a portal and informing its private label business, which it plans to expand. It can share insights into details such as style, fabric, price point and fit, down to nuances such as areas in which the item might be too tight or too loose. It worked with Pistola, which specialises in denim, to launch two exclusive brands and has done this with many others as well. “This allows them to not only grow with us but also to grow their own brands,” Choy said.
Simulating the future to inform inventory decisions
After refining its recommendation system, which is used by stylists and clients, Stitch Fix created a recommendation engine that its merchants can use to help inform inventory decisions. This includes insights into which products should be repurchased and opportunities in the assortment that can be filled by private-label pieces or pieces codesigned with partner brands. Choy, who oversees the inventory and styles to which Stitch Fix stylists have access, and Searcy, who is responsible for the personalisation algorithms that get the right styles to the right people at the right time, worked together closely on this latest technology.
To build it, Searcy’s team essentially reverse-engineered the recommendation system to create a simulator to predict what would be surfaced to each client in the future. Then, the merchant team can get a sense of future demand, even up to 12 months in advance. It is able to run through billions of possible scenarios in a way that wouldn’t be humanly possible.
This is especially helpful for items that are less obvious to the merch team; one graphic tee, for example, only performed moderately well, and the merchant team wasn’t inclined to re-buy that item. The simulator, however, knew better and identified that for some clients, this item filled a need that no other item in the inventory could provide.
Naturally, the ideal mix involves human guidance as well, Choy said; machines can’t predict changes such as pandemics or the influence of movie releases. In fact, Searcy says that some of the most valuable learnings come after the human merchant team is able to identify something that the simulator misses.
“A big focus is reducing the amount of time it takes for us to go from idea to confidently deciding on an investment in a style, to make that time as short as possible,” Searcy said. Already, Stitch Fix is looking at testing new items via Style Shuffle, meaning soliciting ratings on items that aren’t yet in its inventory. Ultimately, Choy added, this helps decrease waste while improving sell-through. “We want to make sure that we’re driving the greatest productivity in everything that we buy,” she said.
Since developing the prediction tool, the company has seen strong performance for items that are recommended by the tool. “When the tool is used to recommend styles, we’ve seen early indications that our clients love them and our stylists are picking them out. Those are really exciting signals that the items we’re buying are succeeding,” Searcy said.
Comments, questions or feedback? Email us at feedback@voguebusiness.com.
