The sudden influx of cold-shoulder tops — blouses with cutouts that expose part, or all, of the shoulder — found on racks everywhere from Macy’s to Zara, can be chalked up to the same pattern found in nature that gave elephants their trunks and giraffes their long necks.
At least, that’s how Stitch Fix chief algorithms officer Eric Colson explains it.
“Nature has a selection criteria: It’s called survival. Circumstances that lead to survival, like a giraffe’s long neck, are going to pass on a certain gene,” said Colson. “We notice the same thing occurring in our inventory — any trend emerges this way. It starts spontaneously, like a mutation. The first time you see a cold-shoulder top, it’s weird and different. But if it’s successful, you see more and more styles as people buy more.”
Putting the life cycle of fashion trends in line with Darwinism seems hyperbolic. But Stitch Fix, drawing from the comparison, identified an opportunity. The company gets a massive amount of customer data from a variety of sources: the initial customer survey, taken upon sign-up, that touches on style preference and sizing; feedback on boxes; discussions with stylists; and social media, as customers can link their Pinterest feed to their Stitch Fix account. With every style box it ships out, its intel grows.
The team realized Stitch Fix could apply a survival-of-the-fittest strategy to its inventory in order to create variations of blouses, jackets and pants that predict and fulfill trends as customers are demanding them. The result: a line of clothing exclusive for Stitch Fix customers, called Hybrid Design, produced in-house using an algorithm that parses through customer data to come up with new combinations of clothing traits that previously didn’t exist in Stitch Fix’s inventory.
“We have a similar selection criteria as Mother Nature, except it’s not survival. It’s customer feedback,” said Colson.
“Trillions of potential options”
Stitch Fix, which doesn’t disclose revenue or its membership count, has raised $42 million from investors and, according to Forbes, brought in a projected $250 million in 2015. The company, founded by Katrina Lake in 2011, employs 5,700 nationwide, including a 75-person data science team and more than 3,450 stylists who work with customers on their boxes, or as they call them, “fixes.” Not a subscription service, Stitch Fix considers itself a styling service, as boxes are shipped on-demand rather than on a monthly basis.
Stitch Fix’s clear advantage over traditional retailers is the transparency it has with customer behavior through every part of the purchase path. Macy’s, for instance, only knows what a customer ends up purchasing and whether or not they end up returning it. Some retailers are taking it a step further, using RFID technology to track data around what’s going into the fitting room and not being purchased. But Stitch Fix can identify what a customer is in the market for, what they purchase, what they don’t keep and why.
“Stitch Fix is taking in so much information very quickly about how its customers are responding to what they put in front of them, and then they’re changing their strategy just as quickly,” said Forrester analyst Tina Moffett. “They have response data, customer sizes and style preferences, and based on that intersection, the personalized offers are basically endless. That’s extremely powerful.”
Today, Stitch Fix has more than 400 brand partners that fill its monthly inventory. It uses machine learning and historical data to determine how much of a single product to order, where to sort new product fulfillment and when to donate old styles. As customers continue to receive monthly shipments, every item they decide to keep or send back presents a new data set.
Based on that data, Colson saw that Stitch Fix’s algorithms could parse the data and determine, by a set of traits, what customers wanted in a blouse or a pair of pants. The problem: It doesn’t always have a matching item in its inventory. There was whitespace in its product offerings that the algorithms identified. Colson said that instead of seeking out more brands to bring on as partners, he wanted to take out the guesswork.
“For every one blouse, there are six or seven neckline options, three or four sleeve lengths, then tons of potential patterns and colors,” said Colson. “If you multiply those together, there are literally trillions of potential options to be made from those combinations. Only a tiny fraction of those combinations have ever existed. Who’s to say there isn’t something out there that would be a total hit with our audience that just doesn’t exist yet?”
In June, Stitch Fix launched its first exclusive brand, produced in-house by its Hybrid Design team. The company uses a set of models and algorithms inspired by the same genetic modeling seen in natural selection to design pieces using new combinations of patterns, colors and styles. For instance, an exclusive blouse for Stitch Fix with polka dots and a high neck is made from two “parent blouses.” Its team of designers gives the final approval. They don’t all work; some designs have been “wonky,” according to Colson. Right now, about 17 items designed by an algorithm are live in Stitch Fix’s inventory database.
Two “parent” blouses, and the resulting Stitch Fix design, below
“We’re uniquely suited to do this,” Colson said. “This didn’t exist before because the necessary data didn’t exist. A Nordstrom doesn’t have this type of data because people try things on in the fitting room, and you don’t know what they didn’t buy or why. We have this access to great data and we can do a lot with it.”
Building customer relationships on data sets
Before Hybrid Design — what Stitch Fix calls its in-house design team that’s part machine, part human— Stitch Fix used a series of other algorithms to support its entire business chain.
An algorithm sorts through the information customers enter in a survey when they sign up for Stitch Fix, using natural language processing to parse through entries like “I dress casually for work,” and “I don’t like to show off my arms.” That intel, along with measurements and monthly reviews, determines what items go into customers’ boxes. A different algorithm assigns a customer’s monthly box fulfillments to a specific warehouse and also assigns a Stitch Fix stylist to a customer, based on the stylist’s expertise and the customer’s needs and preferences.
Fellow retail startups Dia & Co and True & Co have use similar customer data sets in order to inform their designs for customers in the plus-size and lingerie markets, respectively.
“Companies like Stitch Fix and True & Co are raising awareness on how useful data can be in retail,” said Nicole Ferry, partner and executive director of strategy at brand engagement firm Sullivan. “Creating their own designs based on what they see happening in their inventory, identifying the gaps they see, that’s a huge advantage over a department store or a Banana Republic, who have no idea why someone might try on and not buy.”
“We’re not going to push our own brands”
Stitch Fix doesn’t broadcast its in-house brands much with customers, other than to flag that certain pieces are exclusive to Stitch Fix.
“One of our guiding principles at Stitch Fix is that algorithms and expert humans are better together,” said Daragh Sibley, manager of data science at Stitch Fix. “Our algorithms use statistical methods to extract insights from large volumes of data, which are provided as recommendations to expert humans.”
Sibley added that the machine is meant to augment, rather than replace, the human designer, by drilling customer feedback into specific tweaks, like a too-long sleeve, or an awkward, billowy fit.
Ultimately, Colson sees the Hybrid Design line remaining a small, supplementary portion of Stitch Fix’s core business.
“We don’t focus on these designs as Stitch Fix exclusives, which goes back to customer relevancy as our main goal,” said Colson. “We’re not going to push our own brands on people. If we can increase the assortment by filling in the gaps, it’s going to remain a pretty small subset of our overall inventory, while letting us move faster and smarter.”