Stitch Fix, which 10 years ago launched as a data-enabled styling service selling clothes by the box, touts its AI to investors, but doesn't seem eager to speak of it to customers.
About three weeks ago, the apparel e-retailer told its stylists to refrain from invoking its technology when dealing with customers who say their preferences are being ignored, according to internal communications obtained by Retail Dive.
“Now, not only do they not want us to mention an algorithm, they want us to own the choices it makes,” one stylist, speaking on condition of anonymity, said by email.
In an internal message sent to stylists in mid-April, the company said in part: “Data science and Stylists have always worked hand-in-hand… When that partnership doesn’t result in the best client experience, Stylists should take ownership of the disappointment, no matter the role the data played.”
Additionally, stylists were instructed not to use internal terms “unknown to clients such as ‘AGP’ or ‘algorithm’ or explain our algorithmic technology in their notes to a client.”
“Data science and Stylists have always worked hand-in-hand… When that partnership doesn’t result in the best client experience, Stylists should take ownership of the disappointment, no matter the role the data played.”
Internal memo to stylists
According to Stitch Fix stylists interviewed for this story, the company previously did instruct them to let displeased customers know that its algorithm is always learning, and doesn’t always get it right.
As with the text on its customer-facing retail site — where it says it employs "real people who really get your style," referring to the nearly 4,000 stylists who curate its boxes — the new directive stands in contrast to how the company emphasizes data science in investor materials like its annual report, comments to analysts during earnings calls and communications with business reporters.
In an emailed statement for this story, Stitch Fix said that its “combination of human touch and data science is unique in the industry; we were the first retailer to hire a team of world-class data scientists to enable personal styling at scale and today we are still the only retailer that employs 1000s of style experts to build and maintain trusted, one to one relationships with clients and help us deliver a completely differentiated, entirely personalized shopping experience.”
Wall Street vs Main Street
Given Wall Street’s affection for tech stocks, it stands to reason that Stitch Fix would promote itself to investors as a data-rich disruptor of retail.
“Wall Street looks favorably on firms with a technology play, and tends to cut them more slack than it does traditional retailers,” GlobalData Managing Director Neil Saunders said by email. “In some cases part of the rationale is that technology gives companies a strategic advantage, and also that it could be resold or licensed to other firms creating an additional stream of revenue. For some, like Ocado or Amazon, this plays out and it is quite legitimate to see those players as technology-first retailers.”
The apparel box business, practiced by Stitch Fix and a few other companies, is predicated on a marriage between data and the human touch. The company gathers information like preferences, size and budget via style quizzes, and bolsters that with clues from purchases and returns, but enlists people to ultimately fill the boxes.
Speaking at the Milken Institute Global Conference last week, Stitch Fix CEO Elizabeth Spaulding described the setup much the same way that founder and former chief executive Katrina Lake always did.
“From the very beginning, I think one of the differentiators of our model has been really, really embracing human-in-the-loop machine learning,” she said during a panel on Retail’s New Reality: Technology, Innovation, and Consumer Behavior. “If you think about why the whole styling model works — and every quarter on record, we’ve improved our keep rate, which is how many items people keep in this box of products sight unseen — is where we’re leveraging the power of data science, feedback from our stylists, feedback from our client community."
"I think one of the differentiators of our model has been really, really embracing human-in-the-loop machine learning."
CEO, Stitch Fix
But the model’s track record has been bumpy, leading Stitch Fix to introduce the ability to shop directly from its site, and forgo the boxes, in order to stoke sales. In the months since that move, the box business has suffered more than executives had anticipated.
Stitch Fix’s tech advantage is “less convincing” than at other retailers, according to Saunders.
“While it is creative with technology, it isn’t a pioneer and what it has done with algorithms and AI still has a lot to prove,” he said. “This is likely one of the reasons Stitch Fix is keen to downplay any issues with their technology even though there are clearly some problems.”
Stitch Fix appears to have fallen from grace on Wall Street. At press time, its share price is hovering around $10, down from nearly $100 just over a year ago.
A preview of the problem
The company didn’t address Retail Dive's questions about its new directive to stylists, which is centered around “Fix Preview.”
That option, introduced last year, is an email sent before a “Fix” is shipped that allows subscribers — those who participate in its decade-old box business, as opposed to its much newer e-commerce site — to choose or reject suggestions for items that could go into their box. The added step has contributed to higher retention rates, Spaulding told analysts in March, according to an earnings call transcript from Seeking Alpha.
The preview can be a win for customers as well because there’s a greater chance they’ll keep something from their box. If they do, they can use the $20 styling fee toward the purchase; if they keep all five items they also get 25% off the entire cost.
Fix Preview has also introduced new potential pain points, however, according to stylists who spoke on condition of anonymity to protect their jobs. Stitch Fix’s pitch to customers is curating apparel that fits them well, and meets their preferences. When clients are miffed by the preview suggestions, they tend to assume that their stylists are ignoring their communications and disregarding their preferences or needs, the stylists said. It can be especially irritating when all 10 potential options are accessories like bags or ties, or the same garment in various colors, they said.
It’s the company’s tech that is usually responsible for poor selections like that, according to stylists. Successful tech-generated suggestions also raise other questions, they said.
“The algorithm is often (usually) selecting all of the items for the Preview — this is what stylists aren’t allowed to talk about or blame,” one stylist said by email. “We’re still charging clients a $20 styling fee, but leading them to believe a human stylist is always involved in the process even when that’s not the case. So what’s the styling fee for? Seems really unethical.”
A Stitch Fix spokesperson said that the Fix Preview is generated using various combinations of technology and stylist expertise, depending on the individual client. The company didn’t elaborate on how often the preview is surfaced entirely by the company’s algorithm, entirely by the stylist or some combination.
More than the algorithm
Being more transparent around its AI, rather than less, may actually be more helpful to Stitch Fix, experts say.
Technology in general, at least for now, isn’t great at handling the many mercurial factors around apparel and style, according to Piyush Patel, chief strategic business development officer at AI platform Algolia, which specializes in search and discovery.
“In fashion, I think there’s such a subjective element of what is right,” he said by video conference. “It’s going to take the input and give you the output, that’s all. It doesn’t have a preference, humans have a preference. If I didn’t have my cup of coffee this morning, I might think differently about stripes versus not stripes. Or something I saw, where stripes didn’t look good on somebody else. Maybe I just don’t like stripes, who knows? So many subjective elements go into fashion that I don’t know that you can trust either humans or data.”
Companies like Stitch Fix that present themselves as tech-forward often face elevated expectations from investors and consumers alike, according to Patel. Yet when it comes to making choices in areas like clothing, tech is less nimble than humans, who are aware of the subtle shifts in trends and understand the emotional response to style, he said. One way to manage customer expectations is to pair recommendations with the reasons why they surfaced.
“Put tags under it that say, ‘This was based on the colors you’ve chosen. This is based on the size you’ve chosen,’” he said. “Let the consumer drive the data that’s being used to make recommendations, and show them why. Then it’s not creepy. And then it can be wrong, and humans can accept it.”
“Let the consumer drive the data that’s being used to make recommendations, and show them why. Then it’s not creepy. And then it can be wrong, and humans can accept it.”
Chief Strategic Business Development Officer, Algolia
Steve Zisk, senior product marketing manager at customer data platform Redpoint Global, said it’s understandable that the company doesn't want stylists to point to the algorithm when a customer complains.
“Both customers and brands themselves find it easy to blame the tech,” Zisk said by video conference. “But blaming the tech doesn’t help because the tech is a reflection of somebody’s idea of what was supposed to make things better. The difficulty with that message is, I as a customer don’t really care. What the customer wants to hear is ‘I’m sorry.’”
Still, having stylists “take ownership of the disappointment, no matter the role the data played” is irresponsible, unfair and ultimately will hurt workforce morale, according to Colorado State Business School Professor Jonathan Zhang.
Rather than admonishing stylists to minimize tech’s role, companies may get further with customers by being more up front about it, he and others said.
“The algorithms are not perfect. Algorithms have certain skill sets, like humans have certain skill sets, and both make mistakes,” Zhang said by phone. “I think it’s important to avoid this illusion of technological superiority.”
Even if Stitch Fix could perfect its algorithm, however, it has bigger issues that would remain, including longstanding challenges that all retailers face, Saunders said.
“There are concerns around demand for their service, consumer appetite for apparel in general now that inflation is taking hold, and the underlying economics of their business model,” he said. “Having the best technology in the world doesn’t necessarily resolve those things – it just acts as a fig leaf to hide problems elsewhere.”