Stitch Fix is a data-driven clothing company using AI at virtually every aspect of its business, with its own algorithms department staffed by 100 data scientists.
Algorithms help pick out clothes sent to customers in the mail, choose the clothes kept in inventory, assist with client communications, and have even started to design clothes. A computer vision algorithm ingests the Pinterest Pin boards to keep track of things customers found online that they love.
“Remember, this isn’t like Tinder swiping on a phone. We’re sending real things to clients who are standing in front of a mirror in their own house, and that leads to really rich feedback,” Stitch Fix chief algorithms officer Eric Colson said onstage at Transform, an AI-focused VentureBeat event in Mill Valley, California. “It’s that feedback and client data that’s really the game changer, which even I underestimated.”
Becoming an AI company is the approach du jour today, but it’s not for everyone, Colson said.
To decide if you should consider the same, he suggests considering your organization structure. As chief algorithms officer at Stitch Fix, Colson reports directly to CEO Katrina Lake and is at the table at C-level company meetings.
There’s also a separate department just for algorithms — they’re not folded into other teams.
“You find it often buried in marketing or engineering or finance or awkward other places, and we felt like it had to have its own thing, because data science has its own tooling, workflows, ethos, all this stuff that has to be unique to it,” he said. “You don’t want to inherit the skills and work style of a parent organization. That’s what would happen if it was under engineering or marketing is, you get those ethos pushed down on you.”
Companies heavily reliant on branding for their livelihood may also see little value in becoming a purely data-driven company since branding decisions don’t require much data.
“I don’t think Prada needs officer-level representation for data science,” Colson said in response to a question from VentureBeat founder Matt Marshall. “Their big investment is brand. That’s their major asset, and I think they should keep nurturing and fostering that. They also have talented merchants coming up with the latest styles and so forth, and that’s where they’re going to differentiate.”
Companies should also consider whether they have 1) high frequency events that can be measured and 2) unambiguous feedback on how things are going.
Each item of clothing comes with dozens of measurable attributes like the distance between buttons or fit in the shoulders.
For feedback, one of the most valuable sources of data, Colson said, are extensive notes from clients about what they do or don’t like about the box they’ve received in the mail. Humans listening closely to these notes is a way Stitch Fix tries to differentiate itself and compete in an environment that includes Nordstrom-owned Trunk Club and Amazon’s similar service Prime Wardrobe.
Paying close attention to this feedback allows Stitch Fix stylists to be aware of context for special requests, like a recent customer who asked for an outfit for her ex-boyfriend’s wedding.
“This changes the context, right? Our best natural language processing engines aren’t going to get that right, they don’t have the ability to empathize and know what that means, we need humans for this. So that’s one thing — the rich context — you need human stylists for,” he said.
People are also important for final curation of boxes put in the mail, as well as building relationships with customers.
“Those are things we’ll always need to have that are stylist-driven and human,” he said.
Some human signals, however, are taken with a grain of salt. It’s not uncommon for men who are 5’10” to round up their height to six feet tall, and for women to report an age 10 years younger, Colson said.