Algorithms and humans work in tandem at Thread, a company that could very well be the pioneer for the fashion tools of the future. The 500,000-users-strong clothing recommendation service currently focuses on menswear, but they are working to develop a version for women as well--because thanks to machine learning, they’re growing faster than ever.
Pexels.com
Machine learning is a type of artificial intelligence that allows computers to constantly learn and adapt as they are exposed to new data, without the necessity for each step being explicitly programmed. They do this by detecting patterns in the data and changing their actions based on that information.
Thread is using machine learning to create an ultra-personalised fashion service for their customers. It’s like having your own personal stylist--but it’s not just accessible to the rich and famous, because this personal stylist is free.
CEO Kieran O’Neill, a 28-year-old serial entrepreneur who’s already built and sold two startups, explains, “The way it works is, you do a quiz on the site, we match you with a stylist, they'll briefly review your information and put in some high-level information about what they think you would suit and things to avoid. And then the algorithm will go through the quarter-million products that our partners have and find the best ones for you based on the instructions from the stylist, and email them to you. You look at them and rate them and [the algorithm] learns based on that.”
This team effort--man and machine--is how Thread can serve half a million customers and counting...with an in-house team of only eight stylists and 10 in the tech department. Those numbers are fleshed out a little by the 50 freelance stylists who contribute, but still: Thread is a small company doing big things.
They used to be smaller still, as a YC-backed company “doing things that don’t scale” (a hallmark of the YC experience): handling everything manually, starting with a small user base. “It was totally unscalable and unprofitable for the first six months,” says O’Neill.
Introducing machine learning helped scale the business by decreasing the amount of human labor that went into selections. But even now, the stylists, who come from impressive backgrounds like GQ, Burberry, and celebrity styling, sit down with the head data scientist once or twice a week to critique what kinds of results the algorithm shows.
It’s likely you’ve encountered similar concepts already, whether you’re aware of it or not--from social media curating what’s shown in your feed to the results that display at the top after a search engine query. Applying those processes to fashion, though, presents a unique set of challenges.
“Fashion is so much harder because it's just so much more nuanced,” O’Neill explains. “People have hundreds of preferences about clothes that they don't even really know themselves. You show them a shirt and they'll say, ‘Oh yeah, but I'm not a fan of that shade or the stripes are too wide or I don't like the color,’ and that's just one category of item.” He also notes the tendency of fashion to change all the time, and for preferences to be different based on different locations, both of which complicate matters even further.
To deal with this, the algorithm has to contain many different layers. The question was, according to O’Neill, “How do we actually build the basic form of AI to learn across thousands of users and thousands of items and make it extremely personalized so every user has a different experience?”
It took them several years (and hiring AI juggernaut Ed Snelson) to build the answer, but the underlying AI is now capable of using an overall model to understand a user’s basic behavior, blended with “sub-algorithms” that add a greater amount of specificity.
Once clothing suggestions have been made, users can then choose which items they want to be sent. After experiencing subscription services where clothes are sent automatically, without that element of choice, O’Neill jokes, “One of the truths of the internet is that guys hate returning stuff.”
There’s no doubt in O’Neill’s mind that machine learning is an incredibly powerful tool when it comes to growing a business. “It's been transformative. There's no way we could be at 500,000 users right now if we didn't have this. We would need to have literally thousands of stylists by this point. Instead we have eight. That is huge.”
