Zalando Uses Machine Learning to Take the Guesswork Out of Shopping
While fashion is sometimes seen as rarified and impenetrable, the reality is that most people engage with it on a daily basis. After all, everyone has to put on an outfit each morning—and those clothes have to be chosen and purchased somewhere. In Europe, that place is often Zalando. Founded by business school friends Robert Gentz and David Schneider in 2008, the publicly traded company (it IPOed in 2014) has 26 million customers in 17 countries, making it Europe’s biggest online-only store for clothing and shoes (and, more recently, beauty products). The site offers 300,000 items from around 2,000 brands, along with free shipping, same-day delivery, and 100 days to return unwanted purchases. Like many online retailers of its size and user-friendliness, anticipating customer behavior—what they will want to buy, and when and where and how they will want to buy it—is crucial to Zalando’s success.
Viacheslav Inozemtsev is a data engineer working on Zalando’s data lake, a repository for all of the information the company has gleaned from its many, many interactions with shoppers. “We are building the data infrastructure to allow all the teams to have easy access to this data, to be able to start and reach some production state quickly,” says Inozemtsev. The data lake’s purpose is to “enable all the teams, from customer-facing applications down to business intelligence and logistics inside of the company, to be machine learning-driven, data-driven, and now also AI-driven.”
Zalando is applying machine learning to more straightforward issues, such as fraud detection and predicting seasonal demand. It’s also using algorithms to make more subtle tweaks aimed at improving the shopping experience, such as providing accurate package delivery times and using past orders to help customers select the right clothing size. “You don’t see a huge impact in any particular aspect when you are dealing with Zalando,” says Inozemtsev. “Basically, with every experience that you have with Zalando as a customer, you get small improvements every time. And it accumulates.”
The company’s data scientists have also ventured into the notoriously difficult-to-quantify sphere of taste. “There are models currently at Zalando that basically understand fashion. They understand style, they understand silhouettes of various people, they understand how patterns fit together, and they can recognize fashion in pictures everywhere,” says Inozemtsev.
“There are several recommendation engines running every time you visit our website,” says Inozemtsev. “We can definitely recommend whole collections to you now, which means that we can recommend an entire [outfit] that you could wear on an occasion.” Eventually, Zalando might be able to direct a user to what they’ll want to wear next month—or even next year.
While Zalando is already Europe’s leading shopping destination, its ultimate goal is to become the continent’s fashion platform. To that end, Zalando has collaborated with a number of startups in the fashion space, offering them infrastructure and millions of new customers in exchange for tools and services that will enhance the user experience. Zalando is also partnering with brands interested in selling products through the company’s store while maintaining control over pricing and offerings. Meanwhile, Zalando has begun moving beyond fashion with its zIMPACT program, which offers guidance and financial support to startups that use technology to increase supply chain transparency.