“Everything Is a Recommendation”
The next generation of online recommendation engines is less obvious, but more pervasive.
When Barneys New York launched a fashion line by the Oklahoma City Thunder basketball star Russell Westbrook last July, executives didn’t know exactly who would buy those clothes. They didn’t need to. The answers quickly emerged from an online shopping innovation that’s often overlooked: the recommendation engine.
Traditional versions of the technology are simple. Tell Netflix your feelings about a few movies and it suggests more. Read a product page at Amazon and it shows you similar alternatives. These are the tools that helped make those companies huge. But today, new technologies and much bigger arrays of available data are taking recommendation engines like the one Barneys uses to a new place, making them less obvious to the user but more important to website operations.
One example is how recommendations may show up as auto-completing search results. After a shopper at Jenson USA’s online bike shop enters the first two letters of a search for “full face helmet,” the recommendation system displays a list of helmets in an order based on that customer’s profile. At Neiman Marcus, each shopper’s online experience is similarly customized according to the person’s behavior on previous visits—and even in the current one.
Better tagging technologies let retailers dig more deeply into the specific design details of clothes. That’s how they can highlight new designers like Westbrook to appropriate customers, zeroing in on specific features of his designs. It’s similar to the way Pandora groups sound-alikes to build audiences for musicians.
When more sophisticated recommendation engines entice casual browsers with such tailored page selections, the chance they will buy something triples, says Matt Woolsey, executive vice president for digital at Barneys.
“The old way of making recommendations online is about catching up to the customer—you let them tell you about themselves and chase them,” says Woolsey. “We’re trying to use big data to get ahead of the customer.”
The technologies that make online recommendations as well-tailored as a Barneys suit are big-data software like Hadoop and the hardware to run it on, says Joelle Kaufman, chief marketing officer of BloomReach, a startup based in Mountain View, California, that is one of about three dozen vendors in the field.
Location-based data from mobile phones can play an important role, too. Other sources of consumer insight just beginning to inform these new engines’ recommendations can include purchase history from offline stores and social-media history.
A quick run through the Barneys site illustrates how it works. Woolsey and I each went to the menswear page and clicked on the same $150 watch. Since my limited browsing and purchase history focused on less expensive items, I got a list of watches ranging from $95 to $250 as counter-suggestions at the bottom of the screen. Woolsey, who acknowledged cheerfully that he probably dresses better than most reporters, was shown watches costing between $330 and $1,100.
Making this possible are parallel-processing technologies that process massive amounts of data quickly, says BloomReach’s Kaufman. Emerging systems can propose dozens of different algorithms to choose the next page the consumer might see.
At Neiman Marcus, BloomReach’s technology can change what types of clothing appear on the womenswear page after just a few clicks. After Kaufman clicked on three sweaters, a tab for Jimmy Choo shoes disappeared, replaced by a gateway to sweaters on sale.
“That’s instantaneous machine learning,” she says. “Everything is a recommendation.”
Become an MIT Technology Review Insider for in-depth analysis and unparalleled perspective.Subscribe today