What if there were a website where you could ask for recommendations on almost any topic—where you should go for dinner, what camera you should buy, what philosopher you might be interested in reading— and get answers customized to your exact taste?
Such was the vision of the four cofounders of Hunch.com: Matt Gattis, Tom Pinckney, and serial entrepreneurs Chris Dixon (founder of SiteAdvisor) and Caterina Fake (cofounder of Flickr).
Supported by a team of algorithm-obsessed MIT grads and approximately $19 million from heavy-hitting VCs, they built the site and it debuted in June 2009. Hunch now claims the ability to predict the average person’s preferences in a wide range of areas—political orientation, cars, food and much more— with 85 percent accuracy.
How it works: users go to the site and answer, at minimum, 20 questions that help develop a taste profile (500,000 people have signed up so far, with the average user answering 120 questions). Hunch then makes recommendations based on the user’s profile, plus the preferences of people with similar profiles. If the user signs up through his or her Facebook page, Hunch has access to all of his or her friends’ “likes” as well, and figures those into its algorithms to create its growing Hunch Taste Graph. (Counting those Facebook connections, Hunch claims that nearly 400 million people are represented in its Taste Graph.)
Hunch was originally envisioned as a destination site, and to an extent it still is. But for now the company’s road to profits involves partnering with e-retailers and sharing revenue from Hunch-recommended purchases.
Of course, there’s no shortage of companies that offer personalized suggestions to shoppers. Using algorithms and your past buying history, Amazon recommends books, iTunes and Pandora recommend music, Netflix recommends movies, and so on.
The hottest trend, though, is social search, or recommendations informed by people to whom you are digitally linked in some way. This fall Microsoft and Facebook announced they were teaming up to allow users of the Bing search engine access to relevant “likes” selected by their Facebook friends. (For example, a user can look up the film Iron Man and get not only information on the film, but also see which of his or her Facebook friends liked it.) Google promptly promised enhancements to its own social search efforts.
Hunch sees itself as standing apart from larger recommendation services because of its superior predictive ability. “By going across multiple domains—from movies to books to restaurants—we believe we can be smarter in each one of those domains,” says CEO Chris Dixon. If a site uses only a person’s movie preferences to predict what movies he or she will like, some good recommendations will inevitably be overlooked. “By sort of radically spanning the input set so that we look at your political orientation, what car you drive, what kind of salad you like, what TV shows you watch—all those things together can actually let us predict what kind of movie you’ll like,” Dixon says.
“For us, it’s all about correlating the affinity between the person and the thing, and that thing could be a TV, it could be a venue, it could be a restaurant, it could be a travel destination.” Some of the correlations Hunch has discovered so far include the findings that people who like Twitter are likely to be interested in visiting the Museum of Modern Art; that people who believe in UFOs prefer Pepsi to Coke; and that people who swat flies rather than shoo them are likely to read USA Today. Dixon also points out that only the largest e-commerce companies have the ability to build an internal team for “machine learning” (computer algorithms that improve 18- automatically through experience). “As big as Amazon is, it only accounts for seven percent of online e-commerce,” he says.