Ever wondered where to grab lunch? Or which DVD set to buy? Or even what to wear on–gasp!–a first date? “Just ask Hunch,” says Flickr cofounder Caterina Fake, whose latest startup helps users make decisions by comparing what it knows about them to what it knows about others like them. The year-old site, which has attracted more than 1.5 million users and roughly $12 million in funding, assembles each user’s “taste profile” by asking him or her a series of questions. FastCompany.com gave Fake the same treatment.
After Yahoo acquired Flickr in 2005, you spent roughly four years working there. Why did you leave to found Hunch?
It’s not like I woke up one morning, got into the tub, and shouted “Eureka!” While I was at Yahoo, there was an absolutely laser-like focus on beating Google at search. But my background is in social software, social networks, and user-generated content. So I started working on social search, which I thought at the time was the most important, interesting, and significant piece of what Yahoo was focused on. That got me thinking about a site that could learn about an individual, and as a result be able to make great recommendations of things that person might like. Eventually, I left to cofound Hunch [alongside Hugo Lio].
How does the site work?
First, we ask people a lot of questions about themselves–demographics, political views, aesthetics, personality, all those kinds of things. And we try to make them fun and engaging.
No wonder I was asked about alien abductions.
[laughs] Right. We don’t have any psychological biases, and we haven’t hired any social scientists or anything. All of the questions are user-generated.
What happens next?
Then, we look at all the data for correlations. And some are quite funny and surprising. Turns out that people who broke their legs as kids are much more likely to like Madden football games than those who didn’t. Entrepreneurs are significantly more likely than non-entrepreneurs to have used a fake ID when they were underage. And people who wear cufflinks several times a month are much more likely to be thrown out of bars for rowdy behavior.
How do you take stuff like that and use it to recommend, say, a hotel?
Once we’ve developed a taste profile for you, our algorithms can start extrapolating. Obviously, the cufflink-wearer who’s getting thrown out of a bar for rowdy behavior is someone who likes to party. So we can infer that they’d be more likely to want a hotel that’s downtown and has an active nightlife. The algorithms work for all sorts of things: which college you should attend, which cookbook you should buy, where you should retire. And it’s all coming from the correlations. We don’t impose any of our own opinions.
That sounds similar to the social graph, which defines people through their network of online contacts.
It is, but our ultimate vision is to create a taste graph, which we believe is more significant. My mother may be the person with whom I have the closest correspondence–so she’s closest to me on the social graph–but the shoes we’re going to buy or the hotels we’re going to stay in are not necessarily commensurate.
Tell me the last thing you Hunched–assuming that’s the right verb.
[laughs] Yeah, it works. Honestly, I just got back from lunch with a friend in the neighborhood, and I used Hunch to find a restaurant.
And how’d it work out? Did you like the food?
We actually ended up at a restaurant adjacent to the restaurant. After we met, my friend said, “You know, I’m more in the mood for…”
No, seriously, I use Hunch all the time. The other day, I was looking for a laptop bag, and bought the one recommended by Hunch. And I promise you, it worked out great.