Just two weeks after launch, in measuring its success, Hunch has determined that its decision engine has had, well, measured success. According to a blog post by co-founder Chris Dixon, Hunch is doing "pretty well, although we still have a ways to go."
Hunch answers users' questions by asking the user more questions. For instance, you can ask anything from "what author should I read?" to "what is my bedroom personality?" Hunch then asks you a short series of questions, comparing your responses to others' and narrowing the possible field of outcomes until it comes up with a series of prioritized answers or recommendations.
Hunch has created its own metric, which measures the number of times users click "Yes" to one of the top three results without clicking "No" to any of the others. When the site launched in beta with 500 topics, the success rate was about 70%. Since going public, the site is up to 3,500 topics with a success rate of 81%, not too shabby for two weeks' work. In those two weeks, the site has seen 1.6 million topic plays and received 1.3 million feedbacks, enough data that Dixon feels confident in the numbers.
Hunch also allows users to create a profile on which they can answer questions about nothing in particular, allowing the site to create a profile of preferences and tastes, helping the engine hone its recommendations further. So far, users have answered 20 million of these questions and offered 3.8 million feedbacks on the 63,000 results currently in the system. As Dixon points out, that's about 62 feedbacks per result, indicating a good deal of user engagement with the site.
Hunch also launched a new prioritizing feature that allows you to indicate a preference when you can't have it both ways. For instance, when asked "What's the best watch for me?" Hunch counters by asking what price you're willing to pay and if you want a watch that's simply functional, sporty, or elegant. If you choose "under $40" and "elegant," Hunch can't reconcile the two, so it asks which is more important to you, elegance or price?
Keeping in mind that the more results it generates, the more precise the engine becomes, the numbers suggest Hunch is off to a strong start. The goal, Dixon says, is a 95% or better success rate. We have a strong feeling Hunch just might get there.