Netflix found a cheap fix for innovation

Netflix started its business as a DVD rental company a couple of years back. They cleverly combine a net-based ordering system with a well thought-out physical distribution approach for the flicks on DVD. You select your movies from their website and put them in a queue. Every time you send a DVD back, by dropping it in a pre-addressed, pre-paid envelop in your mailbox, you get a new one. No late fees. Great for people like me who want to bring DVDs on a trip or just forget about them. Last year they introduced movies online to watch either on your PC or TV, via a small WiFi connected box. Guess what? They actually have old Dutch favorites like Soldier of Orange and countless other non-mainstream movies.

When you log on the first screen that comes up is “Movies you’ll love”. I have rated around 400 movies so far (nowhere near the guy who rated 5,000 movies in one day). The ratings vary from 1 to 5 and I generously bestowed Ben Hur, Citizen Kane and The Godfather with the maximum number of stars. My ratings are continually compared with people who have similar tastes and by applying some complex algorithm they predict which movies I might be interested in. They show both the overall rating and the one they believe I will give the movies I am considering for watching. Interestingly these two are rarely the same and my ratings are consistently close to their predictions.

A while back Netflix invited anyone with a beautiful mind to participate in a contest with $ 1 million prize money to come up with an algorithm that would improve the accuracy of their current system by more than 10%. In order to do so they made millions of data records available and tested the algorithms submitted against the actual ratings submitted by customers. According to Business Week, the winning team, which includes scientists from AT&T (T) Research, Yahoo's (YHOO) Israel lab, and computer scientists from Austria and Canada, blended more than 700 different statistical models into their formula. The next contest will be for algorithms that predict the popularity of new movies, based on your rental history, demographics and other profile attributes.

Netflix has stared into the future and seen what is happening in the media world. As content proliferates and the lines between user and producer blur, it will get harder and harder to separate the wheat from the chaff. While the studios will continue to generate blockbusters with big stars and maybe even come up with an original plot, instead of a comic book rewrite, we can expect more and more interesting low budget movies for smaller audiences. Movie business is even worse than the fashion industry and the few hits have to make up for multiple misses. But Netflix has something the studios don’t: a thorough understanding of their customers’ tastes and interests down to the individual level. Shortly, they will have a way to predict whether a movie will make it or flop. And they pay only $500K to get there, while the big studios keep shooting in the dark.

Netflix has understood that they don’t need to hire armies of PhD’s in statistical analysis to get what they need. They just “crowd source” their innovation. According to the New York Times, thousands of teams from 186 countries made submissions. The winning group is a merger of different smaller teams, who initially competed against each other. A Survivor-like situation emerged with different individuals and groups trying to form alliances to beat the others.

Google, Amazon and Netflix run highly profitable, multi-billion businesses based on a simple principle: attract as many users as possible and have them interact on your site, gain a detailed understanding of their needs and interests and analyze this against huge quantities of data on their peers to create the best value propositions. There are some lessons here.