$1 Million Snowflake Prize

by

When explaining showing example of how The Snowflake Effect is already at work I often use Netflix as an example.  This DVD movie subscription service has been a huge hit since it first began by eliminating the need to make the trip to the video store and by eliminating any chance of late fees.  They did this through an ingenious combination of old and new by using the postal service to mail DVD’s to your home and by having a simple per month subscription fee that entitled you to keep the DVD’s for as long as you liked before mailing them back. 

To get started you went online and created an ordered list or queue of movies you wanted to watch.  Depending on which subscription level you chose you could have 1-3 DVD’s at a time and so to start they mailed you the first 1-3 DVD’s on your list.  You could keep the DVD’s as long as you like and whenever you were finished you sent them back in pre-paid mailers and they would send the next one in your list to you.

Handy to be sure and the service was a huge success from the very beginning.  However the real value turned out to be a little noticed feature at the time which was the feedback loop that they built into the system.  Each time they received one of the DVD’s you sent back they would send you an Email to confirm that they’d received it and tell you they had sent out the next one on your list.  Then they added the real value item, a simple 5 star rating system asking you to indicate how well you liked the movie you had sent back.  Netflix then took this preference data and used it to create an additional  list of Netflix recommended movies. 

Based on talking to many people who used Netflix, it was typical to pay very little attention to this additional list at first but after some time of using the service they would start to have some  difficulty choosing good movies to add to their list and so they would try some of the ones from the Netflix recommended list.  This would continue for a while and then because you were asked to rate each movie after watching it, people would begin to notice that more and more of the movies they really liked were the ones Netflix had recommended.  Netflix had developed was a movie recommender technology they called CinematchSM and most people found that Cinemax was better than they were at choosing movies they’d love!

To their credit, Netflix soon began to realize that their true and lasting value proposition was NOT delivering DVD’s via the mail or even avoiding late fees.  The real values was in helping people resolve the “paradox of choice”, Netflix lists over 100,000 movie titles, and growing, by helping them consistently choose movies that THEY really loved to watch. .  In fact Netflix as recently struck deals with cable TV and other companies to deliver their movies directly and almost instantly to your home via the internet and so the mailing service will likely soon be a thing of the past.  

In looking for ways to improve on their ability to deliver on this value proposition Netflix began to pay more and more attention to Cinemax and then they got REALLY smart and decided to “crowdsource” the next big improvement in Cinemax by creating a contest they called the “Netflix Prize” which offered one million dollars to the first person or team who could improve Netflix recommendations by 10%.  And therein lies the story I’ve been fascinated to follow since it started back in October 2006.  In February this year (208) Wired magazine had an article “This Psychologist Might Outsmart the Math Brains Competing for the Netflix Prize” wrote up a good account of how the competition had taken off with thousands of entries submitted by everyone from large corporations to research departments to single individuals who were from countries all over the world.  One individual, and the feature of the Wired article identified himself simply and quite accurately as it turns out, as “Just a guy in a garage”.

To help the competitors, Netflix did something which has turned out to be a “prize” in itself to the data mining world at large when they posted what is apparently the largest dataset to ever be published, consisting of 100 million of the preference ratings from Netflix customers.   This enables contestants to write their recommender algorithms that are more and more accurate at recommending movies that users will like.  When competitors submit their latest algorithm, Netflix tests it against a different set of ratings data which they keep secret and the post the results of this testing to the Netflix Prize site.  The competition is still running and you can keep up with the progress of the top contenders on the Netflix Prize Leader board.  As of this writing (Nov.14, 2008) they leading entry is at 9.44% and so while the last 1% of improvement is estimated to be more difficult than the first 9%,, the steady progress would seem to indicate that the prize will soon be awarded.

An additional item of note is that the participants have taken a surprisingly open approach to the competition by openly posting details of their methods and many are analyzing these and building upon them for their own models so there is quite a cyclical improvement happening.  Netflix also took what I thought was a very smart and novel approach in that the winning team retains ownership of the solution they come up with and must license it (non-exclusively) to Netflix. And according to the Wired article;

“The company is already incorporating some of BellKor’s ideas into its own system and in the future may buy code from other contestants, as well.”

For me this is great fun to watch not only for this specific contest but also for an intriguing and replicable way to promote innovation and creativity.  This is proof positive of the extremely tangible value there is in amplifying The Snowflake Effect and moving us further along the continuum towards the end goal of “just right”.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: