How to improve the Netflix algorithm? A competition

$1M for a recommendation they did not use

In 2006, Netflix offered a $1 million prize to anyone who could improve the algorithms used to recommend movies to customers. Essentially, the solution had to beat the in-house software by 10%. It was a close battle among more than 1000 teams, but a seven-person team called BellKor’s Pragmatic Chaos (yes, a crazy geeky name) won the prize in 2009.

Seems odd, doesn’t it?

Why pay $1M for a recommendation you don’t plan to use. After all, Netflix’ net income was only $115M in 2009. Netflix cited a few reasons on their tech blog:

Too much work to implement the changes
The product had changed (Netflix started streaming video in 2007)
New customer behavior data (e.g., how long people watch shows)
New modes of distribution (e.g., Roku and Xbox in 2008)
International expansion in 2011 (Now in 47 countries)

However, Netflix did get something out of it

Publicity. Lots of publicity. The Netflix Prize received a lot of positive press from the New York Times, Bloomberg Business Week, and other news media. Netflix gained tens of millions of impressions over the 3 years of the contest. In fact, they use the tech blog as a means of word-of-mouth recruitment.

Consultants can all relate

Consultants reading this blog can empathize with this situation. We have all been on projects where we did excellent work, but the client did not act on the recommendation for one reason or another. There are so many possible reasons:

The timing was not right
There was not enough budget to invest
The client only wanted validation of their hunch (they never planned to act on it)
There are politics you are not aware of
The client was scared to make a change
The recommendation was too complex to implement

Sometimes, the client will make the change. 3 years later

I was on a project for a Fortune 500 client in the middle of winter. We put together a pricing recommendation for the client that was well-researched and thoughtfully accounted for the company’s organization and culture. The client did not implement the recommendation immediately. In fact, they did not implement it until 3 years later.