4 Comments

Summary:

Netflix CEO thought he could do a better job at developing a recommendation algorithm than his engineers. He failed – and the episode shaped the way the company has looked at data ever since.

Reed Hastings of Netflix at NewTeeVee Live 2010 in San Francisco

Netflix may owe its love of big data to CEO Reed Hastings’ 2005 winter vacation: Businessweek took a detailed look at Netflix’s history and the strategic tech decisions the company has made over the past few years this week. One of the previously-unreported tidbits was related to the Netflix Prize, which the company used to encourage researchers to develop a better recommendation algorithm a few years ago. Turns out that was a direct response to the way Hastings spent his holidays eight years ago.

Apparently, Hastings disagreed with his engineers about the best way to serve up recommendations. He believed that Netflix could just recommend new DVDs based on the star rating people gave movies. As in: Want to watch a new movie? Then check out these titles that others with similar interests have rated highly. His staff disagreed, and wanted to look at a whole range of other indicators, including the things people searched for on Netflix’s website.

From the story:

“Hastings spent two weeks over his Christmas vacation pounding away on an Excel spreadsheet with millions of customer ratings to build an algorithm that could beat the prediction system designed by his engineers. He failed.”

Of course, that lesson – more data is better – has been a key part of Netflix’s streaming business. The company is tracking all kinds of usage behavior, including every time a subscriber pauses or skips a movie, the order in which titles are consumed and more.

Businessweek reporter Ashlee Vance goes on to say that Hastings’ failed holiday hackathon led to the creation of the Netflix Prize. This $1 million competition pitted teams of researchers against each other with the goal of improving the Netflix algorithm by at least 10 percent. The irony is that Netflix never actually used the winning algorithm, because it had shifted most of its efforts towards streaming by the time the contest finally concluded.

You’re subscribed! If you like, you can update your settings

  1. Interesting. Thing is, even before I deleted every last one of my movie ratings a couple of months ago, Netflix’s recommendations for me were comically bad. Often I could see the logic–or rather illogic–behind them, and it was silly. For me, liking a movie doesn’t translate into liking other movies that share certain “tags” you can attach to them, whether based on the people associated with it, the genre, or anything else. I guess that logic must work well enough in the aggregate, though, since it is now being used to engineer new “creative” products I will personally never watch. (Loved the original House of Cards to death. The trailer on the new one was enough to make me remove that version from my queue.)

    I can’t do anything but other forms of tracking, but I find all this data collecting disagreeable enough that I decided I didn’t need to be volunteering more data, especially when doing so resulted in no value to me at all. Took me a long time to get rid of all my ratings, but I did. One less (useless) profile of me out there.

  2. netscape search interface both in wii or pc – really sucks!!! In WII the ONLY criteria to use for search is movie name. You cannot search by any other attribute. For all the talks about brilliant engineers workin in netflix I cannot understand why their search engine is so bad.

    1. – netscape search interface both in wii or pc – really sucks!!!

      um, did you mean Netflix?

  3. The claim that the winning algorithms was not relevant for the online stream makes no senses. I think that Netflix didn’t use the winning algorithm because it didn’t contribute to the overall experience, which is needed for higher consumption/satisfaction.

Comments have been disabled for this post