This shows that most large-scale collaboration fails
Past News ReleasesRSS
Houston, TX (PRWEB) September 21, 2009
With the winner of the first $1 million Netflix Prize (http://www.netflixprize.com) announced and a new contest under way, there are lessons for business about using large-scale collaboration for innovation.
Success through global, large-scale teams doesn't happen by accident or simply by throwing groups of smart people together. It requires vision, leadership, and commitment, according to Greg McAlpin, who played a major role in guiding the team that reached the top of the leader board in the first Netflix Prize.
McAlpin is a software engineer and consultant based in Magnolia, a Texas town near Houston. His team is The Ensemble (http://www.the-ensemble.com).
The intense rivalry of minds attracted more than 51,000 individual contestants grouped into 41,000-plus teams based in 186 countries. Although these tens of thousands of participants put countless hours of thought and calculation into the problem, only The Ensemble and one other team attained the goal after close to three years of effort.
"This shows that most large-scale collaboration fails," McAlpin emphasizes.
Other knowledgeable observers agree. "There is this misconception that you can sprinkle crowd wisdom on something and things will turn out for the best," Thomas W. Malone, director of the Center for Collective Intelligence at Massachusetts Institute of Technology, told The New York Times. "That's not true. It's not magic." (http://tinyurl.com/nclqof)
Online movie rental firm Netflix, Inc., launched the contest in 2006 as a way to focus worldwide talent and ingenuity in the field of data mining on improving by at least 10 percent the accuracy of software that makes viewing recommendations to Netflix subscribers, based on their movie choices.
This June, BellKor's Pragmatic Chaos submitted an improvement of 10.05 percent. According to the contest rules, this set off a 30-day countdown to the close. In a true nail-biter of a finish, The Ensemble at the last second topped the other finalist by 1/100 of a percentage point on the public leader board. Both teams met all of the requirements to qualify for the grand prize.
McAlpin joined the fray in earnest in January of 2009. Drawing on his extensive experience in collaborative systems design and development, he set up a web-based file-sharing platform to enable the group to work together no matter where its members were located. McAlpin's critical collaboration infrastructure included a specially developed module for statistical analysis of data uploaded to the website.
The next month, McAlpin invited other contestants to join him. The four who responded named their team Vandelay Industries ! in a nod to comedian Jerry Seinfeld.
By March, Vandelay Industries ! hit No. 15 on the leader board, despite the fact that none of the members had a background in machine learning. Matrix factorization proved an effective approach to the problem. McAlpin used several variations of this technique to stay in the top 0.02 percent of all contenders. Even while focusing on the technical challenge, he was planning ahead, thinking of ways to improve the team's performance enough to capture the prize.
McAlpin suggested reaching out to the highest-ranked individual on the leader board and offering to work with him. The members of Vandelay Industries ! also wanted to join forces with a competitor called the Grand Prize Team. McAlpin represented Vandelay Industries ! during inter-team discussions.
Trust was a major issue. McAlpin urged his team to bite the bullet and give the Grand Prize Team full access to all Vandelay Industries ! files. That act of disclosure paid off. McAlpin's counterpart on the Grand Prize Team suggested a formal exclusivity agreement to help bring stability to the potential team's membership. McAlpin agreed and the two negotiated the wording. Each then took the accord back to his team and persuaded all members to sign it.
When the two sides merged into The Ensemble, their performance improved dramatically. "Once we got past the fears, it was a good mix," McAlpin says. "I wanted to solve conflicts so we could move forward."
In the end, The Ensemble comprised 30 members from Australia, Canada, China, Greece, Hungary, Israel, Poland, The Netherlands, and the United States. McAlpin also is pleased that the file-sharing website helped ensure that breakthroughs during the contest were noticed.
He knows that the skills and contacts he made during the contest will be invaluable throughout the rest of his career, no matter what his job title.
"In six months, this group has done the impossible," he says. "We met the challenge of making a 10 percent improvement."
gmcalpin at scbglobal dot net
(Bio and photo: http://budurl.com/dfzr)