If two heads are better than one, are 45,000 heads spectacular? That’s the promise of crowdsourcing. Companies can find new ideas faster and sometimes at a lower cost than internal innovation.
[Related: Crowdsourcing Offers a Tightly Focused Alternative to Outsourcing]
But putting business problems out for public brainwork could expose sensitive information and strategic plans. And contest-winning ideas, developed in isolated laboratory conditions with squeaky-clean data, can’t always be translated to the unpredictable real world, says Anand Rao, a principal at PricewaterhouseCoopers.
Still, some large companies see crowdsourcing as a valuable tool for finding answers to questions that involve analyzing lots of data.
The Crowd Takes Flight
General Electric recently worked with Kaggle, a vendor that manages crowdsourcing contests, to find ways to make flying more efficient. GE customer Alaska Airlines agreed to provide data, mixed with GE’s own statistics, about planes, flights and other aspects of flying.
Once the data was scrubbed of anything proprietary or private, GE shared it with Kaggle members, which include coders, statisticians and professors, and other external developers and data scientists. Contestants analyze the data and devise algorithms to answer business questions.
The contest wasn’t over at press time, but early results showed that the best entries predicted flight arrival times within four minutes of actual arrival–three minutes better than tools now used by air-traffic controllers.
“I do believe we have the potential to see people come up with something groundbreaking for us,” says Bill Ruh, vice president of GE’s global software and analytics center.
Insurance giant Allstate is also using Kaggle contests for big-data analysis, but along the way has learned to disguise sensitive data, says Eric Huls, vice president of product research.
In Good Hands With Crowdsourcing?
In one contest, Allstate sought better ways to predict which auto insurance customers are most likely to submit a bodily injury claim, based on what cars they drive.
Allstate gathered historical data on such claims, including vehicle make, model and horsepower. Rather than reveal sensitive information to outsiders (Kaggle claims 45,000 problem-solving members in 100 countries), Huls’ group masked the data. For example, references to “Ford Focus” became “27” and horsepower was assigned a variable, not a real measure.
For three months, 202 players on 107 teams competed to produce the most accurate algorithms. Allstate split $10,000 in prizes among the three participants whose predictions scored highest for accuracy. “Matt C.,” the winner, came up with algorithms 271 percent more accurate than the ones Allstate had been using. He won $6,000.
Running a contest is “an order of magnitude” less expensive than hiring outside consultants to supplement internal analytics staff, Huls says. He also likes the diversity of approaches taken by the contestants. Allstate has incorporated the contest results into how it evaluates customers, he says.
Kim Nash is managing editor of CIO Magazine. Follow her on Twitter @knash99.
Follow everything from CIO.com on Twitter @CIOonline, Facebook, Google + and LinkedIn.