Skip the navigation

Who's the Smartest of Them All?

Social software uncovers the true experts.

January 10, 2005 12:00 PM ET

Computerworld - Before building a walkway through a garden, college campus or office park, an experienced landscaper carefully studies where people walk naturally to discover the best routes between popular destinations.
Hewlett-Packard Co. researcher Bernardo Huberman and his team at the Systems Research Center at HP Labs are using a similar strategy to study how e-mail flows through organizations. The idea is to uncover natural "communities of interest" that can be tapped to make smarter decisions and more accurate business predictions.
Using an algorithm that measures "betweeness centrality" -- a measure of the prominence of individuals in a social network -- Huberman and his team classified hundreds of thousands of e-mail messages by how they traveled within certain HP divisions. They discovered that day-to-day work was often accomplished by self-selected teams of people who don't show up as a group on a formal organization chart. They theorized that members of the groups actually made up de facto teams of experts whose business decisions would outperform those of the formal experts.
To prove the theory, Huberman and his collaborators had 15 HP managers distributed around the globe place bets on projected monthly revenue and profit figures for an HP division. The research team developed an algorithm to account for variations in the managers' attitudes toward risk. As an incentive, Huberman also provided the managers with a small amount of cash that would increase or decrease, depending on the accuracy of their predictions.
Accurate Predictions
In the end, the group of managers consistently predicted the financial outcomes more accurately than an expert financial software tool the division had been using to forecast the figures.
Huberman says the test could also be conducted by pitting the informal group against a formal group of decision-makers, and the results would be the same. The reason is that the information used to predict a business outcome is aggregated from the best possible sources, even though their high level of knowledge may not be reflected in their job titles.
He also notes that only nominal incentives are needed to persuade undeclared experts to do their best. "Just putting up a little bit of money -- less than $100 -- makes people behave differently," Huberman says. Moreover, money isn't necessarily a requirement. "People in companies are concerned about their status. If they predict well, call them 'dukes' or 'barons.' There are ways to enhance people's status other than giving them financial compensation," he says.
Brian Whitworth, a researcher and assistant professor of information systems at the New Jersey Institute of Technology in Newark, has written several academic papers



Our Commenting Policies