On the first Tuesday of every month, 10 or so commodity managers from across Hewlett-Packard's hardware divisions dial in for a conference call - but the civility often ends there. For an hour or more, they bicker, squabble, and joust over one seemingly innocuous question: What will the price of DRAM memory chips be in one month, three months, or six? "Usually, it's the loudest, most obnoxious guy who gets heard," says HP research scientist Leslie Fine, who's studied the process.
Jawing about tomorrow's weather for an hour might sound more intriguing. But at HP, the DRAM powwows often turn into shouting matches for a simple reason: After each meeting the managers vote and then put out an official forecast that 70 HP buyers rely on to price more than $50 billion in HP computers and other hardware-often months before the chips that go in them are bought. If the forecasts miss by even a few cents, the difference, which can add up to millions of dollars, comes out of HP's slim profit margin for hardware.
Bernardo Huberman, a senior fellow at HP Labs, believes there's a smarter way to make predictions that affect a company's bottom line - and he and Fine have made guinea pigs of the DRAM squad to prove his point.
These days, after each meeting, the 10 managers and 10 other colleagues from around the world log on to an internal website and enter bets on chip prices. Each "player" gets 100 tickets to place bets on different price ranges. At the end of the quarter, the winning player gets up to $250.
Huberman's experiment is just a few months old, but already his betting market for DRAM prices is batting .750 against the status quo. So far it's beaten the official HP forecast six out of eight times, and tied on the other two.
Removed from the closed-door setting of executive meetings, where personality and ego can skew honest opinion, the new forecasting tool "works better than the best person," Huberman says.
More important, it's lending credence to the notion that online betting and similar types of so-called prediction markets aren't just for Wall Street and Las Vegas. They're evolving into a potentially powerful management tool for making calls on everything from new hit products to next quarter's sales numbers.
The wisdom of the corporate crowd
The concept is also gaining traction elsewhere in the Fortune 500: Best Buy, Corning, General Electric, Google, and Microsoft have begun trying other types of internal markets to divine everything from the future price of LCD TVs to the number of consumers who will buy forthcoming products.
"You can forecast anything with this," says Robin Hanson, a George Mason University economist who began doing pioneering work with prediction markets 20 years ago. "The biggest payoff will come from asking the biggest questions."
The basic idea behind a corporate predictions market will resonate with anyone who understands the premise of James Surowiecki's business best-seller The Wisdom of Crowds: A thousand heads are smarter than one. Prediction markets are more dynamic than a simple poll because those with stronger convictions can have a greater effect on the outcome by putting more at risk.
Financial markets have a long history of flaunting their predictive powers. In New York City during the late 1800s, betting on presidential elections wasn't just allowed by the government, it was hugely popular and very accurate. Self-appointed "betting commissioners" took wagers on Wall Street in the form of futures contracts, and the New York Times and other papers published the odds.
Economics professors Paul Rhode and Koleman Strumpf have calculated that in the betting markets between 1884 and 1940 - when a gambling crackdown and scientific polling put an end to free-market betting-candidates with betting odds of 60 percent or greater on Election Day won nine times out of 10. "A phenomenal success rate," Rhode says.
The concept lay dormant until 1988, when a handful of University of Iowa economists, surprised by Jesse Jackson's win in the Michigan presidential primary, set up a legal market for elections. Called the Iowa Electronic Markets, it eventually allowed people to buy as much as $500 in stock - initially through on-campus terminals - in presidential and congressional candidates. To date it has traded predictions on more than 250 candidates in 12 countries and has proven a better forecaster than the best opinion polls.
Soon after the Iowa markets launched, Hanson wrote a series of papers suggesting a new variation on the basic concept: "Idea futures" that might better predict answers to important scientific questions such as global warming. The rise of the Web, of course, made subsequent versions of the Iowa markets and Hanson's markets easier to create and popularize, since anyone with a browser could play.
HP, meanwhile, was already looking ahead to business applications. Even before Huberman joined in 2001, another HP Labs researcher, Kay-Yut Chen, ran a set of experiments in the late 1990s in which company managers successfully used a market to predict monthly computer sales by allowing workers to trade stocks that corresponded to various sales forecasts.
The market beat the official forecast six out of eight times, but Chen couldn't get senior management to consider it anything more than an experiment. The idea was confusing to many, training people took time, trading was infrequent, and most employees weren't natural traders. "You'd be surprised how bad they are at it," Fine says.
When Huberman - a physicist who had done groundbreaking research on computer networks at Xerox Parc - came aboard, he arrived with ideas to solve those problems. One was to make forecasting a simple betting game instead of a full-fledged trading market. That made it less work for players. Another was to smooth out the differences between players by first determining how risk-averse each is likely to be; that would help enable smarter predictions with smaller, more manageable groups of people.
The 'Brain' in action
The improvements were folded into online software that Huberman and Fine called Brain. Huberman got a small group of finance managers to try it out with bets on predicted monthly revenue and operating profit for the then $12 billion-a-year HP Services division.
Their bets turned out 40 percent more accurate, or $10 million to $50 million closer, than the company's official profit forecasts. The results were "astonishing," says Pierre Legoff, the HP Services finance manager who sponsored and took part in the experiment with 13 of his colleagues.
"To create a traditional forecast process from the bottom up would have been highly labor-intensive and time-consuming. For 12 months, our market almost always came in closer to the outcome than the best individual, predicting revenue and profit within small ranges."
Huberman's work might have gotten more notice from corporate brass had it not been for the distractions following HP's merger with Compaq in 2002. The results of the early Brain experiment caught the attention of two top-flight VC firms, Kleiner Perkins Caufield & Byers and Sequoia Capital, which were interested in building a startup around Huberman's approach. But that deal fell through, and now Huberman is more hopeful about turning it into a separate practice within HP Services, which has big corporate clients keen to test Brain themselves-including an oil company that wants to use it to predict leaks at its refinery.
While Huberman pushes ahead with Brain, other companies are also gaining traction with internal markets. At Google, for instance, Bo Cowgill, a 24-year-old project manager, has run more than 100 online markets for 350 predictions, such as launch dates of new Google services and how many people will use them. About 1,700 employees, or a fifth of the workforce, are players. Cowgill says the highest-priced stock in any market bears out 70 percent of the time.
At GE, researchers are toying with prediction markets to coax more and better ideas out of scientists. Last year managers at GE's computing and decision sciences group created a stock market for 150 lab scientists to enter and rank product ideas - and got execs to sign off on $50,000 in research funding as the grand prize.
Wall Street rocks research funding
Of the 150 people in the lab, 85 participated, including managers, researchers, and even interns. The managers seeded the market with 10 ideas, and the staff added their own. Two weeks later, 62 stocks were trading - each with its own description page where anyone could blog about it, anonymously building a case for it or tearing it down. (Anonymity helps ensure that employees can trade freely without worrying about political pressure to support their bosses' ideas.)
And the winning idea - a software algorithm that teaches computers how to learn - came from a researcher, not a lab manager. "We found we had more ideas that were generated as a result of this process than traditional brainstorming," says Christina LaComb, the GE scientist who set up the market.
With blue-chip goliaths showing people how to mobilize brainpower, you'd think more would be racing to push ahead. Problem is, the better the markets get at performing business tasks, the more they make managers look expendable.
"If you have a culture in which vice presidents have been making the decisions for decades," says Thomas Malone, a professor at MIT's Sloan School of Business, "the very idea that you would let just anybody participate threatens a lot of people."
Huberman, too, knows it's early, but he looks back on how much he's accomplished since the Compaq merger. Back then, he couldn't get any of HP's execs to even look at his work. But he kept e-mailing, giving talks, and following up with those who would listen. And eventually many did. "Companies should take risks," he says. "Otherwise Hewlett-Packard would still be selling oscilloscopes."
Searching for online gold.
Blogging for dollars.