DECISION SUPPORT

21. When Good People Make Bad Decisions

by Michael Schrage

New economic research affirms what most executives suspect: even when given good data, people make bad decisions. We misunderstand, misinterpret and mismanage important problems. It's not that we're stupid; our thought processes have bugs. There are technologies emerging now to help us fix those bugs. Decision support software has been around for years, but its latest incarnations tap pervasive computing power to generate thousands of scenarios covering all conceivable real world contingencies. Products such as @risk and XL Sim allow people to test their assumptions through what is essentially repeated rolls of the dice - random number generation. The technique - known as Monte Carlo simulation - can turn ordinary spreadsheets into probabilistic wind tunnels for designing good decisions.

"With any luck, the rise of cheap, easy Monte Carlo simulations will reduce the number of stupid decisions managers make by relying on simplistic averages," says Stanford University consulting professor Sam Savage, the creator of XL Sim and a pioneer in spreadsheet-based statistical literacy. Consider the case of a Silicon Valley product manager who has just been asked to forecast demand for a next-generation microchip. "The guy typically will offer a forecast range between, say, 50,000 and 150,000 units," he says.

The problem, says Savage, is that the boss doesn't want a range. He wants a number. So the manager says: "100,000", the average. So the boss plugs that figure and the cost of building a 100,000-chip-capacity plant into a spreadsheet. The bottom line is a healthy $10 million, which he reports to his board as the average expected profit. Assuming that demand is the only uncertainty and that 100,000 is the correct average, then $10 million must be the best guess for profit. Right? Wrong. What Savage calls "The Flaw of Averages" ensures that average profit has to be less than the profit associated with the predicted average demand. If demand is less than 100,000, then profits will be lower than $10 million. But the profits can never be higher than $10 million because the maximum capacity of the plant is based on a flawed average. Consequently, the product manager's correct forecast of average demand leads to an inflated forecast of average profit.

Savage predicts that in time every executive managing plant capacity, every investor with a stock portfolio and every employee with a retirement fund will be running Monte Carlo simulations to test their intuitive assumptions about average returns and average losses.

But that's just one example of how people trip themselves up. There are others. A thick body of research affirms that individuals think about their purchases differently depending on whether they use cash, credit cards, debit cards or cheques. They also put a much higher value on something they already own than on the same thing if offered for sale. This "prospect theory", which won the 2002 Nobel Prize for professor Daniel Kahneman and cited his research partnership with the late Amos Tversky, helps explain why people hold on too long to losing stocks or pay way too much to insure themselves against small losses.

MIT Sloan School of Management professor Dan Ariely and his graduate students are experimenting with an "electronic wallet" that would advise users on how best to spend their money. For example, your wallet may run a Monte Carlo simulation displaying how, based on your payment behaviour, you would save $150 over the next three months by using cash for a purchase instead of your Visa card. In other words, tomorrow's technologies will load the dice in favour of people not repeating the sort of silly statistical mistakes that lead to Nobel Prize winning research. And that will merit a prize of its own.