The Real Limits of Prediction

Back in July 2005, I published a column titled The Limits of Prediction that addressed barriers to the adoption of predictive analytics. I've continued to think about the topic even as politicians come up with yet more wishful-thinking panaceas like off-shore oil drilling to reduce gasoline prices and business — for instance, the real-estate, financial, and automotive sectors — reports huge losses derived from in-hindsight foolish decisions. All those folks spend massive sums on IT and analytics. Why didn't they know better?Take automotive. A 2001 CIO magazine article says "Ralph Szygenda, group vice president of information systems and services and CIO at General Motors, does not suffer fools or tolerate excuses." Szygenda must find it very, very difficult to meet with his bosses. They're the folks who recently drove GM's share prices to 1950s levels — the stock has since recovered a bit — based on abysmal sales of the low-fuel-economy vehicles GM is tooled to produce. Do you suppose the root was poor predictive capabilities — a Web search turns up a number of stories documenting GM's analytics use — or management that thought they knew better?

An April interviewer (eWeek, GM`s Ralph Szygenda Drives IT Innovation) asked Szygenda, "GM's stock was down last month, and estimates were not good for the rest of this year, even for 2009. Is there something you can do in IT to help?"

His response: "We already are. This business is running much more efficiently. Every year, this business is reducing cost. Billions of dollars have come out of the cost structure because of the process changes." Szgenda elaborated, "I have always pushed three things: standardization, simplification and collaboration. You need all three things to innovate." The stakes are huge: "To build vehicles, you're taking $90 billion in materials and services."

I'm sure that Szygenda's efforts over the years included steps to reduce time required to retool manufacturing in response to shifts in demand, but efficiency clearly isn't enough if you're headed in the wrong direction. How is it that large, technology reliant organizations can bet so poorly?

I don't mean to fault Ralph Szygenda. GM's problems most likely stem from from scenarios and forecasts that were dismissed or insufficiently hedged and not from analytical short-comings. As the saying goes, a fish rots from the head down. There's little analytics can do to help when leaders won't listen to the numbers or act on what they show.Back in July 2005, I published a column titled "The Limits of Prediction" that addressed barriers to the adoption of predictive analytics. I've continued to think about the topic even as politicians come up with yet more wishful-thinking panaceas and business reports huge losses derived from foolish decisions. All those folks spend massive sums on IT and analytics. Why didn't they know better?