When to Forget the Rearview Mirror

How can marketers predict whether audiences will pay to see a new film or download a new song? Such forecasts are notoriously tricky. Academic researchers call films and songs “fashion products,” because their sales are driven by volatile consumer tastes. Fashion products often have short life cycles and rely on impulsive purchase decisions. One way to improve forecasts is to seek more data. But new research suggests that with fashion products, this doesn’t always help. IE Business School’s Matthias Seifert and a team of colleagues examined the joint role of historical and contextual data in human judgments about which products will take off and which will sputter. For example, how does a music company weigh historical data (“Taylor Swift’s previous album sold X copies”) against contextual data (“We plan to spend Y marketing this album”) when evaluating upcoming releases? Because creative industries are highly dynamic, historical data isn’t always useful on its own. Although you might assume that, say, a movie with a big-name actor will be a hit, studies have found that star power isn’t actually a significant predictor of box office receipts. And contextual data tends to be subjective and qualitative, which increases the complexity of the judgment at hand. Seifert’s team studied predictions about where pop singles would enter the Top 100 chart. First they interviewed 23 senior managers at four major record companies to identify “predictor variables”—factors that ought to correlate with success. These included the marketing budget for each single, who else was releasing singles the same week, and whether the artist was an established performer or a fresh face. Then they gave lists of forthcoming singles, along with predictor-variable information about them, to 92 A&R managers—people who scout and recruit musicians. Over a 12-week period the A&R managers completed four online questionnaires about the likely chart-entry positions of the singles they’d been assigned, using the predictor variables to make their forecasts—210 in all. After that the researchers sat back to see how each single performed. Finally they categorized each predictor variable as historical or contextual and ran mathematical analyses to see which type had worked best—and how the two had worked together. They found an interesting wrinkle: We make better judgments about volatile demand when we consider only contextual data. Historical data seemed to impair the A&R managers’ ability to interpret context. Because algorithms can forecast linear relationships much better than the human brain can, Seifert’s team suggests letting a computer make decisions in stable environments where the predictions depend solely on historical data. But when asking someone to make a judgment call in a volatile environment, consider withholding historical information so that he or she can focus on contextual information. More data isn’t always better. About the Research: “Effective Judgmental Forecasting in the Context of Fashion Products,” by Matthias Seifert, Enno Siemsen, Allègre L. Hadida, and Andreas B. Eisingerich

When to Forget the Rearview Mirror

Partner Center