Tom Davenport, who holds the President's Chair in Information Technology Management at Babson University, as well as recently being rated the highest-ranking business academic in Ziff-Davis' listing of the 100 most influential people in the IT industry, took the time to speak with us about how companies use predictive technologies to help consumers choose products.
Why are predictive models necessary for consumers?
Basically, it's been a fairly hit-or-miss process to find out what consumers really want. Many industries haven't done a good job. In our research, we primarily focused on industries that produce cultural products—books, movies, music—and those industries have a really low hit rate. The vast majority of them lose a lot of money. So in industries where there are such low batting averages, we believe that predictive analytics will help.
It's also good for consumers because there are lots and lots of offerings available and nobody has enough attention and time to sort through them all, so if you can get some help with what you like—learning about particular attributes—then it's going to be good for you. If you're going to invest a couple of hours and 20 bucks into watching a movie, you'd prefer to see one you like.
How do predictive technologies help companies gain competitive edge?
It's based on data and analysis of what customers have liked in the past, and if companies assiduously use this analysis, they can dramatically improve the rate at which their products are successful and picked up in the marketplace. Most industries have a pretty good understanding of this. One of the things people think Obama did so well was behavioral targeting of people online. Or, Amazon has been using predictive modeling successfully for over a decade. Some industries resist—with cultural products there is a bias against prediction. There are lots of efforts to use it, but it's all pretty nascent. It's going to take off, though, at some point.
Could you tell me about a company that has succeeded using predictive models?
If you look at Will Smith, he's been successful in his relatively informal approach to classifying what movies will succeed and then deciding which roles to take. Studios are more conservative, but some like Lion's Gate are more aggressive than others, and have been pretty successful, but they could do more. There are the companies that distribute cultural products. Netflix argues that customers like its choices for what to watch 10% more than they do when they themselves choose. Blockbuster and Overstock have gotten good results using predictive
recommendation systems. In general, it's being used more on the consumer/distributor side than it is on the creator side.
You've mentioned that Hollywood has overall been resistant to this idea of using predictive modeling. Do you think they'll come around to the idea of using analytical tools to predict artistic successes?
I do, because the stakes are just too high. Only 6% of movies make a profit. If you're a baseball player with that batting average, you wouldn't last. Or if you were a doctor and only 6% of your patients survived, that'd be a disaster. The economics are so compelling that movie studios are going to have to move in this direction. They feel it's art, and not science, and so it can't be predicted, but that's an outdated proclamation. There are some services out there that will help studios if they want. The appetite is not that strong for it, but it will gain.
What are some of the things organizations who want to incorporate predictive technologies into their businesses need to consider?
Well, there are various techniques, so think about which works for you. The most common is collaborative filtering: Customers who bought the same thing that you bought also bought this other thing, and you might want that. Amazon had some problems with not incorporating whether the purchase was for yourself or for a gift. You've got to work pretty hard to make sure these are your actual customer preferences. But in the last month or so, they introduced that approach.
Also, if you need to recommend cultural products, one popular way to do it is through defining attributes. For instance, say you're interested in movies that have strong characters and a female lead and a love story and lots of dramatic action. It's very difficult and time-consuming to make those classifications, so then if you're a company, you look for already existing databases, as there are for films.
Or, Pandora is an online radio station that has classified all of its content. People tend to like it, but it's a very expensive approach and not that great a business model. Recommendations by themselves don't really pay off. If you have another source of revenue—like distributing books or movies—then recommendations can be a good adjunct, but nobody has made money selling it by itself. It's both expensive to get it classified in the first place, and the whole model of distributing music online is somewhat problematic anyway. The great outcome for them would be to get snapped up by Apple, which has a poor classification system. The combination would be dynamite.