2007 was a terrible year for movie stars. But not for Will Smith. That year, Smith's film "I Am Legend" set a box-office record for a December movie opening, taking in $77 million. He continued with that success in 2008. His movie "Hancock," despite receiving poor critical reviews, grossed more than $625 million worldwide. In fact, the movies Smith stars in have higher opening weekends and average box-office receipts than movies with any other male lead.
Does Smith know something other actors don't? Perhaps: When Smith began his Hollywood film career, he and his manager studied a list of the 10 top-grossing movies of all time. "We looked at them and said, OK, what are the patterns?" Smith told Time magazine. "We realized that 10 out of 10 had special effects. Nine out of 10 had special effects with creatures. Eight out of 10 had special effects with creatures and a love story." Smith sees himself as a "student of universal patterns" and studies box-office results after every weekend. He's clearly an astute observer: He has a track record of choosing films that deliver $120 million or more.
Smith's ability to predict which movies will succeed belies conventional wisdom about predicting consumer taste. We see such predictions as an art, not a science. And, of course, creative judgment will always play a vital role in shaping and marketing cultural products. But today the balance between art and science is shifting. Companies now have access to data and sophisticated technologies that allow them to weigh factors and consider evidence that before were unobtainable.
As a result, creators and distributors of cultural products are spending more time attempting to predict how successful their products will be—before, during and after they are created. Such knowledge is only going to become more critical as offerings proliferate. So producers must figure out how to wisely invest in products in a world already cluttered with them. Likewise, given how many options they have, consumers need help deciding what media they'll most likely enjoy.
Useful consumer recommendations really came into being in the late 1990s, when Amazon.com pioneered "collaborative filtering." This software makes recommendations by analyzing consumers' past choices and making correlations with other products they may like. Collaborative filtering can be useful in pointing shoppers toward products they didn't know existed, but it's also limited. It has no way of knowing, for instance, when someone has purchased an item for someone else—as a gift, perhaps, or on someone else's behalf.
More recently, Netflix has had success with another form of collaborative filtering. Its software produces movie recommendations by correlating a data set of more than a billion customer movie ratings. Another example, the TiVo "Suggestions" feature, selects shows it predicts consumers will like based on their viewing patterns and ratings of other programs.
There are other predictive technologies that exist. Another approach focuses on an item's attribute. A movie, for example, might be classified by its length, genre, theme, tone or reviews. Analyzing the movies a customer likes could lead to recommending other movies with similar attributes. This is currently being done for music. The online radio station Pandora and the musical software recommendation company Echo Nest have classified thousands of songs using many different aspects —including timbre, key, tempo, time signature and orchestrations.
Some companies are also beginning to add social networking as a means of recommendation. The idea is that if your friends like certain songs and movies, perhaps you will, too. And if you and a stranger like the same movies and songs, perhaps you should become friends. Netflix, for instance, has a "Friends" service that lets customers share movie preferences and reviews with a community.
Each approach has varying strengths and weaknesses. Collaborative filtering, for example, requires a substantial amount of data on past purchases. Neural networks also require a large amount of data. Attribute-based recommendations require that someone classify products according to key attributes—and these attributes can be difficult to develop. Prediction markets need a large number of independent participants and often must offer prizes or rewards to attract enough people. Overall, the best recommendation tools perform a balancing act: They connect to consumers' sense of individuality as well as their group identities.