In this week's New Yorker, Malcolm Gladwell writes about a company which has devised a formula to predict a movie's box office success (also available as a PDF). This company, Epagogix, has a proprietary formula that relies on grading the movie along a huge variety of script or narrative elements (themes, characters, locales, costumes, and on and on), then runs them through a neural network which spits out a predicted box office take.
It sounds like some holy grail for the movie industry, but many caveats apply, as you'd expect. The script has to be graded on those script elements by human beings who read the script. They say the process takes their team a day to do, but if the formula relies on people with particular aesthetic tastes or training, that isn't scalable.
Another issue is that the neural network can only be trained based on past results, by feeding it old screenplays and their box office results. If the public's tastes change, the network has to be retrained, though I wouldn't expect this to be a huge problem.
Finally, if the formula works purely off of scripts, a lousy production might still torpedo the box office. Again, this isn't a problem for Epagogix since it sounds like they're being paid for pre-production script consulting.
I'd be curious to see them use the network to put together an ideal script, a Frankensteinian list of what narrative elements would go into a movie with the highest potential box office. The idea of storytelling being reduced to some formula spit out by a computer is repulsive to the artistic side in me but thrilling to my contrarian, scientific half.