There’s a lot of literature that claims a successful Hollywood script can be broken down into a formula.
The most famous is the screenplay guide Save the Cat! by guru Blake Snyder, which has swept through the screenwriting world with its minute-by-minute formula for how to wow the audience.
Now one startup is taking a crack at the screenwriting formula from a machine learning perspective. Vault, an Israeli artificial intelligence company, has created a program that claims to be able to tell whether a film will be a hit or a flop, simply by reading the script.
But how?
David Stiff, Vault’s CEO and co-founder, says it hinges on an intensive analysis of 300,000 to 400,000 story “features,” which can be things like themes or level of violence. All these story features are pulled from the script by his program with no human input.
Vault trained its AI using script data from films going back to 1980, when Stiff says there was a shift in Hollywood toward the “blockbuster” model. The team fed the system the script, allowing it to compare data points to the box office performance data.
Stiff now claims his algorithm can predict the box office performance of a film with 65% to 70% accuracy. This an extremely high percentage given that only 20% of movies make their money back, he says.
When I press Stiff to reveal what factors are most important to a successful film, he cites themes. “If we take out themes from our predictions, our rates drop dramatically,” he says. This makes sense. There are themes that have been recycled from the time of Ancient Greece to now, and they still move us.
But one aspect Stiff thinks Hollywood puts too much emphasis on is the star power of actors. His AI can also suggest actors a studio could cast, based on the script, but the focus is on saving money. Stars can be useful at the box office, but a series of high-profile flops from actors like Johnny Depp prove that even an acting legend can’t save a sinking script. This thinking runs counter to the prevailing wisdom in Hollywood, which places screenwriters more toward the bottom of the food chain.
Stiff says if he had to write a film based on what he’s learned, he’d use no-name actors in an action comedy with a budget of $30 million. But he stresses that the formula is too complex to be “gamed” in a straightforward way. It would be easier for him to select a portfolio of 10 movies, he says.
This nod to diversification harkens back to the team's background in algorithmic trading. While Vault works with both studios and investors, Stiff envisions the program working particularly well for people who want to wade into the film-funding marketplace.
Vault started its analysis efforts with film because the team thought it would be hard, Stiff says. There are so many elements that go into a film besides the script, that if they could nail this, they could easily move into other industries. To that end, Stiff says the company plans to move into TV and even publishing.
And when has Vault been the most wrong?
Stiff said his AI thought that the latest Terminator movie was going to be a hit, but it’s looking like a disappointment at the box office. Maybe it was just a case of robots favoring robots.
You can see Vault’s full 2015 predictions, a mixture of indie and studio films, by clicking here.
Join the conversation about this story »
NOW WATCH: An incredible true story is brought to life in the riveting first trailer for 'The 33'