‘Not one person in the entire motion picture field knows for a certainty what’s going to work,” William Goldman, the Oscar-winning screenwriter of hits like “The Princess Bride” and “All the President’s Men,” wrote in his 1983 book about working in Hollywood. “Every time out it’s a guess and, if you’re lucky, an educated one.”
But after months of late-night coding sessions in a small North End apartment, three recent college graduates think they’ve removed much of the guesswork from box office prognostication, unveiling a computer algorithm they say can predict how a film will fare long before the director ever calls “action!”
Their program recently calculated that “Kevin Hart: What Now?” and “War Dogs” would gross about $11.79 million and $13.85 million, respectively, in their first weekends. The actual hauls were $11.98 million and $14.3 million. Other guesses weren’t quite as close, but the software — inspired in part by Netflix’s pioneering “what to watch next” algorithm — is getting better all the time as it digests more data. Overall, its first-weekend revenue predictions are off by a median of just $2 million.
The graduates’ nascent company, Pilot Movies, earlier this month released a preliminary version of its profit-predicting Web app to potential customers. The firm plans to add more features and a slicker interface before pitching studios and investors later this year.
Pilot Movies is a bootstrapped dream for three young computer-science majors and film buffs who weren’t even born when Goldman wrote the 1969 blockbuster “Butch Cassidy and the Sundance Kid,” and the odds that they turn math into gold seem long. But their product is timely, arriving amid a wave of attempts to extend computers’ reach further into the realm of the subjective — What makes a successful movie? What do you want to see in your Facebook feed? — by harnessing massive sets of data and advanced statistical techniques.
“The fundamental statistical methods that we use in our models are the exact same ones that folks use to win millions of dollars on DraftKings and that folks like Nate Silver use to predict elections, but applied to a different problem and different data set,” said Pilot cofounder and chief executive Alan Xie, a 22-year-old who started working on the formula while a Harvard undergrad. “It’s not magic.”
Online streaming services have been using similar “big data” for years to help decide which projects to green-light or license. For example, Netflix optioned “House of Cards” after noticing that users who watched the show’s British incarnation also loved films starring Kevin Spacey.
Most major movie studios, on the other hand, have been less enthusiastic about algorithms. Some predict box office returns by analyzing social media posts and online searches in advance of a film’s release, but Xie argues those numbers simply measure the effect of movie ad campaigns, a kind of self-fulfilling prophecy.
“It’s like a [political] poll the week before the election,” said Xie, who started the company with fellow Harvard students Alan Rozet and Tianxing “Vincent” Lan. “We can tell you who’s going to win during the primaries.”
Studios also worry about soulless robots interfering with the creative process. But despite recent headlines about computers rewriting (or even writing) scripts, and unlike some of its competitors, Pilot won’t tell you to add an ice-skating scene to your rom-com or which actress to cast as the sympathetic best friend.
“Pilot isn’t a platform that usurps the creative process,” Xie said. “It actually simplifies it by abstracting the financial uncertainty and complexity of filmmaking into a one-click piece of software.”
Pilot’s algorithm works by comparing a potential film project to a massive database of information about every movie widely released since 1990. Input the cast, director, writer, budget, rating, release date, genre, and plot summary, among other variables, and the app spits out projected box office takes for the all-important first weekend and the full domestic theater run. Change one element and a new estimate is immediately generated.
Other companies offer similar predictive services, but Xie says Pilot is different because its software-as-a-service model allows users to experiment with endless combinations right in their Web browser — no need to wait for pricy consultants to prepare a new report. He also insists the math under Pilot’s hood is more robust than competitors’ formulas.
Behind the scenes, Pilot’s algorithm harnesses so-called network effects.
‘Pilot isn’t a platform that usurps the creative process. It actually simplifies it.’
For example, rather than automatically deem Matt Damon a 9 out of 10 on a scale of relative profitability, it measures how Damon’s box-office power changes depending on the film’s genre, director, and cast.
Xie plans to hawk Pilot’s software, which will cost several hundred dollars for a monthly subscription, to small- and mid-size studios, talent agents, financiers, insurers, and other players in the movie industry that need an idea of a project’s financial potential. Xie hopes they’ll come to adopt it as a standard decision-making tool.
Despite the ingenuity of its software, Pilot faces substantial obstacles, including its founders’ status as young outsiders in an industry built on personalities and old-school schmoozing.
There are also well-funded competitors such as Scriptbook and Epagogix that also use algorithms to forecast box office results. And, some industry skeptics said, box office returns are just really hard to predict.
“We all to hope to find that magic formula to actually predict the box office,” said Ben Spergel, president of media and entertainment consultancy SpeakEasy Research. “But we’re far from being able to do that simply because so many other variables come into play. At best you can sort films into buckets of relative earning potential.”
The biggest unknown, of course, is execution. A bad acting performance can torpedo a film that looks great on paper. There’s also the unpredictable shifting of release dates as studios jostle for prime weekends. Plus, actors in small-time films probably have few previous appearances, making it harder to predict their future performance. And studios may be reluctant to rely on a calculation whose margin of error is too wide.
“We have enormous funding and a team of 100 people and 12 PhDs in various computational sciences and we still don’t do a great job,” said Matt Marolda, who heads the Boston-based in-house analytics team at Legendary Entertainment , one of the largest film companies to embrace data. “The other challenge is industry acceptance: This kind of analytical approach is just not typically used in a lot of contexts right now.”
Spergel thinks analytics companies such as Pilot will eventually combine forces with more traditional creative advisers who speak Hollywood’s language.
“In a perfect world, these two forces would stop competing,” he said. “They need each other.”Dan Adams can be reached at firstname.lastname@example.org. Follow him on Twitter @Dan_Adams86.