The process used to create a major motion picture hasn't changed much in a century. Someone has an idea that they turn into a screenplay, which is then edited, developed and handed to a director, who brings it to life. The only difference now is that there's plenty more focus-grouping, audience analysis and number crunching to ensure each film is a hit. Except that doesn't really work, since 2016 alone has seen scores of movies unceremoniously crash and burn. But maybe that will change with ScriptBook, an algorithm that its creators say can spot most turkeys before they've even been made.
The Belgian startup was born out of one of Hollywood's greatest catastrophes: the Ben Affleck–Jennifer Lopez train wreck Gigli. ScriptBook CEO Nadira Azermai was a college student when she got the chance to intern on the film for a couple of weeks. Inspired by the film's failure, she wrote her thesis in applied economics on a way of using machine learning to develop a tool that would enable producers to avoid box office bombs. And then, in her own words, she "left it [the idea] in my bottom draw to go work for a bank" for a few years.
When she decided to revisit the idea, it took her (and her burgeoning team) a year of research and development. The finished product was ScriptBook, a machine-learning platform that -- so Azermai claims -- knows what makes a good screenplay. "In the first six months, we did a lot of exploration, taking 4,000 scripts and 10,000 movies with metadata to see what parameters came out," she said. The result is an algorithm that knows what has worked before and can judge a screenplay against 220 parameters that are used to calculate its theoretical financial performance.
The sort of report a studio executive using ScriptBook could expect to receive.
Some of the insights were derived from common sense as well as the established "rules" of screenwriting sold by Robert Mckee, Syd Field and Blake Snyder. That means following Joseph Campbell's monomyth structure, ensuring that your lead character is sympathetic and goes on a hero's journey. But likability was a hard substance to quantify, and one that ScriptBook's algorithm initially struggled with.
Azermai cited Die Hard as an example, since, on paper, "the lead character -- he's not a scumbag, but he's an unlikeable, disgruntled cop." So, to teach ScriptBook, "we hired people to annotate the data set and answer questions on if the main character was likable." Once the system had looked at the human input, it can then classify these traits automatically, as is the case with most deep learning systems.
Scripbook isn't about picking winners so much as it is about avoiding losers, which is a huge issue for even a big studio. "We did an impact analysis" on the slate of major studio movies released in 2014 and 2015, says Azermai. "ScriptBook would only have green-lit 42 out of 70," with only a handful of false positives, but she claims that her product avoided the biggest flops. And saving studio cash could be a big business: the cumulative deficits caused by 2016's ten biggest flops -- as charted by Forbes -- was a whopping $100.9 million.
Azermai references the 2015 remake of Point Break as evidence of the sheer power of the platform that she's put together. A fan of the original film, she ran the screenplay for the remake as soon as she got her hands on it. The algorithm, however, determined -- months before release -- that the film would gross only $31 million in the US. In her telling, the result "made me really doubt our system," because the remake seemed like a slam dunk. When the film debuted that December, its domestic gross, as calculated by BoxOfficeMojo, was just $28.8 million -- more of a flop than even her system could determine.
There's a boatload of ifs and buts, but if ScriptBook works, then Azermai believes she could have a massively successful product. After all, spending a million dollars on script analysis is chump change compared with eating a hundred times that in losses. Even if the studio was used only 20 or so times a year, that's still enough for ScriptBook's creators to kick back and relax. If you look at some of the studios' lineups (oh hey, Warner Bros.) for the next few years, it's clear that there's a need for this sort of QA checking.
Nadira Azermai and her team at ScriptBook.
ScriptBook isn't the only game in town: Tools such as ScripThreads can analyze a screenplay and visualize its storyline and character interaction, while Slated offers screenplay analysis based on the pooled review scores of three unnamed studio development team employees. But Azermai's product is the only one that offers an apparently concrete prediction of a movie's potential success.
The platform will be launching in the near future, and while its initial pitch will be to studios, it won't stay that way. Azermai told Engadget that the company is "working on" a more limited tool that'll be offered to screenwriters. Rather than the more detailed financial analysis, it'll offer a generic metric of a script's quality and likelihood of success. Thankfully, Azermai promises that it'll also be cheap enough for most dirt-poor typewriter junkies to afford.
ScriptBook's claims are big and broad, and if it works, it certainly has the potential to upend the way movies are made. The only problem, for now, is that it's nearly impossible to demonstrate or prove that the results it produces are legitimate -- for instance, Slash Film decried its claims as "complete and utter B.S." Azermai isn't worried about the "older creatives," who are skeptical, because it's Hollywood's decision makers -- the accountants -- who will determine the product's success.
There are still a thousand unanswered questions as to how ScriptBook actually works, and few are currently forthcoming. We were allowed a day to play around with the firm's embryonic user interface, but any attempt to test the system with a fresh screenplay was resisted. All the same, the company is getting ready to unleash itself upon an unsuspecting world, and if we see the quantity of flops decreasing in five years or so, we'll know who to thank.