Using computing and risk evaluation strategies that bring the techniques of Moneyball to the movies, a British company says it can help film studios determine whether they have a blockbuster such as Avatar, or a financial disaster like the legendary flop Heaven's Gate – before they commit to a production.
Epagogix, which takes its name from Aristotle’s discussion of epigogic learning (basing decisions on past experience), says it can take scripts for film or television and tell studios and producers whether they score well on a large range of specific script elements. The better the score, the more likely the film will be box office gold. And they’ll predict box office returns, too.
Nick Meaney, chief executive of Epagogix, says the company grew out of a handful of critical observations and events back when he was working for a major risk management company.
A former colleague told him about neural networks, computing systems which operate in a similar way to the human brain. They can evaluate a wide range of different elements of a problem at the same time, while also learning from previous "experience" – all the data fed to the system over time. Such systems are at the heart of Big Data, searching for meaningful patterns.
“It was very clear to me that neural networks were a fantastic way to make predictions, in real time,” says Meaney. And potentially, they could do this more accurately than humans, without humans’ innate biases.
Neural networks are a key tool in the risk assessment industry now. However, nobody at his former employer seemed particularly interested at the time.
A conversation with a manager in the company’s Hollywood division is what really got him thinking. The insurance and risk industry calculated premiums for a given film based on the likelihood of accidents and problems – whether someone might slip and fall – not on the chances of the film being a success or a flop.
“That struck me as being a really big opening,” he says. Why not use a neural network to analyse the elements of the script – the single most important element of the film, he notes – to predict its success?
Meaney mentioned the idea to retired Washington DC lawyer and film fanatic, Dick Copaken, and introduced him to a college friend who was developing a system for analysing scripts, and Epagogix was born from their collaboration.
Meaney, who will speak at the upcoming Digital Biscuit (Digitalbiscuit.ie) technology and creativity conference in the Science Gallery in Dublin, says the company built an initial neural network out of historical film data and found that it could predict the success of any given film within a margin of 12 per cent, 85 per cent of the time.
Nine years ago, they headed off to Hollywood to show the system to studio heads. They got a lot of meetings with top figures but, he says, eventually they realised they were being shunted around. Only one of the top six major studios took them seriously.
Epagogix has worked with that studio ever since, as well as with a range of other large studios, independent producers, and television companies. He says he can’t reveal names or companies, nor precisely how the system works.
But he says that one script that performed poorly on their analysis and was therefore dropped by their big client,was eventually picked up by a rival major studio, which lost $80 million on the film.
And – by contrast with the advice of the studio’s human advisers – they accurately predicted that changing the big name actor in the studio’s superhero franchise would not affect the film’s appeal or box office take.
A few years ago, Epagogix piqued the interest of writer Malcolm Gladwell. It was Gladwell who told them that they were just like the analysts in Moneyball. Gladwell wrote a long piece about the company for the New Yorker in 2006, setting them to analyse the original and the shooting scripts for the 2005 Sean Penn and Nicole Kidman film, The Interpreter. They noted the weaknesses in the film that made it perform poorly at the box office and highlighted changes they would have made to improve its chances.
Just as expert baseball scouts doubted – wrongly – Moneyball's statistics geek's predictions on which players could produce a winning baseball team, Meaney says, "oftentimes our analysis is put at odds with what expert insiders thought about a script. But we found you have to stop thinking about gut instincts."
While he won’t go into the details of the process, he says that every script is first read by their human experts, who analyse and score a large number of variables, broadly including elements such as the location, the lead and key supporting characters, the twists and turns of the plot, “and lots and lots more stuff.” Plot, he says, is the most critical.
That analysis is then run through the neural network, which looks at the scoring numbers, and makes predictions about the marketability of the script, and the ultimate box office take for the film, if it were to be produced.
It isn’t a straightforward process in any way. “The importance of those elements varies, according to other elements,” says Meaney. “It’s the overall context.”
The company produces a report that includes action points for general ways in which a script could be altered and strengthened.
The end result is a challenge to screenwriter William Goldman’s famous quote that when it comes to predicting box office success, “No one knows anything.” Meaney argues that his “no one” doesn’t include a well trained neural network.
Gladwell spoke to a number of anonymous studio executives for his New Yorker piece, who noted how eerily accurate Epagogix's analysis could often be.
Meaney doesn’t claim the system is perfect, not least because so many elements go into the successful filming and editing of a script.
And he concedes that the process is very much geared towards box office success, as opposed to critical esteem. The predictions also are primarily good for Hollywood films aimed at US audiences, not European art house productions.
Still, he says, a quirky, independent script may well score highly because the neural network has so much data in it. Unless the film is totally unique in myriad ways, the network should recognise the elements of a strong film.
That’s one reason why he argues that the system is not geared towards conveyor-belt sameness in films, or stifling creativity. Rather, he says, it actually enhances the position of the script writer, who typically sits fairly low on the Hollywood totem pole.
“The script is pretty much the central point of the film – not the talent, not location. I would say what we do is we read scripts, after many eyes have already read the scripts, and help bring the best out of the script. We think we help the writer to do the best they can.
"Even Charles Dickens had editors."
Nick Meaney will speak on on the topic, Friend or Foe: Do Neural Networks Help or Hinder Creative Expression? at the 2014 Digital Biscuit conference at the Science Gallery, Dublin, January 24th. More details and tickets at Digitalbiscuit.ie