Updated: Jun 7, 2019
AI researcher and the Director of the AI and Neuroscience in Media Project at USC’s Entertainment Technology Center.
1. What does Machine Learning actually means for Brands?
Machine learning can be very powerful for brands. Think about it like a microscope: if applied properly (and that’s a big “if”) machine learning can help brands look much more deeply and scientifically into their audiences, their own story, and the cognitive relationship between the two. This is the Holy Grail of marketing: for over a century, brands have told stories, and sometimes these stories have resulted in increased brand awareness, equity, and sales. Nobody has ever been able to scientifically explain what types of brand stories generate what type of customer affinity, and how that translates into purchasing behavior. Machine learning can help brands investigate these processes, sometimes very deeply, and that’s what we do at Corto: we apply computer vision, natural language processing, and many other ML and AI techniques to help brands understand what kind of narrative structure generates what kind of cognitive affinity, and how that translates into customer behavior.
For example, we use ML to very granularly analyze all the visual and audio attributes of film trailers or video ads and analyze those attributes in relation to conversations about those trailers and ads: there we find a lot of extremely valuable data about what attributes of the brand resonate in what way with what audience segments. How a certain look stacked with a certain edit pace and a certain music tonality generates a specific narrative point or emotion. It’s very powerful.
That said, I want to be clear that machine learning isn’t a magic wand, and often, much more basic statistical methods generate higher returns than very involved methods. I’ve created millions of dollars of value for some brand clients just using linear regression, which can be done in Excel.
But a lot of this success is predicated on the presence of audience data, and not just polls or panels: you need actual audience conversations, because these are very rich in meaning. But if nobody’s talking about you, there’s no data, and without data there isn’t much you can do.
Get the Issue to read full interview