Supervised and Unsupervised ML Applications in Media & Entertainment

Video

Transcript

Unsupervised in the middle of the text of the correlation, right, there’s lots of business operations, business questions, rather, that we didn’t think to ask and then about segmentation, so not based on demographics. So, for example, people who love horror movies or baby boomers, the factors that really correlate to revenue churn risk NCLB. So when it comes to predicting churn, we look at the first thing we look at is the AI versus the existing models, for example, accuracy. The percentage of users with predicted high risk within a segment say for a particular pricing, your product, false positives or false negatives, the ACC performed better. And then we look at things like arawa, how much revenue loss and we prevent or how much revenue gain in terms of percentage and overall dollars with AI. And then we look at some interesting data points that most companies weren’t tracking before we looked at distribution. So how does retention compare across audience segments like hoppers versus free trialist versus loyal viewers? And we look at consider how much revenue are you making per title in your library? We can do that using data. Maybe there’s something that you should license if you own it or absolutely renew it if your licensing the content based on that data and then lastly, anomalies. So properly train them out and sniff out something that seems off like what characteristics are recurring among the highest risk customers. Maybe it’s a data point. You weren’t looking for like unintentional churn due to credit card expirations or correlated to the cancellation of a show that was popular with an audience. So, again, the licensing thing, if that show goes away, you may see a percentage of churn rise because of that title going off your service.