Video: How JUMP Uses Machine Learning to Measure Engagement
Learn more about dynamic ad insertion at Streaming Media's next event.
Read the complete transcript of this clip:
Suzanne Rainey: One of the ways we're using machine learning is with clustering. So we look at it in three ways. The first is by engagement, so we look at your video services, all of your data, and then we determine how engaged the customers are, or the viewers.
So in this case, you can see 64% are inactive, 18% are local, 15% are freezing, which means they're not using it very much, and then 3% are sleeping, which they're not using it at all. And as you might have seen in the first video, all of this data is immediately actionable and you can use a marketing automation program to go directly to your users and market to them.
We're also looking at it by content type, so we have 35% that are addicted to series and movies, 24%, and then just a fraction at broadcast, and we're looking at it by genre. So you can take any of these clusters and then market directly to your consumer so it's allowing you to personally market at scale.
We're also looking at customer lifetime value. So you can see in this example, the average is ... I can't see it on this screen, but it's like $84 or something, and you can see how it skews, and you can see several other analytics that we have about how much their lifetime value is based upon how much they're watching.
And we're looking at the user experience. So this is pretty small here, but we look at the pathway to their first to viewing content, and you can see this is by device, and if it's above average, it means it's taking them a lot longer to get to the piece of content, which is bad for your viewer. And so you can dive deeper into it and go into a heat map, and actually identify the bottleneck, and then go directly into your front end and make adjustments to improve it. And then we have analytics that will show you how it's improved after you've made this choice.
Netflix says they attribute over 75% of their views to their recommendation engine, and I think that most of us know that it doesn't really work that well. So there's a lot of frustration in the market about recommendation engines. And most recommendation engines are just looking at the content and the user, but what we're doing is we're looking at time of day and the day of the week that they're using, and which device they're viewing it on, and where they're viewing it.
So it puts it in a lot better context to be able to make relevant recommendations, and we're also enhancing the metadata with deep learning, as I mentioned, and using image recognition to also enhance the metadata.
Related Articles
Sling TV EVP and Group President Warren Schlichting talks churn in this clip from his opening keynote at Streaming Media West 2018.
29 Nov 2018
Sling TV EVP and Group President discusses the impact of dynamic ad insertion in this clip from his keynote at Streaming Media West.
21 Nov 2018