How Machine Learning Is Making Video More Valuable

Why is Netflix successful? It’s not a question that lends itself to a simple answer, of course, but it’s a question that many in the professional video industry have been probing once it became clear that Netflix was more than a pipeline for shoveling DVDs to home viewers.

Many, like investor Mario Gavira, have argued that Netflix is successful because it’s less an entertainment and content company than a data company. Indeed, Netflix was hardly the first company in the world to offer streaming video. It didn’t invent the idea of pairing licensed Hollywood fare with compelling original content. It’s far from alone in delivering a service using a pleasing user interface.

No, what makes Netflix special is its use of data. Netflix has tons of it. What users watch, when they watch it, how long they watch it for, what parts of the country they’re in and a lot more. They’ve used these kinds of granular insights not simply to cough up “watch next” recommendations but to tailor the creation of its original programming to ensure a higher-than-average success rate.

Netflix has access to this data because, unlike legacy broadcasters and content companies, they’ve been collecting it ever since they went over the top. This trove of data has given them a huge head start over traditional companies that rely on less precise metrics to divine customer behavior.

But that lead is eroding, thanks to improvements in machine learning that enable video content owners to compete with—if not beat—Netflix at its own game.

Many media companies sit on a vast trove of content with often rudimentary metadata attached to it. Machine learning—specifically natural language processing and machine vision—can  improve the quality of that metadata simply by scanning the contents of the video and automatically appending more content-specific tags. This makes video data more easily discoverable, both within a media company and in any public-facing consumption environment (we wanted to open a bar under that same name).

But machine learning can do more than simply add basic tags to video. Using a technique called sentiment analysis, it can also pull greater context from videos. It can classify video by the types of moods and emotions it portrays.

Content owners can use machine learning to fully catalog and organize their assets, but it requires some connection to end viewers to fully unlock the value of this data. By pairing insights from viewers—whether it’s collected in web or mobile browsers, media players or smart TVs—machine learning can then be tapped to analyze viewing habits and inform content owners not simply of who’s watching, but the kinds of videos they watch. Armed with these insights, they can improve their own recommendation engines as well as the quality and precision of the advertisements they’re serving to customers.

It enables every holder of large video assets to become, in effect, their own Netflix. While it’s still in its early days, the adoption of machine learning technologies is likely to accelerate as video owners see the value of the data it unlocks. The end result will be more binge-watchable content, which may not be precisely what the world needs, but will certainly be welcomed by Netflix’s numerous competitors.