One of 2017’s highlights for me was participating in TNW New York. The event was packed with amazing content and inspiring speakers discussing what’s “Now” and “Next” in digital. I was fortunate enough to be one of the thousand attendees who walked away with new perspective and insights into IoT, AI, data, and blockchain.
Netflix’s Head of Machine Learning, Tony Jebara, was one of the stand-out speakers. His talk on machine learning for personalization really got me thinking about the future of marketing, user engagement, and retention amidst these emerging technologies.
As we all know, humans are storytelling animals, originally relying on spoken word handed down from generation to generation. While a story typically has a beginning, middle, and end, the manner in which it’s told can impact both the audience and storyteller. The nature of two-way conversation adds another dimension to what otherwise would be a seemingly linear story. Facial expressions, body language, and clarifying questions all help shape the experience. It also lets the storyteller know if you like what your hearing, and more importantly, will listen again. With the advent of books, tv, and film, we’ve lost much of that back and forth interaction. That doesn’t mean, however, that the two-way interaction is gone for good.
Netflix is now able to add a layer of “reactions” back into storytelling to exponentially improve personalization, recommendations, and the overall user experience – all through machine learning.
My guess is that most people know that when they “like” or “dislike” a movie on netflix, their action will influence what types of movies will be recommended going forward. As marketers, we also assume that any digital content platform will try and track as many direct (or implied) user interactions as possible. However, the depth to which Netflix collects a user’s behavioral data to personalize their experience is pretty remarkable. I think it’s safe to assume that they’re tracking not only what you watch and why, but also where you prefer to do your watching.
All that being said, Netflix knows A LOT about its customers. So what do they do with all of that data?
A simple example of this can be seen when you examine how personalization of artwork for every title in Netflix’s library can become both meaningful and uniquely useful to each customer. Given the huge variance in individual preference, obviously it would be in Netflix’s best interest (as well as the end user) to optimize artwork for each member in order to highlight the aspects of a title that are most relevant to them.
Let’s consider how personalizing the image for the movie Forrest Gump could work. Keep in mind that Netflix has dozens of candidate artwork images for each title. Someone who watches a lot of action/war movies may be more interested in Forrest Gump if they’re shown artwork showing Gary Sinise leading a platoon into the jungles of Vietnam. That very same movie title may be presented to an avid fan of Castaway with a representation of Tom Hanks running cross-country with long hair and a shaggy beard.
Most of the guesswork is eliminated through sophisticated machine learning algorithms that leverage “contextual, multi-armed bandits.” This approach prevents large samplings of users from missing out on the personalization benefits during the machine learning process. Something that’s just not possible with simple A/B testing.
This is just one small example of how Netflix is leveraging massive amounts of user data and a platform that actually gets smarter the more you interact with it to benefit its members. It’s tough to judge a book (or movie) by its cover, but at least now that cover will be personalized.