Machine Learning and Chill | Exploring the application of AI at Netflix

December 17, 2019

Machine Learning and Chill | Exploring the application of AI at Netflix

From its humble beginnings as a rent-by-mail DVD service to a household name as the leading internet entertainment platform, Netflix has arguably become the epitome of personalised user experience and content recommendation. Without a doubt, Netflix has proven to be a successful venture and continues to gain subscribers every month, but how exactly does a small DVD-rental company from Scotts Valley grow into a billion-dollar enterprise?

Right off the bat, Randolph and Hastings (the minds behind Netflix) collected user interaction data, in hopes of gaining deeper insight into how customers use the DVD-rental software. They began their data collection process by means of telephone surveys and even occasional house calls, but luckily as technology evolved, so did their data collection methods. Two decades later, Netflix is still focussed on improving user experience through key software-usage-insights and thanks to emerging technologies such as AI and ML, Netflix is able to offer each subscriber a unique experience based on granular subscriber base knowledge.

Use cases of AI and ML at Netflix:

Tailored Movie Recommendations

It’s perhaps not lost on anyone that if person A and person B opened Netflix at the same time, both would be shown different program recommendations. This might seem obvious enough on the surface but taking a peek under the hood tells an entirely different story. The Netflix recommendation system is algorithmic, but what increases the relevancy of these recommendations is a combination of ML and AI. The algorithm changes constantly and uses onsite user behaviour to determine which programs a specific user will be interested in. The algorithm learns and evolves as new data is accumulated, thus the more time you spend on Netflix, the more relevant program recommendations will become.

Thumbnail and Artwork Personalisation

When browsing through a vast array of binge-worthy options, users tend to pay attention to the program thumbnail in order to determine whether or not it’s worth watching. Netflix realised that titles alone can’t persuade users to watch a program, so they turned towards dynamic personalised thumbnails. The specific thumbnail chosen is algorithm-based, in that the user’s preferences and past viewing history is used to select a thumbnail with the highest chance of converting. Each Netflix program has a diverse range of possible posters, each catering to a specific group of viewers, and as the algorithm gathers information and learns more about the viewer, as well as which thumbnails users better respond to in general the images refresh and change in realtime.

Netflix has also gone as far as to personalise program trailers, in an effort to push mass personalisation even further. House of Cards, a Netflix Original political thriller, has been one of the biggest content investments made by Netflix. What separates the personalisation approach Netflix took with regards to House of Cards, is the variety of trailers produced. Each trailer created, was aimed at specific program preference groups, in an effort to spark views from a broader subscriber range. Trailer personalisation further improves user experience, but creating several trailers for all of the programs on Netflix would drive costs through the roof. In order to improve user experience as well as lower the costs involved in creating a variety of trailers,  Netflix announced that they are investing in AI technology which will be able to create a large portion of movie trailers programmatically, giving trailer creators more time to only focus on the creative process.

Optimal Streaming Quality

Streaming quality is a vital metric which directly contributes to user bounce rates. Netflix aims to provide viewers with the best experience possible, but maintaining this standard whilst subscribers stream over 140 million hours of content every day can prove challenging. One only needs to look as far as the recent shakey launch of the Disney+ streaming service to understand how even multi-billion dollar companies can struggle with the challenge of optimisation at scale. By making use of AI and ML solutions, Netflix is able to predict future streaming demand and position assets at strategic server locations ahead of time. By pre-positioning video assets closer to subscribers, viewers are able to stream high-quality video content at peak times without the slightest interruption or dip in quality.

Improving video streaming quality over smartphones has also been addressed by means of AI solutions. Netflix has adopted technology that improves video encoding whilst also decreasing data usage. The AI Neural Network algorithm used, encodes videos scene-by-scene, limiting the size of the video without compromising the quality of the video. The effectiveness of this bandwidth sparing solution was first demonstrated on the trailer for “Iron First”, which resulted in a 50% reduction in streaming speed requirements, without sacrificing the trailer quality.

By offering a tailored user experience to over 158.3 million subscribers, Netflix has proved that granular personalisation is achievable regardless of customer base size. AI and ML are powerful tools with millions of application possibilities, but connecting the dots between business needs and AI solutions are where most companies tend to fall short. Linking business needs with the right mix of AI solutions is one of Swipe iX’s specialities and as an AWS partner, we’ve helped numerous companies solve core business problems through cloud solutions.

Even the most seemingly trivial of business problems (if addressed properly) can make a big difference, so contact us when you’re ready to take your business to the next level.

Hendri Lategan


Swipe iX Newsletter

Subscribe to our email newsletter for useful tips and valuable resources, sent out monthly.