Last night, I struggled to find a solid chick flick movie among several friends that none of us had seen. While scrolling through my Netflix account, there were specific recommendations based on other movies I had chosen. There were the classics that many of us had seen countless times, but very few that none of us had ever seen and that could appeal to us…until one from 1993 with Demi Moore, Robert Redford and Woody Harrelson was suggested and it fit the perfect bill. The question begs, why and how did Netflix suggest this exact film among the countless chick flicks out there?
Netflix invests substantial money into capturing data and then driving revenue off of it by giving users what they want often before they even know they want it. Approximately 75% of all streamed movies and shows are recommended. Netflix captures data and knows what you play, searched for, rated in addition to the exact time and date as well as the specific device you have searched on. They then can utilize the metadata to make specific recommendations and continually improve the recommendations over time and at different times depending on the day of the week. Netflix categorizes recommendations based on specific programming you have recently watched and rated and can make recommendations so that you can watch that perfect rom com or action flick depending on the setting and mood you are in. However, it can predict that on a Wednesday evening you are more likely in the mood for an unwinding comedy than the drama date night you watched last Friday.
Netflix offers an integrated subscription model offering both DVD by mail as well as streaming services streaming thousands of movies and TV shows to users on demand. With over 60 million subscribers, they have 3x as many as the next competitor – Amazon instant video. Netflix offers the largest content of any of the largest streaming giants (estimated to be double or triple the size of Amazon – closet competitor) but also is known for its own content such as House of Cards and Orange is the New Black.
The company is continually attempting to enhance its recommender systems. Back in 2006, Netflix announced a machine learning and data mining competition worth a $1M prize with the goal to crowdsource a movie recommendation algorithm to deliver a 10%+ improvement in prediction accuracy over the existing system, which was finally awarded in 2009. The company did enact many of the recommendations such as the winning solution called the Feature-Weighted Linear Stacking to combine predictions from multiple predictive models (using items such as popularity, predicted rating, correlation to those with similar interests, past ratings, perceived gender, age, etc.) to produce final recommendations. While Netflix has continued to spearhead competitions, today the company cares much more about the user experience, satisfaction, and retention. To continually improve these metrics, the company harnesses various algorithms across identical subsets of populations to measure differences in metrics. In addition, they fully understand that the UI is the sole channel to interact with a user, which means it must be a fluid process for the data and algorithms to be updated and altered.
Netflix’s superior machine learning capabilities are likely why they have one of the lowest churn rates in the industry. According to a recent RBC survey, churn rates were near record lows with almost 75% of those polled saying they were “not at all likely” to cancel Netflix subscriptions. Netflix has continued to be disrupting the cord-cutting and entertainment industry, but can data continue to be the way forward? Will their new focus on content creation mixed with metadata drive enough new users and retain existing users? The outlook is very rosy for Netflix but with Amazon and others such as Apple and Google who can leverage their existing tech skills and pour tons of money into the TV space, Netflix cannot afford to take any rest breaks.