Netflix: Optimizing the user experience and investment decisions using Machine Learning

Netflix, most people’s first paid streaming provider, completely revolutionized how we consume video entertainment content. At this point in your HBS experience, mere weeks before graduating dear reader, you have probably read a case about Netflix in at least a third of your classes including FRC, FIN1 and our favorite DIT. This level of coverage is entirely deserved given how successful the company has been, how culturally significant it has become in reshaping the video entertainment experience, and most importantly how innovative the company has been.

We all know how Netflix creates value, by charging monthly subscription fees from users to give them access to their gigantic content library. I will not belabor the point in recognition of the audience’s high degree of familiarity with Netflix’s background and will instead zoom into how machine learning (ML) is used to create and capture value.

Netflix ML Value Creation and Capture:

Netflix was an early adopter in the tech space in collecting huge amounts of data on user behavior including “the time and date a user watched a show, the device used, if the show was paused, does the viewer resume watching after pausing? Do people finish an entire TV show or not, how long does it take for a user to finish a show and so on [1].” This data is collected on all 150M+ paying users resulting in a huge amount well-structured information that can be meaningfully deployed. This information is used to inform Netflix’s famous “recommendation algorithm,” which Netflix claims is responsible for over 80% of its streaming duration [1][2].  This algorithm uses machine learning to predict what you may enjoy watching based on your behavior on the platform, your searches and interests, what people who have similar tastes also watch etc. Netflix claims that this customization algorithm is responsible for the low churn rate compared with their competitors (<10% for Netflix, compared with over >15% for Disney+, and >20% for Apple + and HBO Max) [3]. Churn is one of the most sensitive inputs for the valuation of a subscriber-based business and thus this customization algorithm is a huge part of Netflix’s success. For example, if I were to search Invincible, amazon Prime’s new animated comic series (which I could not more strongly recommend), I would see results of animated, comic book, and hero movies and shows. The algorithm has been trained to relate that search item with similar content as well as my watch history (mostly sci fi and superhero mythology).

Beyond the recommendation algorithm, Netflix uses AI/ML to enhance their product in several other ways. Firstly, based on the historical “click” data generated by users, Netflix can predict which thumbnails, the pictures attached to a movie or show title, will work best on a specific user to increase customer engagement [4]. Another example is using historical streaming data to better under predict and manage streaming demand which has the effect of increasing streaming quality and improving the customer experience [4][5]. What that means in practice is that Netflix caches data in servers closest to the users who likely will want it to improve load times. From the standpoint of the customer facing product, these uses of AI/ML improve the experience and maintain customer retention which drives down the churn rate thus creating and capturing customer value.

Pathways to a Just Digital Future

Watch this tech inequality series featuring scholars, practitioners, & activists

Switching gears to the “upstream” uses of ML/AI, specifically in content creation as opposed to the “downstream” delivery mechanism, Netflix continues to leverage their data collection to improve decision making. For example, based on the data collection and use of ML/AI predictive abilities, Netflix can size audiences for types off content [6]. Put another way, predictive modeling can better inform decisions around investing in content creation to optimize the bang for their content creation buck.

Outlook: Challenges and Opportunities

The biggest challenge facing Netflix is the threat from new entrants ranging from Apple+, Disney+ and HBO Max, to Amazon Prime and Hulu. The streaming wars are a reality. This is especially a concern when considering that HBO and Disney are consistently some of the premier content creation studios. However, Netflix has entered the content creation game, and is using ML to be more efficient with content creation investments. The biggest opportunity Netflix has at present is their strong retention and brand name. If they can produce comparable content to their streaming rivals, they may continue to lead, however, if they produce mediocre and forgettable content, a popular criticism, then they may suffer from a slow steady decline that no amount of data collection or ML can prevent.

Sources:

1. Dixon, Michael. 2021. “How Netflix Used Big Data And Analytics To Generate Billions – Selerity”. Selerity. https://seleritysas.com/blog/2019/04/05/how-netflix-used-big-data-and-analytics-to-generate-billions/.
2. Chong, David. 2020. “Deep Dive Into Netflix’S Recommender System”. Medium. https://towardsdatascience.com/deep-dive-into-netflixs-recommender-system-341806ae3b48.
3. Adgate, Brad. 2021. “A Challenge For Video Streamers Will Be Keeping Subscribers”. Forbes. https://www.forbes.com/sites/bradadgate/2021/01/15/a-challenge-for-video-streamers-will-be-keeping-subscribers/?sh=20eda709202b.
4. “How Netflix Uses AI And Machine Learning”. 2020. Medium. https://becominghuman.ai/how-netflix-uses-ai-and-machine-learning-a087614630fe.
5. Ekanadham, cHAITANYA. 2020. “Using Machine Learning To Improve Streaming Quality At Netflix”. https://netflixtechblog.com/using-machine-learning-to-improve-streaming-quality-at-netflix-9651263ef09f.
6. Dye, Melody, Chaitanya Ekanadham, Avneesh Saluja, and Ashish Rastogi. 2020. “Supporting Content Decision Makers With Machine Learning”. Netflixtechblog. https://netflixtechblog.com/supporting-content-decision-makers-with-machine-learning-995b7b76006f.