Netflix Uses Machine Learning to Cut Costs and Retain Customers

Netflix is an on-demand video subscription service offering unlimited viewing for a monthly fee in virtually every market in the world except China. The company currently has over 137 million accounts across these markets in which it competes with other online streaming services and linear TV networks, [3]. As Netflix seeks to continue to grow across markets in this increasingly competitive entertainment space, it faces several challenges that almost necessarily require the use of machine learning to address. These challenges include, but are not limited to, increased consumer demand for personalized content, growing diversity of customer needs and preferences in content, and tightening competition on content licensing.

To satisfy an increasingly broad set of customer tastes in a cost-effective and personalized way for its diverse customer base, Netflix needs robust analysis of an incredible amount of data, which can only be conducted with artificial intelligence. Netflix’s management is currently implementing the use of machine learning in the areas of content acquisition and production, and customer experience, which includes content recommendations and streaming quality.

The company has seen that even its largest titles, which are viewed by tens of millions of people, account for a very small percentage of overall streaming hours so it understands that it is the combination of pieces of content that attract and retain customers. Since tastes are broad, even within a single market, Netflix offers a wide breadth of programing in order to maximize the size of its customer base [4]. To satisfy this broad range of customer tastes in a cost-effective way, Netflix uses machine learning to determine expected hours of viewing for each piece of content, estimate the cost per hour viewed, and compare it with that of similar content deals [7]. Additionally, the firm uses predictive models to understand customers, such that there is a large enough set of content that meets their preferences without necessitating the renewal of any specific title [7]. This cost-effectiveness is particularly important as increased competition bids up licensing and renewal agreements.

To further reduce reliance on outside studios and ideally reduce content costs, as well as strengthen brand loyalty, Netflix has focused on building out its own original content [4]. This content production is another area in which Netflix leverages machine learning [5]. The company uses machine learning not only to identify potential projects its customers will like, but also to make cost-effective business and technical decisions. In the pre-production decision process, Netflix combines and analyzes various data sets to predict the cost of numerous attributes of the production process, such as content, location, and schedule, and optimizes decisions given resource constraints such as time, cast, and locations. To extract as much value as possible from these production investments, Netflix tries to make content accessible to as many viewers as possible. In order to do this, Netflix needs to prioritize markets given the previously mentioned constraints. Machine learning informs this strategy by predicting which languages a piece of content will me most popular in, in which locations, and amongst what groups [5].


In addition to having the right content to acquire and retain customers, Netflix needs to deliver the right content to the right customer. The firm uses machine learning to address these challenges. To direct customers towards the content they want to watch, Netflix relies on a recommendation engine that uses “implicit and explicit” data to identify trends. Implicit data is information that users share with Netflix, such as like/dislike feedback, and explicit data is behavioral data that includes which shows users watched and how fast and frequently they watched those shows. The engine recommends shows to different users for very distinct reasons based on identified tastes [6]. In an effort to further personalize the user experience and maximize successful recommendations, Netflix uses machine learning to match the tastes of a customer to the form in which content is recommended to them. This makes it so that different customers are not only promoted the same content because of distinct reasons, but also promoted the same content in distinct ways. Netflix does this primarily by promoting, for the same content, different images that appeal to different preferences [1].

In addition to what Netflix is currently doing, the firm needs to ensure that it is searching for new data sources to refine its algorithms. Competitors like Amazon have data unrelated to video that can inform their recommendation models. Netflix should seek to partner with companies like Data Wallet to acquire this additional data. Netflix should also strive to seek causality with its algorithms and ensure they are not self-reinforcing to eliminate bias.

How can Netflix ensure its algorithms identify causality?

(784 words)


[1] Foreman, Holden. “Netflix Machine Learning Director Talks Personalization Software.” The Stanford Daily. October 02, 2018. Accessed November 13, 2018.

[2] “How Netflix Uses AI to Find Your Next Binge-Worthy Show.” The Official NVIDIA Blog. June 01, 2018. Accessed November 13, 2018.

[3] “Long-Term View.” Netflix – IR Overview – Profile. Accessed November 13, 2018.

[4] “Netflix Q3 2018 Shareholder Letter.” Netflix Investors. October 16, 2018. Accessed November 13, 2018.

[5] Netflix Technology Blog. “Data Science and the Art of Producing Entertainment at Netflix.” March 27, 2018. Accessed November 13, 2018.

[6] Plummer, Libby. “This Is How Netflix’s Top-secret Recommendation System Works.” WIRED. August 21, 2017. Accessed November 13, 2018.

[7] “Top Investor Questions.” Netflix – IR Overview – Profile. Accessed November 13, 2018.


Pinterest uses machine learning to expand their offerings for consumers and marketers


Chasing Medicare Fraud with Machine Learning

Student comments on Netflix Uses Machine Learning to Cut Costs and Retain Customers

  1. Thanks for such an interesting dive into Netflix’s use of machine learning. I appreciate the multitude of ways Netflix is applying machine learning. They have definitely been ahead of the industry in evaluating consumer content preferences and using those insights to evaluate content acquisition or production decisions.

    I wonder if there are more ways Netflix should be applying machine learning to other areas of their business. For example, in addition to the quantitative insights that it gains from consumers’ like/ dislike feedback, I wonder if they could also gather qualitative feedback from viewers about their shows. That qualitative feedback could still be mined using machine learning, and would compliment the quantitative data. I believe the combination of quantitative and qualitative could potentially further refine the algorithm and begin answering the causality question you raised.

  2. Great overview of how Netflix is leveraging machine learning, and especially liked your point about AI application to market development/penetration. Regarding leveraging the technology to make content development decisions, however, I wonder how restrictive such an approach (using, presumably, historical data) could be on the content they end up creating. To borrow an ideology from the fashion world, customers don’t set the trends, fashion houses do. There is always an inherent risk of using historical data to predict future outcomes, but even leveraging the predictive capabilities of machine learning may leave a striking amount of creativity, innovation and non-linear decision making that could result in exceptional profitability and success unexplored.

  3. Enjoyed reading about machine learning in Netflix. In addition to customer retention and engagement, which are obviously critical, is Netflix also using machine learning to look into customer acquisition? For example, could it analyze existing movie/TV watching habits, advertisement responses, or other trends to determine how and where to best advertise particular shows? Finding external data to drive these changes may be challenging or expensive but could be instrumental to Netflix’s growth, especially as it nears saturation in certain countries.

  4. Interesting article! It was fascinating to see the various ways in which Netflix uses machine learning to acquire or produce content, and serve its customers. One thought I had reading this would be to understand to what degree are the datasets informing Netflix’s algorithms proprietary versus external. It is clear from the essay that Netflix has been successful at applying machine learning (in many ways) to conduct business, but the above inquiry would further indicate the relative advantage Netflix may have in doing so in comparison to competitors. The underlying assumption here being that external data, being publicly available, is not a source of differentiation, but proprietary data may be.

  5. Interesting article! Your question about striving to not self-reinforce bias in content is a great one. The main issue I see with Netflix’s use of machine learning is that AI can’t deduce or create data from something that doesn’t yet exist (whether it be a genre, new type of show/movie, more diverse casting or characters, etc.). I hope this is something Netflix is taking into account when using their existing data to determine which new shows/movies to green-light. You also brought up an excellent point that some of Netflix’s competitors have access to lifestyle data beyond just viewing data. I imagine there is a correlation here as well, the question in my mind is this: is there enough value in additional purchasing and customer behavior data to justify Netflix looking at data outside of their own compiled information?

  6. Very interesting article! Reading through all the technological advances in Machine Learning and predictive modeling that Netflix is leveraging to produce its own content and cater to a large audience, I am curious to see where this trend will take entertainment in the 10 to 50 year timeframe. One potential issue that I see with companies like Netflix, Amazon and probably soon Disney leveraging data more and more, is that original content gets produced more or less as a direct aggregation of user preferences. As companies seek to cut their costs and appeal to as broad of an audience as possible, will we be left with mind-numbing, crowd-pleasing entertainment? Looking at the type of movies and shows that appeal to the widest audience, I am worried that we will stop producing independent, thought-provoking content that does not correspond to the vast majority of user preferences. Another risk we run is to re-create the Facebook “echo chamber” in the entertainment space. If I am only ever exposed to content that I am naturally inclined to watch, how can I get a more complete view of reality and learn about people, ideas, and societal events that may not be “pleasing” for me to watch?

Leave a comment