(CS320) - Value of Data & AI

Lecture 2: Steve Eglash1
Disclaimer: These notes have not been subjected to the usual scrutiny reserved for formal publications.
They may be distributed outside this class only with the permission of the Instructor.
2.1 Goals of the Session
The subject of today’s class is an examination of how Netflix, the streaming video content
company, uses recommendation data and algorithms to create value. Netflix uses a variety
of different technical approaches, primarily for content recommendation and search, and
there are distinct negative and positive consequences to these approaches and their use at
Netflix’s scale. We will explore these consequences and then try to generalize our
observations to other companies to understand the impact of recommendation algorithms
across a variety of industries and sectors.
2.2 Netflix: Background Information
Netflix is, today, one of the largest and most successful entertainment and media
companies; it is the seventh largest internet company by revenue in the world. Netflix’s
primary business model is a subscription streaming service, and the company has 150
million subscribers and $19 billion in annual revenue generated from monthly fees. The
use of this subscriber model, as opposed to the ad-based revenue model that is much more
commonly seen among data-based internet companies, makes Netflix somewhat unique
amongst its peers.
Netflix’s biggest cost is acquiring and/or creating content for its streaming platform. It is
important to understand the costs at any company that we are examining, so that we
better understand the specific challenges and opportunities that the business faces—we
will do this for each company observed during this course.
Initially, when Netflix launched in 1999, the company provided a mail-based DVD rental
service. Over the course of the 2000s and 2010s, Netflix transitioned to providing streaming
CS320: Value of Data & AI Winter 2020
Session 2
Lecturer(s): Steve Eglash Scribe(s): Susannah Shattuck

Lecture 2: Steve Eglash2
content online, and it eventually discontinued the DVD-based portion of its business.
Today, Netflix has transitioned again to becoming a major producer of new content, having
developed and launched X shows since X date.
Each time Netflix’s business model evolved, the company faced new opportunities and
threats in the market. From the start, however, working with data and algorithms has
been a critical component of their business. We will examine how Netflix’s evolving use of
customer and third party data has informed its position in the market as a viewing
platform for movies and television shows.
2.2.1 Netflix’s Goals
Today, Netflix’s executives might say that the goal of the company is to become the largest
producer and distributor of television shows and movies. In order for Netflix to succeed at
this goal, the company needs to be able to offer engaging content to its subscribers who
come to the company’s website for content. A corollary goal to this primary goal is to keep
those subscribers on the website without getting distracted and dropping off to another
website or source of entertainment.
2.2.2 Netflix’s Challenges & Advantages
There are several critical challenges that Netflix faces in pursuit of its primary business
goals. One key challenge is streaming quality and the required supporting infrastructure;
consumers today are increasingly intolerant of slow or poor quality video streaming, and
Netflix has had to invest a lot into its streaming infrastructure to support high quality
and fast performance.
Another key challenge is that consumers are fickle. A subscriber visiting Netflix’s website
at any given time may be in a specific mood or have a specific need that is different the
moods or needs indicated in that same user’s past behavior. Being able to provide relevant
content recommendations for its users, even as their desires are constantly changing, is a
critical challenge that Netflix must solve in order to be successful in its primary goal.
One of the advantages Netflix has in the face of these challenges and increasingly tight
competition in the video streaming marketplace is that it knows a lot about consumer

Lecture 2: Steve Eglash3
preferences for movies and television shows. Its access to data on these consumer
preferences is a competitive advantage—and this is often the case for data-driven internet
companies.
Other media groups are starting to catch on to Netflix’s advantage when it comes to
consumer data. The relationship that Netflix has with competitors in the content creation
space has evolved over time as the company’s business model has evolved; initially, content
creators viewed Netflix as a way to access customers. As the dynamics of the media
streaming industry changed, however, and Netflix moved into the content creation
business, those initial partners started to view Netflix as more of a competitor. Industry
dynamics are constantly evolving across the entire market, and data-driven internet
companies are often key contributors to those shifting dynamics.
2.2.3 Netflix’s Stakeholders
Netflix is beholden to a variety of stakeholders as a business; to understand the challenges
and advantages of their business model, it’s important to examine who those key
stakeholders are and what their interests and needs are. Netflix’s primary stakeholders are
its customers, the subscribers who pay for its streaming service—and these will be the
focus of this class’ discussion.
Netflix is also beholden to its shareholders, as any publicly traded company is. Another
key group of stakeholders are independent producers of content. This group values their
relationship with Netflix particularly strongly, because Netflix provides them access to a
wider audience in a way that traditional movie theaters and television might not.
On the infrastructure side, Netflix is beholden to ISP providers and must work closely
with this group to deliver content more quickly and at high quality. Infrastructure for a
streaming company like Netflix is particularly intensive—a large percentage of the world’s
internet traffic is related to video streaming.
Finally, Netflix must continue to attract and retain another key group of stakeholders: its
employees. Its competitive advantages in the market are dependent in part on its ability
to attract the best developers, engineers, and researchers to continue to develop cutting-
edge approaches to the big technical problems it has to solve to achieve its goals. These

Lecture 2: Steve Eglash4
problems include not only infrastructure challenges but also the development of
recommendation and search algorithms.
2.3 Recommendation Systems At Netflix
Netflix’s success is dependent upon their ability to serve content to subscribers such that
those subscribers stay engaged with Netflix as a platform. Netflix creates value by
recommending personalized content to subscribers; data and recommendation algorithms
are the driving forces behind this capability and, ultimately, the company’s entire business
model.
2.3.1 Data Sources
Netflix relies on a diverse range of data sources to generate its personalized
recommendations. Key data sources include:
• Star ratings—Netflix subscribers can rate movies or television shows based on a
five-star system, and each subscriber thereby builds a personal profile of preferences
as they rate movies and television shows that they have seen;
• Content metadata—each movie or television show has a set of metadata that
provide relevant information about the genre, cast, director, etc., which can be
relevant for making a recommendation to a subscriber;
• Users’ content queues—before Netflix switched primarily to streaming content,
when it was mailing DVDs to subscribers, each subscriber had the ability to
maintain a queue of movies and television shows they would like shipped next to
them, and that queue could be a source of data about individual preferences;
• Source of content discovery on Netflix’s website—subscriber behavior on
Netflix.com can be used as a source of data, including where on the website a
subscriber clicked on a particular piece of content;
• The content—a movie or television show is a collection of data points in and of
itself, and that data can be used to recommend “other movies with car chase
scenes,” for example;
• Publicly available third party data—other relevant data, such as critics’ reviews
and social media response, may be used by Netflix to make recommendations;