In this abridged extract from new book ‘What to do when Machines do Everything: How to Get Ahead in a World of AI, Algorithms, BOTs and Big Data’, authors Malcolm Frank, Benjamin Pring and Paul Roehrig examine Netflix, as a great illustration of a company that has leveraged systems of intelligence, created a business model oriented around them, and upended business as usual.
As of 2016, Netflix represented about 35% of all internet traffic in North America, and about 75 million subscribers worldwide, and has the incumbent TV networks in quite a spin. By dissecting Netflix, we can see the anatomy of "the new machine" in action, and how it can deliver superior customer experiences.
The anatomy of the Netflix system of intelligence
Anatomy element: the app
Most of us experience Netflix as an app. Regardless of where we are or the hardware we use—tablet, mobile device, laptop, set-top box, or VR headset—the only part of the Netflix system of engagement we touch is the app that connects us to Orange Is the New Black (and many more).
Anatomy element: process logic
Do you remember using the print TV Guide to find out what was on television and then having to time your life around when a show was airing? In those bygone days, there was really only one process for consuming media via television or movie screen, and it wasn’t great. Now, of course, whether we are watching on a mobile device on the road or next to our loved ones at home, Netflix provides content in an almost infinite set of variations. It seems so simple now. But it feels seamless and frictionless because Netflix has thoughtfully adapted its entire system to the way we want to consume content.
Anatomy element: machine learning
Netflix’s recommendation engine—basically, a set of algorithms that connect us to content we want—is the most well-known element of the Netflix AI system, but the whole engine is actually much more than that. The core of the Netflix experience connects us to content, distinguishes between family members, and processes billions of events a day related to movies, viewers, payments, and the like. More important, the platform improves over time. As we use it more, it learns about our tastes and serves up the best content available in a highly personalized way. It can distinguish between what we say we like and what we actually like. (Note, for example, that Adam Sandler’s The Ridiculous Six is rated a middling three stars by Netflix users, but in January 2016 was the most watched Netflix program in history.) The core of the Netflix system is a remarkable piece of software design and engineering, which, because it is so good, is nearly invisible.
Anatomy element: software ecosystem
Netflix relies on connections to dozens of other systems to bring us Orphan Black and The Walking Dead (two of the most binge-watched shows). The Netflix team emphasizes open-source software tools such as Java, MySQL, Hadoop, and others. Content distribution is supported by tools like Akamai, Limelight, and Level 3 Communications. It also relies heavily on Amazon’s cloud systems for storage. The point here is not the specific tools Netflix uses but that the company realized early on that it would need to build certain elements of its overall system and that it also could leverage other best-in-class systems to grow faster.
Anatomy element: sensors/Internet of Things
Netflix engages with us—both providing content and collecting data—via our devices, but it also learns about us via the many sensors and data associated with these devices. Other sensors are starting to matter to Netflix. For example, it is exploring ways to connect to Fitbits and even socks to monitor if we’ve fallen asleep.
Anatomy element: data
The Netflix data warehouse stores about 10 petabytes of information. With all that data flowing through the AI engine, Netflix knows you. We’re sorry if that creeps you out, but it tracks the movies we watch, our searches, ratings, when we watch, where we watch, what devices we use, and more. In addition to machine data, Netflix algorithms churn through massive amounts of movie data that is derived from large groups of trained movie taggers.
Anatomy element: systems of record
There’s no way to manage the vast reams of Netflix data without an absolutely top-notch architecture. Netflix started with Oracle, but has moved over to an open-source database called Cassandra. It uses Hadoop for data processing and Amazon S3 for storage. It also links to back-end payment systems so nothing can interfere with our next purchase.
Anatomy element: infrastructure
As of February 2016, all of the infrastructure services Netflix requires are provided by Amazon Web Services.
So.... what are the attributes of a successful system of intelligence?
There’s a big difference between merely having all the necessary ingredients of the new machines and actually getting them to perform at a high level. A system of intelligence that can help you be the Usain Bolt of whatever race you’re in will have all or most of these characteristics:
- Smart, not dumb: One true test of good AI is that as data flows in, it’s smarter tomorrow than today. In every case, the best systems of intelligence suck up data from a wide variety of sources, which helps achieve the data mass required to derive insights and create personalized experiences.
- Open, not closed: Systems of intelligence that achieve their full potential are generally more open than closed. Think of Tesla giving away its patents and Uber with its open APIs; both of these policies have helped generate new solutions built on the companies’ AI cogs and gears.
- Smart hands, not just bots: Most successful ANI systems include people. There are lots of things that machines simply cannot do as well as we can (at least within the timeframe of a meaningful business decision). Game-changing systems of intelligence are built around the integration of AI with humans by combining the best of what computers do with the best of what humans do.
- Narrow, not broad: Even complex platforms like IBM’s Watson or GE’s Predix offer significant business value only when they are configured for a specific process or customer experience. Focusing on a “moment” or a simple activity is one of the best indicators that a system of intelligence might have meaningful impact.
- One-to-one, not one-to-many: While most existing enterprise systems are generally not designed to give us an individualized experience, systems of intelligence most definitely are. Alexa, for example, self-configures to your patterns. Amazon tunes its offerings to your virtual self.
This is an edited extract from What to do When Machines do Everything: How to Get Ahead in a World of AI, Algorithms, BOTs and Big Data, by Malcolm Frank, Paul Roehrig and Ben Pring (Wiley, 2017).
ABOUT THE AUTHORS
Malcolm Frank is executive vice president of strategy and marketing at Cognizant, a global technology consultancy of over 250,000 employees, focusing on sharpening the company’s corporate strategy and global brand. With more than two decades of experience driving digital transformation, Malcolm is a frequent speaker and industry thought leader on IT management issues and was named “one of the most influential people in finance” by Risk Management Magazine. He graduated from Yale University with a degree in economics.
Benjamin Pring is global managing director of Cognizant’s Center for the Future of Work, a thought-leadership team that helps clients leverage opportunities being created as new technologies to make computing power more pervasive, more affordable and more important. Ben wrote the industry's first research notes on cloud computing while a research vice president at Gartner. He graduated from Manchester University with a degree in philosophy.
Paul Roehrig leads strategy and marketing for Cognizant Digital Business. He founded the Cognizant Center for the Future of Work, and was formerly a principal analyst at Forrester Research. Paul also held key positions in planning and implementing global technology programs in a variety of industries. He has a degree in journalism from the University of Florida and holds graduate degrees from Syracuse University.
Malcolm, Ben and Paul also wrote the award-winning Code Halos: How the Digital Lives of People, Things, and Organizations are Changing the Rules of Business (Wiley, April 2014).
