Why Netflix Won at Personalization - and How Personalization Helped Them Win

Netflix turned 25 this week - and over this time they have become the masters of personalization while creating new categories and disrupting multiple existing industries from the neighborhood movie store to traditional TV & cable models. Over this time, they’ve consistently engaged users (including me!) with high quality and relevant recommendations - all driven off Netflix’s vast datasets that include incredibly rich views of their customers. And while high quality personalization has been a Netflix staple, their trajectory as a business is anything but, having undergone multiple significant pivots over the past two decades plus.

I remember in high school and later in grad school getting these little red packages with DVDs in the mail and trying to understand their business model. Effectively you could have three DVDs out at a time. Watch one, send it back to Netflix, and then you can choose another one on their website. Netflix paid for two-way shipping - the price wasn’t per movie but a monthly recurring subscription cost.

A decade after its initial launch in 2007, it also started streaming movies. Their CEO Reed Hastings realized that this is where the world was heading and even went so far as to try to move the DVD service under a different business name (“Qwikster”), but that was quickly kabbashed. 

And then, another half-decade later, Netflix made its most recent transformation to one of producing original content. Streaming as technology was becoming commoditized, and competitors were showing up. Netflix had a hypothesis that they could create content that was designed specifically for their audience and their platform. And their initial launch was a smash hit with “House of Cards” in 2013. Binge-watching was officially a thing.

Over the past 25 years, Netflix went through two major business pivots - a highly unusual feat. 

So, how did Netflix win?

The short answer is that it all comes down to data and their ability to build killer personalizations that drive binge-watching and engagement that today has become synonymous with the Netflix experience.

From day one, Netflix has always had a subscription model. They’ve never charged on a per-movie or per-show basis, always on a per month, “watch as much as you can basis.”

And with this value exchange, Netflix understood that if their customers aren’t watching a sufficient number of shows per month, they’d simply cancel their subscription. Unlike their initial set of rivals - Blockbuster Video or the local video stores - having great content and a comprehensive movie selection wasn’t sufficient. Netflix users weren’t driving to a store to browse thousands of movies with desperate hopes of finding one to watch tonight.

Netflix turned this model on its head. And today, when a user opens up Netflix on their iPad, they’re presented with a few rows of highly personalized recommendations out of the thousands of possible shows that Netflix has in its inventory.

Netflix turned this into a science because, to win over the past 25 years, they realized they had to become extraordinarily good at this. The combination of their pricing model, their new form of content delivery, and their massive scale today set the stage to build the best single example of personalization anywhere today.

And Netflix isn’t just the best at personalization - they won largely because of their personalization.

So, how did they do this and build such an amazing machine? 

See my thoughts below - it’s much more than some smart algorithms and data scientists building cool machine learning.

They Use Data to Drive both Personalization & Inventory

In 2011, Netflix started seeing tremendous success and engagement by licensing shows like Madmen from networks like AMC producing original content. Yet, despite their business growth at the time, they realized that their window of opportunity around licensing would be short-lived as rivals started developing their streaming capabilities and services.

While their pivot to original content provided them with better competitive moats around proprietary content, it was also a game changer for their personalization strategy.

At a high level, central to Netflix’s personalization strategy is their recommendation algorithms. And when these algorithms fall short, they do so for one of two reasons.

The first is algorithmic: the shows that are being recommended aren’t the best ones out of the complete set of 13,000+ shows that Netflix today has in their inventory. Netflix today has a powerful AI group today that works on this problem specifically and is continually optimizing its algorithms.

The second reason is more existential: what happens if the algorithm is doing its job and finding the “best” options, but at the same time, the most relevant shows for a specific user just aren’t that interesting? For most personalization applications - and every one of Netflix’s rivals at the time - there was nothing to be done about this.

Netflix realized that their proprietary content & shows could be designed to improve personalization! They built a consumer insights team - which focuses on trends across their customer base and uses data to prioritize which shows to produce next. Effectively, the team works by looking at demand and viewing data from their personalized recommendations and then uses this data to understand what drives engagement and where the gaps are.

The disruptive nature of their strategy hasn’t come without contention. Internally there was strife in the business in aligning between high-paid “blockbuster” stars that Netflix’s content team had under contract - vs. best-performing content & strategies that their data & insights team had concluded (and many times tested) to be the most effective. 

They’ve Built a Great Data Team

Unsurprisingly, underlying Netflix’s personalization expertise lies a great data foundation. And given the proprietary nature of Netflix’s data - which today is comprised of engagement data from 220 million global subscribers watching what’s primarily exclusive content - Netflix is incredibly open about its strategy, infrastructure, and algorithms.

Netflix’s data efforts are expansive - including core infrastructure to house the data for insights, analytics, and machine learning - and also extending into all areas of the business including machine learning for better recommendations, computer vision for improving content production, and encoding research for more efficient & clearer video streaming. Netflix’s systems today process trillions of events per day - a number that demonstrates not just their business scale but also the specificity in which they’re measuring & collecting data. If each of their users logged in daily to Netflix - this amounts to almost 5,000 data points daily per user!

Netflix has truly built science around data - which is pretty clear by looking at some of the stuff they’re publishing. A recent article titled “Reinforcement Learning for Budget Constrained Recommendations” hits close to some of the academic research I did in a previous life. Setting the math aside, this paper goes deep into why simply identifying the “top 10” recommendations for a user isn’t always the best strategy from an engagement perspective. For example, a show like “Stranger Things” is top-rated and has now been out for several years may be on the top of this list. Still, it’s also likely that the user is well aware of it - and perhaps they either have no interest in watching it or have watched it elsewhere already.

Automation, Automation, Automation

At an institutional level, Netflix believes that automation is critical for its success. Before getting into examples of automation for personalization, I’ll give an example of Netflix’s insane devotion to automation in a completely different context.

Netflix was an early adopter of cloud computing - in 2008, they started moving their entire compute infrastructure to Amazon Web Services. A big point of success in their migration was ensuring that their service had a high degree of reliability and up-time “four nines” in their technical jargon (i.e., 99.99% of the time, Netflix is available to customers).

Foundational to this mission of reliability is understanding how Netflix responds when individual servers and machines go down. To test this, most businesses will schedule a day of “fire drills” where they test how the system responds to taking individual machines offline and making them unavailable. Does the site still render? Can you still watch shows? Are personalized recommendations working?

This isn’t how Netflix did it. They instead built a service called Chaos Monkey that they continually run in the background and effectively disables machines in their Amazon infrastructure. It’s a virtualized analog of someone walking through a data center and unplugging services. And it’s fully automated - and the monkey is always running around. I’ve never worked at Netflix, but I can only imagine how this acts as a forcing function to build & architect a truly resilient system.

With this as a backdrop, let’s look at how Netflix approaches its adtech & martech strategies. Unsurprisingly, their platform strategy focuses on automation and workflows that enable their customer-facing teams to best leverage their proprietary content to engage new and existing customers fully. 

• Workflow Automation: Workflows that make it easier to deploy high-quality content & digital assets. Remove friction around content packaging (image sizes, creative sizes, mediums, etc) - and allow marketing to focus on core strategic content issues.
• Channel Automation: Automated channel syndication across ads, out-of-home, YouTube, and beyond. Netflix lets the technology handle syndication and what is all too often tedious & incredibly time-consuming workaround syndicating content consistently across channels.
• Results Automation: Measure everything, and do so in a way that it automatically gets done. Netflix understands how tedious and difficult this can be - particularly the cross-functional dependencies that are sometimes required with data & analytics resources to pull reports and measure campaign effectiveness.

Their automation strategy isn’t focused on automating marketers - but instead on automating the time-consuming and workflow-ineffective processes that exist in the day-to-day of today’s marketers. Netflix recognizes that time is the most valuable resource for their marketing team. Their technology charter reflects this: “to create technology that will enable our partners to spend more of their time on strategic and creative decisions.”

Culture Starts at the Top - And Reed Hastings LOVES Data

Netflix CEO Reed Hastings once said, “If the Starbucks secret is a smile when you get your latte, ours is that the website adapts to the individual’s taste.”

And undergirding this vision is an intense focus on data. In 2006, Netflix anonymized and open-sourced their review data - and challenged the world to beat their existing “Cinematch” algorithm by at least 10%. The first person or team to exceed this mark would be rewarded one million dollars.

By allowing anyone to build a recommendation algorithm on top of their data, Netflix effectively declared this dataset the best in the world. By this point, Netflix had already won because of their data - and while an incremental 10% improvement would most certainly translate to an increase in engagement and show viewing rates - it was inevitable that Netflix would eventually get there.

After the initial prize was awarded, Netflix proceeded with a second prize that included additional information on their customers, including demographic & other anonymized user-level information. The second prize was ultimately canceled out of privacy concerns following a paper published by two of my former colleagues at the University of Texas and a class action lawsuit.

Regardless, the “Netflix Prize” represented a declaration of data excellence in a way that we’ve never seen before. As Thrillist summarized in a piece chronicling the history of binge-watching (and in their typical over-the-top fashion),  “In a way, Hastings was a tech-age Willy Wonka letting any curious hacker into his digital Chocolate Factory. Instead of a chocolate river, he offered a gushing stream of data.”

And this trillion-event data stream is what enables Netflix’s personalization today.

#netflix #personalization #streaming #data