[itvt] Column: The iTV Doctor Is In!: Patrick Donoghue Discusses Netflix's Whitepaper on its Recommender System

Dear Readers:  

As we get closer to the Tenth Anniversary TV of Tomorrow (TVOT) conference in San Francisco (June 7/8 2016), we have a terrific guest column from Patrick Donoghue, the industry wunderkind who altered television forever with his work at Time Warner Cable and Cablevision. Patrick is now making his services available to the entire industry through Next Stop Willoughby Inc. which he describes as “focused on making our digital lives better and more satisfying by taking away the noise and complexity and helping users get more from technology. We apply user-centric thinking to product strategy, user experience, and the content itself to do what is most important...make users happy.”
Patrick will join Cameron Johnson from Netflix, Bill Mobley from Freecast, Paul Stathacopoulos from Rovi and others on a panel titled: "TV used to be easy. What did you guys DO to it?"
I asked Patrick to comment on a recent whitepaper about the Netflix Recommender System:   http://dl.acm.org/citation.cfm?id=2843948
Here’s what Patrick had to say:
Once upon a time, there was one TV in the house and the whole family sat down on Saturday night, tuned to CBS and watched back-to-back The Mary Tyler Moore Show, The Bob Newhart Show and The Carol Burnett Show. And everyone seemed happy with his or her "choice." There were only three networks on in the early 1970s, so it wasn't difficult to choose, and life was good.
Forty years later, a "Modern Family" has replaced the Bunker family, and the world is a very different place. Today we all suffer from a wealth of options for video entertainment that includes hundreds of networks, video-on-demand, DVR, YouTube, Netflix, OTT, TV Everywhere, etc. There are thousands of shows for us to watch and ironically, we often can't find something as satisfying as The Mary Tyler Moore Show. Take a look at The Paradox of Choice: Why More Is Less by Barry Schwartz to understand the psychology. Helping viewers find a show to watch is not as simple as giving them more options.
Patrick DonoghueSmart people in the industry are getting it right. User interfaces are better. Metadata is more robust; search and recommendation engines are helping. Until recently, we all believed that recommendations would be the solution, but the magic bullet of a TV that was always ready to suggest the perfect show for viewers was a bit too idealistic. Netflix has been at the forefront of recommendation engine development for years and in 2009 famously awarded $1MM to the winner of the Netflix Prize for "substantially improving the accuracy of predictions about how much someone is going to enjoy a movie based on their movie preferences"--and their bottom line indicates it's working.
It turns out recommendations weren't enough. Recommendation engines can learn a lot about us based on what we watch, when we watch it, if we record it, rate it, watch it ten times, and many other data points. In many ways we are creatures of habit, but our moods and preferences are also impacted by other factors. Weather, news, friends, and life experiences all influence what we want to watch. According to a Netflix study, there is an "uptick of romantic video watching during Valentine's Day."
This more human approach to content discovery is an important trend. We are complex beings and even with everything that Facebook and Google know about us, they can't always know what we want. Taking a look at how we find TV shows and movies today gives us a good sense of the targeted browsing methods we have invented as users. Here is how most American adults find something to watch on TV (your sequence may vary):
  • Resume something I have already started on DVR or video-on-demand.
  • Go to the DVR. It will likely have something I want to watch because I chose to record these programs.
  • Go to favorite channels. ESPN, HGTV, Food Network, etc. are curated content that I like and are usually satisfying.
  • Browse New Releases--We all tend to like the same movies and TV shows and this is where we find them.
Understanding this process gives the industry valuable insight into how we discover and choose what we watch. It appears that the big players are adopting this user-centric approach to content display with categories such as "Just In," "Trending," and "Most Popular" making their way to the front along with "For You" and "Search." Variety, it appears, is the spice of life, and based on Netflix data, we need a lot of variety, in a quick and easy-to-consume way. "Consumer research suggests that a typical Netflix member loses interest after perhaps 60 to 90 seconds of choosing, having reviewed 10 to 20 titles (perhaps 3 in detail) on one or two screens."
What should the industry do with all of this? Is big data, artificial intelligence and multivariate testing the answer? Netflix has released an excellent whitepaper, The Netflix Recommender System: Algorithms, Business Value, and Innovation, which finally provides some much needed data on what works and how it can impact your business. A few key takeaways:
  • We have found that shorter-term temporal trends, ranging from a few minutes to perhaps a few days, are powerful predictors of videos that our members will watch, especially when combined with the right dose of personalization, giving us a trending ranker used to drive the Trending Now row
  • The recommender system is used on most screens of the Netflix product beyond the homepage, and in total influences choice for about 80% of hours streamed at Netflix.
  • 20% of viewer choices come from search.
  • The recommender system helps us win moments of truth: when a member starts a session and we help that member find something engaging within a few seconds, preventing abandonment of our service for an alternative entertainment option.
  • The recommender system spreads viewing across many more videos much more evenly than would an unpersonalized system.
  • The lift in take-rate that we get from recommendations is substantial. But, most importantly, when produced and used correctly, recommendations lead to meaningful increases in overall engagement with the product (e.g., streaming hours) and lower subscription cancellations rates.
  • Over years of development of personalization and recommendations, we have reduced churn by several percentage points. Reduction of monthly churn both increases the lifetime value of an existing subscriber, and reduces the number of new subscribers we need to acquire to replace cancelled subscribers. We think the combined effect of personalization and recommendations saves more than $1B per year.
The best news from all of this is the solid evidence that a faster and more personalized experience will result in the ever-important maximization of shareholder value. This will give the product managers and UX people of our industry the budgets they need to keep innovating and improving the experience for users.  
Patrick Donoghue