Netflix Curated Lists
Streaming services have changed how we consume TV and music. Channel hopping through TV or radio stations has given way to consuming the content we want when we want it. This new world of on-demand media, convenient as it is, leaves us with one important question: What exactly do we want to watch or listen to?
To answer this question, every major streaming service boasts the ability to adapt to your tastes and recommend the right content at the right time. Recommendations are either a result of computer algorithms that crunch massive amounts of data in the service of guessing what content you are likely to enjoy, or they are the result of human curation, with selected tastemakers surfacing and promoting exciting content.
The music streaming world, where good recommendations are a prime differentiator between services, has shown impressive creativity in combining these two approaches. In June 2015, Jimmy Iovine unveiled Apple Music as a service where leading music experts would bring a human touch to recommendation, a feat that is beyond purely algorithmic approaches. A month later, Spotify introduced their Discover Weekly feature which was hailed by the press as the solution to human curation at internet scale.
In comparison, TV and movie streaming services, which have in the past relied primarily on exclusive content for differentiation, seem to be lagging behind and have not yet found meaningful ways to incorporate human curation into their suggestions. As part of my final project for Product School, I decided to sketch out what human curation could look like in Netflix, a product I love and use everyday.
In this article, I want to walk you through the feature development process of Curated Lists for Netflix, and provide a roadmap for future development.
Netflix Recommendations and the Moment of Truth
In 2000, Netflix introduced their personalised movie recommendation system which uses a number of recommendation algorithms. Over the years, they have invested a considerable amount of time and money on their recommendation algorithms which are a key part of their business, constantly reviewing, improving and refining them. In 2006, for example, Netflix announced the Netflix Prize, a machine learning and data mining competition for movie rating prediction. They offered $1 million to whoever improved the accuracy of their existing systems by %10. In 2014, Neil Hunt, Netflix’s Chief Product Officer reported that they employ 300 people to maintain and improve their content recommendations, and spend $150 million on recommending movies and TV shows to their members every year.
This money is well spent as even small improvements to the recommendation algorithm can lead to dramatically increased revenue by decreasing churn, the rate which customers stop subscribing to the service. Hunt explained that Netflix has a small window of opportunity to convince their audience to watch something, they call this “the Moment of Truth”. This window usually lasts between one to two minutes, and during this time, a viewer may browse as many as 20 to 50 titles before choosing to watch something, or giving up and doing something else. For Netflix, successfully delivering in this small window of opportunity is essential to business success.
Today, Netflix boasts one of the best recommendation systems, and they continue to get better, but there are a few key problems that their current recommendation engines still fail to address.
Three Shortcomings of Netflix’s Current Recommendation Systems
A Netflix recommendation is based on data that is collected when a subscriber uses the service, such as the history of watched content or ratings that the subscriber gives to movies and TV shows. For new subscribers, very little data is available and recommendation algorithms will typically fail to provide meaningful personalized recommendations. Finding good solutions to this so-called “Cold Start” problem is critical to reducing churn in the first month, when users decide whether they want to extend their subscription past the trial period.
Another major problem with recommendation engines is that there might be few or no surprises. Machine learning algorithms optimize for most hours watched (addictive content), and as Netflix’s own CPO admits, that is not always the best way to retain viewers over time. Where algorithms will tend to recommend shows similar to shows watched in the past, a recommendation by an expert can directs viewers to new and surprising content, provide them with more meaningful viewing experiences. Recommended content may subtly push us beyond our normal viewing habits while expert judgement helps us contextualize our viewing experience. Both of these factors help increase subscriber satisfaction in the long term.
Finally, automatic recommendations may still leave a user with an overwhelming amount of choice. This can lead to a long time spent browsing content, and if the user finds nothing sufficiently interesting after a while, can end in a ‘failed sessions’ where the viewer decides to do something else. Ranking choices effectively and not overwhelming subscribers with titles can help lower time-to-watch and reduce the number of failed sessions, two metrics that Netflix pays close attention to.
Netflix Curated Lists
To address these problems and to help differentiate the Netflix viewing experience, I suggest taking a cue from the music industry and introducing elements of human curation into the recommendation system in the form of curated lists. Human curation alone does not scale, but by enhancing a user’s uniquely personalized recommendation feed with curated lists that the user is likely to enjoy, we can address some of the shortcomings of algorithmic recommender systems while retaining their power and scalability.
Let’s take a look at what that might look like:
A user opens the main movie selection screen of the Netflix app. In addition to the normal feed of recommended and featured content, a few curated lists are shown. Unlike categories, lists consist of an ordered selection of movies. These lists would not change when new content arrives on the platform, but newer lists might be added. The selection of lists shown to each user would be updated on a regular basis.
Once a user selects a list, the list overview opens which provides a short description and may include a trailer-length video with the curator of the list explaining the significance of the selections or providing further guidance to help viewers decide on a movie.
Each list contains between 3 and 10 movies that the viewer can scroll through. Some of these lists, for example “Past Oscar Winners” can be ordered according to Netflix recommendation algorithms, to put the most likely to please movies first. Others like “Evolution of the Zombie Movie” will be curated in a way that incentivizes the viewer to watch them order to get the full experience, from the slow and shuffling undead in Night of the Living Dead to the relentless fast zombies of 28 Days Later.
Each movie includes a short description that explains its inclusion in the list and expands on why this movie in particular is worth watching. In addition, movies may link to trailers and additional information. Ideally, a list should be worth watching in its entirety while also easily allowing viewers to find the movie they’re most interested in.
Curated lists can also be adapted to episodic TV shows, for example, “8 X-Files Episodes You Need to Re-Watch” or “7 Columbo Episodes that Defined the Genre”.
Who is this for?
Curated lists have the potential to be a popular feature among many different types of viewers. Completionist viewers may enjoy the additional incentive to finish watching groups of related content. Novelty-seekers and movie buffs will appreciate being guided to new and interesting content that they may not have otherwise discovered. TV binge watchers may appreciate the opportunity to test drive a few selected episodes of an older show before committing to the whole thing. An important group that will benefit from curation that is especially interesting to Netflix’s business are new subscribers.
Last year, 5.6 million new users joined the Netflix platform in the US alone. Netflix’s main business objective is to retain user subscriptions, especially after the 1 month free trial. While Netflix’s overall churn is incredibly low, newer users to the platform who have just signed up for a free trial are over 200 times more likely to leave than long time subscribers. Surveys of of new Netflix subscribers have revealed that other than financial reasons, and not being interested in movies to begin with (which are difficult to influence regardless), there is a substantial subset that is unable to find the find the right content. This takes us back to a major issue with recommendation algorithms in general, the “Cold Start” problem. Unless viewers spend time rating movies and TV shows that they have already seen, it becomes very difficult to make meaningful recommendations.
Curated lists can help address these problems. Curated lists may be designed specifically for new viewers, by collecting great picks that span a wide variety of tastes together in a single place. Topical lists may be of more interest to new viewers than blind selections of content, for example, a selection of past Oscar Winners right around the time of the Oscars. Lists curated by celebrities or well-known critics may provide viewers with a personal connection to content and help them choose. Finally, in addition to simply giving viewers an alternate way of discovering and browsing content, curated lists provide an editorial platform that helps establish a more personal connection between Netflix and its subscriber base, sending a strong message to viewers that Netflix actively cares about finding content that’s right for them.
If you are interested in learning about the business benefits of curated lists for Netflix, please stay tuned, as I will add the slides I put together for Product School here soon.
I’d love to hear your take on curated lists for Netflix. If you have any comments or feedback, please leave them below or email me directly at email@example.com