Long read: Bland, easy to follow, for fans of everything: what has the Netflix algorithm done to our films?

This is an entertaining and insightful long read about the business of movie making and how Netflix is profoundly changing the way movies are made. In order to help us understand why Netflix is a big deal, Phil Hoad of The Guardian, gives us a bit of context:

“Among the streaming companies, Netflix is by far the most successful. It now has 301 million subscribers worldwide – 100 million more than its nearest competitor, Amazon Prime Video. Releasing more than 100 film “originals” a year, it is more prolific than even the Hollywood studios in their Golden Age peak. It expanded in the 2010s from its US base into nearly 200 countries and operates as a monolithic global distributor of entertainment. While some of its content is purely local, it also aims to select the most promising titles throughout the world and make them available internationally (as happened with the TV show Squid Game and the 2022 Oscar-winning German adaptation of All Quiet on the Western Front).

Netflix’s model, and its enormous success, gives it unprecedented influence over cinema’s future.”

A significant part of Netflix’s success arises from the fact that the firm has understood, defined and classified different aspects of a movie at a level detail no one else had bothered to get into:

“In the late 00s, Netflix’s then director of personalisation, Todd Yellin, set himself a trifling task: completely redefining the taxonomy of how films and TV were classified. He was a dedicated cinephile and director who had made a well-received debut feature in 2006, the Brooklyn family drama Brother’s Shadow. But working at Netflix gave him an opportunity to flex other skills. “I also have a mathematical side to my brain,” he said, “so I thought if you subdivide movies and TV shows into their constituent parts and tagged them accordingly, would that help put the right title in front of the right person at the right time?”

This was his plan for refining Netflix’s recommendation system; the process by which content is sorted and mathematically weighted in order to give individual users the most pleasing selection. Often referred to as “the algorithm”, it actually involves 10 or more interlocking ones.

After putting his toddlers to bed, Yellin would sit down in an old chair and raid his library of cinema books for ideas about how to classify content. He quickly went beyond the repertoire of traditional genres – horror, comedy, thriller – to begin tagging titles by subject-related criteria: “Is it about dancing? Architecture? Marital relationships? Then we’d look at emotions – how dark is it?” For tonal matters, he and his team assigned values from one to five or one to 10.

A new job position – “tagger”– was created to watch and classify Netflix content. Yellin remembers it as painstaking work. He and his helpers eventually devised what in 2014 amounted to 77,000 “altgenres” (there are very likely more now): the categorisations that also, depending on what the algorithm serves you, appear on the Netflix homepage as row labels, the categories of films you’re offered. They run from the blandly familiar (“Adventure films”) to the slightly more specific (“Relentless crime thrillers”) to the infuriatingly broad (“Feel all the feels”). And then of course there’s the “Casual viewing” supposedly rotting everyone’s brains, the likes of The Electric State or Red Notice, an action-comedy that’s a mashup of James Bond, Indiana Jones and Fast & Furious.

Sometimes these row labels are automatically generated, based on underlying relationships between the altgenres revealed by machine learning. Every user is assigned a mathematical “distance” to each altgenre, based on how much or how little they interact with them on the platform. Aggregated across millions of users, this web of consumption patterns reveals unexpected correspondences, overlaps and affinities in viewers’ tastes. One example was the overlap between audiences who liked Formula One and classic rock’n’roll documentaries; in that instance, the recommendation system might generate a category that combined the two.

This deeper data architecture was a gamechanger for Netflix. Originally, the service had generated recommendations based on a five-star system of user ratings, but in 2017 Netflix abandoned this in favour of the altgenre-based system. “Moving from explicit to implicit recommendations was the big shift,” said Yellin. “Recommendations based on behaviour – what you actually watched and consumed, versus what you said you liked.””

Netflix is uniquely well placed to use new classifications and taxonomies to slice & dice a movie in ways never tried before:

“In 2017, Netflix logged 700bn “data events” – interactions with the platform in some form – per day. Not just whether you opted for something in “So completely captivating” or a sports documentary – but what device you watched it on, what time of day you were viewing, how many other titles you lingered over, whether you turned something off early, how many times you rewatched, and on and on. All nodes in the galactic data cloud the services use to decide what films and TV shows to put in front of us.”

The result of these two forces – the first is detailed taxonomies & classifications and the second granular data on viewers’ behaviour – is a very different style of movie making to what Hollywood & Bollywood is accustomed to:
“…Netflix likes to move fast. Within five seconds, to be precise – this, according to the pitch workshop document they hand out to potential collaborators, is the length of time within which the “audience subconsciously decides whether they will watch your show”.

A swift and unambiguous opening is a non-negotiable for the company; most of the film-makers interviewed for this article mentioned it. Screenwriter Aron Coleite was brought on to punch up the 2024 sci-fi film Atlas. His draft originally opened with the film’s star, Jennifer Lopez, interrogating the severed head of a robot terrorist. It was deemed too left-field and Coleite ended up replacing it with a more conventional Swat team raid intro; he was swayed by Netflix’s data demonstrating that viewers need to be hooked within a certain window. He feels that window is getting shorter: “I see it shrinking as attention spans are harder to corral.”

At Netflix, specialist strategy and analysis teams are embedded within every division of the business. The strategy and analysis team in the content division helps value a prospective new title – whether acquired or developed in-house – by modelling its performance based on historical data. The company has been doing this a long time: there are talks available on YouTube going as far back as 2016, in which Caitlin Smallwood, then head of science and algorithms at Netflix, details how a film’s predicted success evolved according to new elements added during pre-production, such as certain actors coming on board, or the reaction on social media to a teaser trailer. (Netflix later clamped down on this kind of disclosure, afraid it might be interpreted as the algorithmic adulteration of art.)

According to Smallwood, this process went as far as assessing pitch decks or scripts for elements that might boost or reduce their appeal. Director Cary Fukunaga mentioned a complex narrative structure in his 2018 big pharma miniseries Maniac being nixed because of the audience loss predicted by the data. “The algorithm’s argument is gonna win at the end of the day,” he told GQ….”

Note: Amazon is part of Marcellus’ Global Compounders Portfolio, a strategy offered by the IFSC branch of Marcellus Investment Managers Private Limited and regulated by IFSCA. Accordingly, Marcellus, its employees, immediate relatives, and clients may hold interests or positions in these stocks. Any references to these companies are made solely for informational and educational purposes, in the context of the article discussed.’

If you want to read our other published material, please visit https://marcellus.in/blog/

Note: The above material is neither investment research, nor financial advice. Marcellus does not seek payment for or business from this publication in any shape or form. The information provided is intended for educational purposes only. Marcellus Investment Managers is regulated by the Securities and Exchange Board of India (SEBI) and is also an FME (Non-Retail) with the International Financial Services Centres Authority (IFSCA) as a provider of Portfolio Management Services. Additionally, Marcellus is also registered with US Securities and Exchange Commission (“US SEC”) as an Investment Advisor.