Understanding Netflix

Bryne Hobart writes:

There’s a good reason to think Netflix can, in fact, build the world’s best movie studio. They have a monopoly on the viewing data of 94m people who watch ~1.5 hours of Netflix per day. This is one of the crown jewels of media data. Nielsen ratings cover more viewing time (for now), but broadcasters, studios, and investors all subscribe to them. YouTube covers more online viewing time, but their content leans towards non-exclusive music content at the high view-count end and user-generated content in the long tail. And even if YouTube knows that cute animal content performs well, it’s hard to translate that knowledge into an increase in dog-on-skateboard or cat-in-weird-pose video content. And while granular stats like percent completion and common stopping points aren’t available, raw viewing data is widely distributed.

From a purely technological perspective, Netflix is one of the most fascinating companies. They are extremely secretive about their recommendation algorithm- one which knows how an individual user will pause/skip or rewind. The recommendation algorithm has the worldwide sampling data getting better with each passing millisecond as users log in to their accounts.

Since Netflix is using paid content, their viewing data is structured: for a given show or episode, they know who wrote it and who starred in it. It wouldn’t surprise me at all to know that Netflix had scene-level data on actors, and that they tracked which actors predict that the viewer will stop watching, and which predict that they’ll keep bingeing.

emphasis mine

I won’t be discussing their “financials” because they don’t reveal the “churn in the metrics”.

The more time you spend on Netflix, the more apparent it is that the company is a black box. This is for prudent strategic reasons. Local TV companies don’t need to know how many subscribers Netflix has in each country. And Hollywood agents definitely don’t need to know which of the people they represent are starring in huge hit shows. Netflix’s decision to stop disclosing churn is a little more mysterious, since it makes every seasonally slow quarter a white-knuckle experience and doesn’t seem to matter much strategically.

Why do you reckon is Netflix important here in the context of AI/ML and medicine? The most “successful” companies have a water-tight black-box which is not immediately apparent to outsiders. Do you think Netflix produces movies and streams? They worked on the most efficient algorithms to stream content and developed an industry standard for algorithms. The most unpredictable human aspect of human behaviour- “choice” is being fashioned by the algorithms in a profound manner that you are not even aware of it.

AI is indeed fascinating. I have several ideas around what Netflix will likely do- but those fall in the intellectual masturbation criteria. I am closely watching Netflix on how it performs.