• You have deals with N big media companies who each have their own restrictions on who can stream what from where. The list is constantly changing, so permissions to view media are locale-specific and revokable; you need a way to say “okay this person is not allowed any more to do that.”
• Multiple-screen detection emphatically needs to be rock solid. Someone is going to unplug their Roku player when their laptop says “you're watching from too many screens,” and by the time they get back to their laptop you need to be detecting them as streaming from 0 screens. At the same time a hiccup in this process shouldn't cause like 3% of your users to get a big streaming interruption as they don't seem to be online.
• You have to recommend stuff based on what this person has watched. An acquisition team needs to do cluster analysis on this to get new stuff to fill all of the different clusters of interests that emerge in your user base.
• People will search for shows you don't have. (Because of point 1, the big media companies only permit access to a fraction of their backlog.) You have to know this media that you don't have access to, well enough to recommend something related that might keep the user on Netflix instead of hopping to another service.
• All of this has to happen on pretty low latencies when someone starts up Netflix. That is, anybody who jumps into Netflix should see a personalized view of what they were watching, what they can watch, filtered by your allow lists and not cached on their device, within just a few seconds.
• All of this has to be portable to all of the different platforms Netflix supports.
The thing is, as a consumer I can't tell any difference between the technical solutions of Netflix vs Disney+, HBO Max or basically any other streaming service at this point, which proves that it's not actually that hard of a problem, at least today. I'm sure they were very innovative when they pioneered streaming but this is 2024. I'm sure their tech is better in multiple ways, but not in ways that actually lead to an edge, or is even noticeable to consumers. Now it's all about what content they have.
For comparison, I can tell a MacBook is better than a Windows laptop in 100 different ways. I can tell that the touch screen UI and self driving features in a Tesla is 15 years ahead a Toyota, but I cannot tell Netflix is better than its competitors.
> You have deals with N big media companies who each have their own restrictions on who can stream what from where. The list is constantly changing, so permissions to view media are locale-specific and revokable; you need a way to say “okay this person is not allowed any more to do that.”
Yeah, international media content licensing is very likely Netflix's moat. I'd wager starting a streaming service, and getting the big studios to host their content is pretty much impossible.
Recommender systems are pretty well studied, and anyways, I'm sure it's significance for Netflix is overrated. People (including me) tend to watch the show everyone's talking about. It's a much more hairy issue for a site like Youtube, where users upload decades worth of content every day, and the site has to figure out what to show to which viewers, where the content's shelf-life might be measured in days, so building up collaborative filtering data might not be feasible.
All the other technical issues are probably non-trivial, but I don't think any of them requires world-changing engineering prowess.
Most engineering is non-trivial. If I attempted to design a washing machine, I'd probably fail miserably, and figuring how to do it well probably took collectively thousands of engineer-years. Doesn't mean it requires exclusively unicorn engineers.
• You have deals with N big media companies who each have their own restrictions on who can stream what from where. The list is constantly changing, so permissions to view media are locale-specific and revokable; you need a way to say “okay this person is not allowed any more to do that.”
• Multiple-screen detection emphatically needs to be rock solid. Someone is going to unplug their Roku player when their laptop says “you're watching from too many screens,” and by the time they get back to their laptop you need to be detecting them as streaming from 0 screens. At the same time a hiccup in this process shouldn't cause like 3% of your users to get a big streaming interruption as they don't seem to be online.
• You have to recommend stuff based on what this person has watched. An acquisition team needs to do cluster analysis on this to get new stuff to fill all of the different clusters of interests that emerge in your user base.
• People will search for shows you don't have. (Because of point 1, the big media companies only permit access to a fraction of their backlog.) You have to know this media that you don't have access to, well enough to recommend something related that might keep the user on Netflix instead of hopping to another service.
• All of this has to happen on pretty low latencies when someone starts up Netflix. That is, anybody who jumps into Netflix should see a personalized view of what they were watching, what they can watch, filtered by your allow lists and not cached on their device, within just a few seconds.
• All of this has to be portable to all of the different platforms Netflix supports.