Tartle Best Data Marketplace
Tartle Best Data Marketplace
Tartle Best Data Marketplace
Tartle Best Data Marketplace
September 15, 2021

You Are Watching Netflix - and Netflix Is Watching You

Netflix's Data Strategy for a New Global Video World
BY: TARTLE

Netflix is one of the biggest names of the Digital Age. It went from being a new way to rent movies, to a streaming service aggregating everyone else’s movies and TV shows, to being a powerhouse creator of original content in its own right. Oh, and it killed a one-time giant Blockbuster along the way and helped spawn a whole new branch of the entertainment world. However, it’s fair to ask if Netflix is really about making movies and TV shows. Is that really their main concern? The answer of course is ‘no’. The company is primarily about making money and for that, they need subscribers and in order to get the largest number of subscribers possible, they make use of a lot of data. 

Naturally, they start with what they hope you want to see and basically spam your feed with a bunch of generically popular content. Overtime, they will try to narrow it down. How do they do that? They make use of algorithms to gather information on what you watch. They also pay attention not just to what you click on, that is both too simple and not terribly informative. How many times have you clicked on a movie only to get about ten minutes in and decide you don’t want to watch it? The algorithms pay attention to that as well. What you watch, how long you watch it for, when you watch it, and all of that goes into the algorithm. From there, Netflix’s hope is that they will be able to find similar movies and put them in your recommended feed. Sounds simple doesn’t it? Yes and no.

For one, there are holes in this system. Not just the occasional recommendation you would never plan on watching but major problems that can break the algorithm all together. Say you have roommates and you all share an account but also have very different tastes in movies? Or you have kids. Chances are you watch different things when they are around. That of course is what the profiles on Netflix and every other streaming service are all about. If they can break out each person individually, the algorithm has a chance to work. However, how many people really bother with the different profiles? I’m guessing it’s not as many as Netflix would like. Also, what if you have a busy schedule and rarely have time to watch a full two hour movie? You only have fifteen minutes here, twenty minutes there and typically work on something while the movie is on. So it might take a whole week to watch one movie. That kind of person likely wrecks the algorithm entirely. 

Not to mention, how long does it take to build up a worthwhile profile of a given subscriber? One week? A month? A year? It will vary from person to person based on how much they watch, meaning how effective the recommendations are will vary a lot from one subscriber to the next. In short, Netflix’s algorithms are extremely inefficient in a variety of circumstances, and that means they are wasting time and money building user profiles that don’t work.

What should they do then? What would be a more efficient means of building those user profiles? Netflix could work with us at TARTLE. They could go directly to subscribers and ask what it is that they would like to watch. Who their favorite actors and directors are. When do they watch? Do they prefer movies or series? Do they like their series dumped all at once or would they prefer a weekly schedule? Netflix could talk directly to its subscribers and get feedback directly from them and so build a far more accurate profile than any other method. This would be faster and cheaper and in the end far more financially rewarding. Which means they could spend that extra time and money making better content to draw in more subscribers. 

What’s your data worth? Sign up for the TARTLE Marketplace through this link here.

Summary
You Are Watching Netflix - and Netflix Is Watching You
Title
You Are Watching Netflix - and Netflix Is Watching You
Description

The company is primarily about making money and for that, they need subscribers and in order to get the largest number of subscribers possible, they make use of a lot of data. 

Feature Image Credit: Envato Elements
FOLLOW @TARTLE_OFFICIAL

For those who are hard of hearing – the episode transcript can be read below:

TRANSCRIPT

Alexander McCaig (00:09):

Data science and Netflix. Yes, ladies and gentlemen, welcome back to Tartle Cast.

Jason Rigby (00:15):

New Mexico Netflix?

Alexander McCaig (00:16):

New Mexico, Netflix. It's not about the movies. Their bread and butter is delivering you movies that they think you want to see.

Jason Rigby (00:29):

The suggestions.

Alexander McCaig (00:30):

It's not, for them, about making the movies. It's about everybody making movies and making sure that the ones that are made are seen by the right people.

Jason Rigby (00:38):

But the content is not the main focus of Netflix. It's to introduce great content so that you get subscribers.

Alexander McCaig (00:45):

Mm-hmm (affirmative).

Jason Rigby (00:46):

They want subscribers.

Alexander McCaig (00:47):

They want subscribers and they get more-

Jason Rigby (00:49):

That's how they grow.

Alexander McCaig (00:50):

They get more subscribers through their data processing algorithm, which is an affinity propagation algorithm. What that's saying is, imagine you have a table, rows and columns, and each one is, say a descriptor and then within that descriptor, you have variants of that description. And then once you're in there, you rate each of those variants within that description. So when you're on your Netflix account and you choose a piece of content to start this off, Netflix starts a table on you. And then from this table, they're going to be like, okay, we know that this piece of content has these identifying factors and these levels of characteristics within it. This person watched it all the way through, start to finish. So our affinity propagation, so it's going to feed into itself, this machine learning model, to say that, okay, they've watched it once, let's tee them up with something that's kind of like it. Now let's also tee them up with something that is not even close to it.

Alexander McCaig (01:57):

Now, now we can begin to compare things across this table. Did they choose the thing that was almost directly associated with the flavor of the thing they watched before or did they go for something completely different? Now when they watched something completely different, did they watch it all the way through? Three quarters of the way through? Turn it right on, turn it right off? Did it capture the same characteristic identifiers and variants that the first one did?

Jason Rigby (02:20):

Mm-hmm (affirmative), yes.

Alexander McCaig (02:21):

Oh no, it didn't. Okay. So maybe we shouldn't give them something totally polar opposite, but maybe something in between. What's like a middle of the road kind of thing. Oh, they actually watched that all the way through.

Jason Rigby (02:30):

Well, it's super important with these because, and you know as well as I do, for me, my table will be really easy, it's just action movies.

Alexander McCaig (02:37):

Yeah.

Jason Rigby (02:38):

But I'm saying, YouTube does this too when they're watching percentage. So I can advertise to somebody that watched 100% of Tartle's video, because obviously that person must be really interested. I can say they watched 10% of it, 20% of it, 30, and so on.

Alexander McCaig (02:54):

Or it was on and they just left it on.

Jason Rigby (02:55):

Yeah. They left it on. Yeah, you have that part too, we've talked about before. But Netflix is knowingness. So it's comes really important with this, so let's say you're in the United States and everybody's watching action movies and they suggest a French action movie to you. And it has subtitles, and so people start to watch it for five minutes and then when they see the subtitles, they're out.

Alexander McCaig (03:15):

Doesn't work.

Jason Rigby (03:17):

So they know that, and so now they can group people into buckets, which we talk about all the time, and say 18 to 24 year old males that watched French action movies don't like subtitles.

Alexander McCaig (03:26):

And when a producer shows up and says, "Hey, we want to list this movie on Netflix." And they say, well, actually we know that these don't do well on Netflix.

Jason Rigby (03:33):

Trust me, they have all the data.

Alexander McCaig (03:33):

And we know exactly how much to pay you, depending on the audience size that we have.

Jason Rigby (03:38):

Why do you think documentaries and podcasts are all about serial killers and murderers and crazy stuff? It's just constantly, I was on HBO Cinemax of the day and I wanted to look at a documentary and I'm like, there's no good, all these documentaries are crazy people doing crazy shit.

Alexander McCaig (03:55):

I have no interest in that.

Jason Rigby (03:56):

Why is everybody watching that?

Alexander McCaig (03:57):

What's that football player that did something?

Jason Rigby (03:59):

Yeah. Yeah. I mean, that one's one of the top.

Alexander McCaig (04:00):

I literally could care less. Their algorithm sucks so bad. I go on Netflix because I heard about it and I want to watch one thing. I almost never, almost never being like, oh, that's a great suggestion. I don't want to watch 99% of the other things they're suggesting.

Jason Rigby (04:16):

Well, that's because they need Tartle. Because then you could ask somebody specifically, do you like crime documentaries?

Alexander McCaig (04:25):

Do you like subtitles? Do you like this? Do you like that?

Jason Rigby (04:26):

And rating one to five, where are you at as far as liking? Least likely to like that, you need to ask people directly.

Alexander McCaig (04:34):

And before you pay a producer to list this thing, that's going to be a bust, don't do it.

Jason Rigby (04:39):

Yeah.

Alexander McCaig (04:39):

They had this one that Tom Hanks, I think it was, or ... no. Who's the guy who was in Oceans 11 and 13? Lead actor.

Jason Rigby (04:52):

George Clooney.

Alexander McCaig (04:52):

George Clooney. George Clooney is in this one where he's up in the Arctic Circle or something like that. They paid all this money to get this movie made, and I'm like, and they suggest it to me, I'm like, oh, okay, cool. I'll check it out. Sucked. I watched the whole thing, it just totally sucked. And then it starts suggesting to me these other films, and it's like, if you had told me the plot ahead of time for this whole film, I wouldn't have watched it. If you told me it wasn't going to be explained, there was no actual start and finish to this story, I wouldn't have watched it. Yeah. I would have told you don't pay the money to it, the whole thing was a flop, but you made it look great.

Alexander McCaig (05:26):

The algorithm sucks, because they actually don't know the flavor of a person's mind. They just use an affinity propagation model in their machine learning, to take a best guess about what you want to see, and then that acts as a key driver. So they use lagging indicators to be their key driver for how they make decisions going forward.

Jason Rigby (05:45):

Yeah. And Spotify is having issues with this. The things that's suggesting to me are horrific. I listen to EDM music, that's it. Maybe a little praise and worship music, because I liked that music, and then everything else, 80% of my listening habit is podcasts. So why are you suggesting top 40, classical music? You're way off. These companies have to realize, the only way you're going to get accurate information is go to the source.

Alexander McCaig (06:16):

Thank you.

Jason Rigby (06:17):

There's only one way.

Alexander McCaig (06:18):

Fix your damn affinity propagation algorithm tables, by putting in the human input. Don't just create a model of computer input.

Jason Rigby (06:27):

So if Netflix came to you and they said, "Alexander McCaig, CEO of Tartle, we want to know how Tartle can help Netflix." So, to be more accurate.

Alexander McCaig (06:37):

Here's the first thing I would ask Netflix.

Jason Rigby (06:38):

To serve people.

Alexander McCaig (06:39):

Netflix, what is it you would like to know? What would you like to know about people? And then they'd say, "Well, we want to know exactly what people want to watch." Okay. What else do you want to know? "Well, we want to know how likely they are to watch something and what their preferred time is to watch something." Obviously, your model is going to be skewed to people watching videos at night because they got home from work and that's their habit. Maybe you can change that so you can get people to watch videos all day long, but you have to understand the habit, you have to understand the flavor of how they do things, not just look at the device that came in, but you have to speak to that individual. You have to create a bridge to them, you have to understand the decision about what they are going to do, rather than wait for them to watch the video and then guess on what they should do next.

Jason Rigby (07:30):

On a marketing side, I would have to say Netflix, you're putting yourself in a box.

Alexander McCaig (07:34):

By putting others in a box, you put yourself in a box.

Jason Rigby (07:37):

You're basically going to turn into who you took over.

Alexander McCaig (07:42):

Yeah. And ...

Jason Rigby (07:43):

Who was it? Video stores?

Alexander McCaig (07:46):

Blockbuster.

Jason Rigby (07:46):

Yeah, Blockbuster. You're all in on one thing. You have to realize you're a cloud media company. And so it's media, how can I serve media in multiple streams to people that are consuming it? You should have bought out Spotify a long time ago.

Alexander McCaig (08:03):

Yeah. And I just, I'll make one last comment and we'll close this out. If you statistically flatten the value of a human being, you statistically flatten your growth, and you statistically flatten the likelihood of your longterm survival as a company.

Speaker 3 (08:24):

Thank you for listening to Tartle Cast with your hosts, Alexander McCaig and Jason Rigby, where humanity steps into the future and the source data defines the path. What's your data worth?