The Data Scientists Quietly Powering Hollywood’s Greenlight Machine

0:04This episode is presented by AMC Network. A new chapter in Anne Rice's Immortal Universe begins with AMC's The Vampire Lestat. Get a backstage pass to the iconic frontman who Pace Magazine calls a Bowie-inspired rocker that will have fans screaming. Don't miss the legendary vampire Lestat de Liancourt in his own electrifying rock saga. Watch The Vampire Lestat Sundays only on AMC and AMC+. Learn more at amcplus.com. This episode is brought to you by LinkedIn ads. Ever invest in something that seemed incredible at first, but didn't live up to the hype? Marketers know that feeling. They optimize for the numbers that look great, impressions, reach, and reacts.

0:45But when they don't show revenue, well, that's a not so great conversation with the CFO. LinkedIn has a word for that, bull spend. Instead, why not invest in what looks good to your CFO? LinkedIn ads generates the highest ROAS of all major ad networks. Reach the right buyers with LinkedIn ads. You can target by company, industry, job title, and more. So cut the bull spend. Advertise on LinkedIn, the network that works for you. Spend$250 on your first campaign on LinkedIn ads and get a$250 credit for the next one. Just go to linkedin.com slash the town. That's linkedin.com slash the town. Terms and conditions apply.

1:27Tom. It is Thursday, July 16th. It's no secret that Hollywood is increasingly dominated by large global streaming services and that those streaming platforms are governed by data. In fact, Netflix often likes to describe itself as having one foot in Hollywood and one in Silicon Valley. But people inside Hollywood tend to know a lot about how the Hollywood side of the company works, yet they know a lot less about the tech side, the data side that influences so many decisions and, when it's working, can help predict how a project will perform. Data has always informed what films and TV shows get made, from the box office track record of stars, maybe their Q rating, to the Nielsen viewership and demographic information.

2:08But since the 2010s and the rise of direct-to-consumer digital platforms like Prime Video, Apple TV +, data has become much more powerful, used to define the content strategies and influence production choices and identify promising genres or even storylines. evaluate talent, and effectively price projects for the platforms. And along with that has come an army of statisticians, data scientists, machine learning specialists, thousands of them across the industry, occupying key roles at the streamers. They're not at the movie premieres or thanked in Emmy speeches, but these data scientists are the quiet powers at the streamers.

2:44And few people in town know much about them, or more importantly, what they actually do. That's why a new book caught my eye. It's called Data-Driven Hollywood, the new data professionals in the age of streaming. It's by Violin Roussel, a professor of sociology at the University of Paris 8. I read it, and it's really about how the, quote, creative choices have been reinvented over the past 15 years thanks to data, for better and worse. Violin talked to about 75 data scientists at the streamers and studios with titles like VP and head of science and algorithms at Netflix. And she chronicles the rise of the data nerds at the streamers, the tension with traditional producers and creative execs.

3:23So of course I had to get her on The Town. Today it's the data that ate Hollywood and how streamers decide what to make and predict what viewers will love. From The Ringer and Puck, I'm Matt Bellany and this is The Town.

3:39Okay, we are here with Violin Roussel, who is professor of sociology at the University of Paris 8 and an author of a very interesting new book. Welcome. Thank you. Thank you for having me. Everyone in Hollywood knows that data is informing most decisions at the streamers and increasingly the studios. Data, data, data. We hear it all the time. We see the engagement report. In fact, Netflix's engagement report is coming out today and we'll get to see the result of their efforts over the last six months. But we don't know exactly how this works. And you wrote a whole book about how data and data scientists are being used at the streamers.

4:25So first of all, how much is data informing every decision at these companies from what shows to develop to what stars to hire? So what I discovered conducting investigation in Los Angeles and in Hollywood regarding data scientists who work for the streamers and now, as you also mentioned it, for the studios. And really, the distinction between the two is not as relevant as it used to be because the studios have also their own streaming platforms and the streamers are also studios. But the difference is really how much data is used. And it's happening, in fact, increasingly. As we speak, it continues to develop, in fact, and now also with AI, which is a new development that I need to do a new book about, to write a new book about and new research.

5:22But from the mid-2010s on, it's been a process of having data and data specialists more and more involved in not only producing numbers that can help for marketing or that can help producers to justify their choices in retrospect, but really involved at the very early stage of deciding what should be made. And that means both the general content strategy of the streamer of the platform, as well as more specific decisions on categories of projects or on one specific project that would be expensive, for instance. So I want to get into the kind of nitty gritty of how this works, because there are, from my experience of just talking to these companies, there are content strategists, there are content finance analysts that look at the economics behind each project and try to leverage data to say, OK, we should be placing more bets here because this genre with this filmmaker actually makes more sense for us as a platform.

6:42There are content planners that look out into the future and say, OK, this is in the ballpark of where we should be placing our bets. We're getting our lunch eaten by Netflix in the podcast and daytime consumption space. So all of a sudden, a year after they notice that data, Netflix is going full speed into podcasts. So explain a little bit to us about the interaction between these content strategists, analysts, and the creative teams. Yes. So first of all, you mentioned different categories, in fact, of data specialists, and that's very true. And in particular, what I describe in my book and what I realized doing the fieldwork is that there are two big categories of data specialists that play a different role.

7:36Some of them are data scientists, sometimes engineers, sometimes people with a PhD in computer science. And they are more on the tech side of things, creating the models that are going to be used for the algorithmic measurements of things later on. And then there are the people who work in the data strategy, content, content data analysis, content planning departments. The names vary a little bit depending on each organization in Streamr. But basically these people act as middlemen, middlewomen between the kind of hard science data scientists and technicians on one hand. And on the other hand, the traditional type of producers that work with the creators, with the artists to produce, to make the content, who have been traditionally the ones that were really monopolizing the power to make the decision before, depending if they were high enough studio heads and studio executives that had enough power.

8:52they were indeed very powerful in deciding what should be made. And now they have to negotiate, in a way, with those data scientists. And in particular, they do it through the content strategy teams and the content strategists, content analysts, that have a business background and are more able, in a way, to translate the language of data in a way that is understandable by the traditional producers, but also the language of the traditional producers to the data team or the data science team, rather, so that they can make the tools in a way that is more efficient for the goal that is. Right. So it's a back and forth.

9:41It's not just we should be making more, you know, comedies with Will Ferrell. It's the production team coming to them and say, listen, we've been pitched a comedy golf series with Will Ferrell. Give us your assessment. And these teams perform analysis based on who the audience is, who is potentially going to watch this, what the cost of the show would be. And then they come up with a price that they are willing to pay for this project if they decided something that's worthwhile. Yes. I would say it's a negotiation, but the terms of the negotiation have changed over time. An internal negotiation.

10:20An internal negotiation, yes. And I would say at the start of this process that I studied, which is in the mid-2010s and late-2010s, the traditional producers, the content people, had a lot of power in those negotiations still. And the data specialists were trying to push their own agenda or their own forms of legitimacy to establish their own forms of legitimacy, their own, I call this new way of defining how content should be selected, a new production narrative. So they are pushing their new production narrative that says that basically it's important to look at the data. And a lot of pushback you describe in the book to that process.

11:11And it was not just, you know, the traditional, uh, I'm the, you know, the creative executive, I'm the producer. I know better than what the data says. But also you describe a lot of demographic differences. You know, these are PhDs. These are scientists. You say they are much more female than traditional Hollywood producers. They're non-white. They are largely immigrants. They are people that are not the traditional Hollywood gatekeepers. And this was a challenge, but I think those days are over. I think these days, everyone accepts it. Even the top producers want to see the data. They may not listen, but they do want to see it.

11:53Yes. And they have to, if not listen, at least they have to take these other key players that are the data specialists into account now. At the beginning of when I started doing this field work, It was in the mid-2010s, and I could hear a lot of producers, studio heads, people with power in Hollywood at the time, tell me, oh, I don't believe in data. I don't believe in data. I don't, whatever they say, I don't care. I know the artists. I have the relationship. I have the connection. I have the experience. And I have an eye for quality. all things that the data specialists don't consider to be a skill.

12:40It's not their skill. And they come from, like you said, they come from a very different background. And they also, they don't come from a field school. And they don't come from the background in which they would value making art or working in entertainment. Well, I think there's still some. Taylor Sheridan would probably still take a ream of data and whip out a cigar, light it on fire and throw it out the window. But I think there are more and more that will. Precisely. There is more and more. You said the key thing, which is that over time, these two groups that are the traditional producers and the data specialists have come to share more common ground, to learn the way of the other, to build this negotiated order in which they can work together.

13:36Because it's absolutely true what you said. There is not as much friction as there used to be. I want to get into how these platforms assign comparative values to projects or even to individual actors. And it's a mix of the internal data that they have about their customers and what people are watching. But it's also third party data. They are getting outside information from Google, from Facebook. Amazon uses IMDB to see what people are searching for and what kind of stars are getting engagement. And they use that data to inform Prime Video. So take us into how this actually works internally at the streamers to figure out what to make.

14:24So at the time, they used a lot of external data. Now all of the streaming platforms have their own internal data. And they use, I think, the external data mostly for information that you cannot get from your own data, which is what will work in the future that we don't yet have. Their internal data only give them information about what they already have. So if they want to try to guess, and that's this, I call that the oracle function, their role of predicting the future, basically. Well, that's what this is all about, trying to predict what's going to be a hit, predicting hits. and you know there's a long history in hollywood of charlatans coming forward and claiming they have the magic secret to predict hits i mean ryan kavanaugh was the famous one at relativity media and he ultimately was proved to be not that um but for as much as the data can inform hit after hit at netflix seems to butt up against that theory adolescence squid game stranger things Are these projects that the data would have said are going to be hits?

15:34That's the interesting fact. These hits that are the most famous of the big platforms, such as Netflix, the one that you just mentioned, they are not necessarily projects that were backed by data or suggested, created, following what the algorithm was saying. Every producer is now cheering in their car that I asked that question because they hate this stuff, you know? Yes, but that doesn't mean that the economic model of Netflix that made Netflix as successful as it is, is not at least party based on their use of data. But it's mostly to detect what is the most successful in what they already have and make more of the same.

16:22Because with the categorization of content, the big databases that they have built based on the titles that they already have in their library of content and creating micro tags, micro genres, micro categories to classify those titles. and then trying to compare based on that the new projects that they are offered. So basically with this system, there is a feedback loop that you can imagine. Like the system is telling you, is suggesting that you're making more of the same that worked in the past. And that's where we get to the taste clusters, right? Yes. Because Netflix uses this system of taste clusters where you say it's about 2 ,000 separate little micro communities that they see in their data that tells them what would appeal to certain types of people.

17:20And explain how that works. So the taste clusters are basically made by combining the subcategories of content that have been created following the process that I mentioned before. These microcategories of content are connected to microcategories of behavior that the people who are on the platform, the subscribers, what they are doing on the platform, basically, what they watch, when, how, in how many segments, everything that defines subcategories of behavior in how you consume the content. Let's use Craig as an example. He watches a lot of NBA documentaries and high school comedies from the early 2000s.

18:15Is that a taste cluster? Like what, like, give us the categories they're using. The categories of content can be what you just mentioned. You know, a movie with strong female lead in the set in the, in France, in the 60s, for instance. So it gets that granular. That could be a subcategory of content. Yes, the categories of content can be extremely granular, even smaller than this. And then you combine that with information about the behavior of people on the platform. So these subcategories of behavior, it's classes of behaviors. It's not groups. That's why I can't tell you it's like white people that age, that level of education.

19:02Oh, yeah. I've talked to Ted Sarandos about this. He's like, I don't know who you are. I only know what you do on the platform. Yes, exactly. So basically, these subcategories of behavior that are not one person or one group either are connected algorithmically with the subcategories of content to produce these taste clusters. But you understand that these taste clusters are very difficult to illustrate. I could show you, I'm not prepared to do it, but I could show you a diagram, what it looks like, but it won't tell you who is in this taste cluster. So then that is used to create the algorithm that predicts what you, Joe Blow, is going to want to see next.

19:48No, it's used to predict what the platform should make. Oh, so that's on the production side. So if we are seeing a taste cluster forming around a particular style of content, we should make more. Yes. Yes. This episode is brought to you by Accenture. When your advertising operations fall out of sync, campaigns slow down, insights get buried, and opportunities get missed. That's why Spotify and Accenture are working together to reinvent the rhythm of ad sales. using automation, analytics, and smarter workflows to simplify campaign delivery and access better data across the business. The result?

20:29Less time spent on operations, more time connecting brands with the moments and fandoms that matter most. To learn more, check out Accenture.com. This summer, FanDuel is the best place to bet on goals. Including equalizers, voles, headers. Every goal is worth more on FanDuel. So let there be goals. New customers get up to$1 ,000 and bet reset tokens when you bet$5 daily. 21 plus and present in select states. Often required. Refund issue does not withdraw, but bonus bets that expire seven days after receipt. Minimum$5 wager required for five consecutive days. Max refund$200 per day. Restrictions apply, including token expiration.

21:03See full terms at Vanduul.com slash sportsbook. Gambling problem? Call 1-800-GAMBLER or 1-800-MY-RESET. All new Sundays at 9, exclusively on AMC and AMC+. I am the vampire Lestat. I'm a rock star now. I'm a real killer. from ann rice's immortal universe comes what vulture calls the most momentous event in fictional rock history thousands of fans love you i want millions it's my era and rice is the vampire list dot all new sundays at nine exclusively on amc and amc plus stream now the example that always goes around hollywood is the house of cards example that you discuss in the book, actually, about how the legend is that when Netflix decided to get into original content, the data people said, we need a politics show set in Washington, D.C., and Kevin Spacey should star in it.

21:59And all of a sudden, this House of Cards adaptation from the U.K. became available, and Netflix saw it and said, this is what our data people want us to make. Let's blow everybody out of the water with a double the bid that HBO is making, and then House of Cards became the first big Netflix show. I've also heard that story is BS. Is it BS? Yes. It's a tale. It's a tale that people tell. Right. Well, it doesn't mean, I've asked Ted about this. He said, yes, Kevin Spacey was a big star in the movies that we had on the platform with Kevin Spacey were big. So he was a good star for us, but I don't think it got that granular at that point.

22:41No, especially because at that point, the data teams were very, very small. It happened in the very early day of using data to decide what to make. So producers, I think, want to know what are the most important data metrics that Netflix and others use to judge their shows? Is it pure audience engagement? Is it completion rates? Is it the so-called decay on the show, like the drop-off rates? Like, what is it that they most care about? Or does it mix? I think it's a mix and it also depends on the moment. You know, when they feel that they have maxed out on how many new people they can get onto the platform.

23:25Do they feel this way or not? It depends on where. In the US, obviously, I think it's the case, but elsewhere, they still have room for growth. So basically, the metrics that matter most depends on the moment and their strategy in the location. where they are looking at the metrics. But it's very interesting. Also, something that I observed during my, when I was doing fieldwork is that everyone else in the other platforms that are not Netflix wanted me to tell them something about Netflix because they believe that Netflix somehow has a secret, a secret recipe to make you successful. Well, you talked to 75 people, including a lot of people at Netflix.

24:12Do they have a secret? No, I don't think so. I think they were smart and they were ahead of others and they took a lot of risks. They were in debts for a very long time. Yeah. And their strategy was successful. Well, the market supported that debt. Yes, eventually. But I don't think that they have a data based secret. Yeah. But the funny thing is now everyone has data. The talent agencies have their own army of data scientists. And you describe in the book this funny thing where it's like it's now data versus data. Whose data is better? So you've talked to people at the agencies who are obviously using the data to try to boost their clients and get more money for their shows and explain why their shows shouldn't be canceled.

25:01And yet they're going up against these big scaled platforms that have, I would guess, better data. Whose data is better, the talent or the streamers? It breaks my heart to say it because I studied the talent agencies for a long time and I really enjoyed it. But it's really difficult to beat the platforms, the streamers, because the streamers, they also are very protective of the confidential nature of their data. They give very little information even to the people that they work with, the external producers or creators, because they know that their data is their power and their negotiating power.

25:44How dare you? Netflix says they're the most transparent platform out there and they release more data than anyone ever has. Data is so precious. Can I tell you? Yeah, it's it's it's it's very global. It's not just Netflix. Sure. No, I get it. So you wrote this book, you researched this for many years. Are you convinced that the algorithm and the data knows best when it comes to what projects to greenlight and what shows to renew? I don't know if it knows best, but it knows or not algorithm and data, but the people, the people behind the algorithms and the data. Because for me, it's very, I'm a sociologist of work and occupation.

26:29So it's very important to study people at work. It's not data or algorithms that do things. It's the people behind them. So I don't know if they know better, but they know differently. They have a different way of grasping what are the audiences and what they want and to contribute to defining it, to making it happen. And it's very, if you think of the way that producers, traditional producers before were deciding what to make, it was mostly by focusing on their relationship with their artists, the people that they had a strong relationship with, that they believed in. And they thought that these artists had a connection with their audience and maybe can command some box office.

27:29But it was really in this duo between the producer and the talent that what was going to be made was decided. And now the power balance is a little different and has shifted in favor of the data specialist. I don't know if it's not more accurate. It's just based on different skills. Yeah. Forget whining and dining the stars. These agencies and producers, they should be whining and dining the data guys. Maybe they do. I had one question. Are there any examples of hits that were primarily a data-driven decision? We talked about Stranger Things and Adolescence and Squid Game and the Taylor Sheridan shows, all shows that might not have worked because of the data.

28:19What about the opposite side? Are there shows that are huge hits that were primarily a data-driven decision that creative executives would not have thought of? I mean, there are shows that were supported. It's difficult to know exactly what is a show that is based on data versus not, because nowadays, most shows, they have to be, you know, they have to be evaluated positively by the data teams to exist. Right. The Greenlight Committees includes those people now. Exactly. So you can always claim that it's because of the data. But a lot of shows that are unusual, that are very different, that you have never seen this type of show before, cannot be really based on algorithms to suggest that this is the best thing to do.

29:10Because there is this, you know, with the algorithmic logic, it is based on what worked before. Okay, so last question. Is AI going to blow this entire thing up? Are the data scientists going to be replaced by AI? And us too. And us too, you and me both? Yeah. No, but that's a separate question. But let's focus on the massive data employee base that these companies have. Is that going to go away? I can't really predict what the future is going to be. Oh, but that's what this is all about. But it's their role, not mine. But what I can tell you is that people who work in these data teams, they are definitely worried.

29:54Because it's true, but it's not only the data specialists that have this kind of concern. It's in many, many sectors that there is a risk that AI be used. Even for producing content, maybe you don't need as many data specialists, but you also don't need a lot of crew people and even actors and actresses anymore. Some would argue that a lot of the content on Netflix is already AI generated, but I assure you it is not. It's not supposed to, at least. Yeah. No, I don't. I don't think it is. But I know I'm working on China. I'm doing fieldwork in China and AI is even bigger. And it's definitely changing.

30:43It's definitely a revolution and it will be a revolution in Hollywood as well. And it's an opportunity for me to do fieldwork again and write another book. I can't really tell you more than that because I have not done this yet. I get it. Well, I appreciate the insights here. It's fascinating. I think all producers and people who are selling projects to the streamers should read your book and at least understand what you're up against here and what you're dealing with when these companies are evaluating projects. So I appreciate you coming on the show. Thank you so much, Matt. We're back with the call sheet.

31:20Craig, you and I saw the Odyssey on Monday night. We schlepped to Universal City. The premiere was in New York. So we went to the press screening. Chris Nolan will not allow us to see this movie in any theater other than his preferred universal CityWalk IMAX 70 millimeter film theater. First off, before we get to the movie, what did you think of CityWalk? You'd never been there. I had never been to CityWalk. You know, kind of like a like a beaten down downtown Disney. Totally. It's one of the worst places in Los Angeles. And that says a lot. Yeah. I love, you know, more power to Christopher Nolan.