Latest / The Operations Podcast with Fexingo: Process, People, and Profit in Modern Business / How Netflix Redefined Content Operations With Data
Transcript
- Lucas: So Netflix spent about seventeen billion dollars on content last year. That's more than the entire GDP of some small countries. And the question is — how do they decide where that money goes? Luna: It's not just throwing cash at the wall, right? They have a system. Lucas: Exactly. And it's a system that's pretty unique in Hollywood. Most studios still operate on gut feel and track record — a producer's past hits, a script's buzz. Netflix built something closer to a quantitative hedge fund for content. Lucas: And before we go deeper — quick honest thing. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making episodes like this. So if these conversations have moved your work forward in some small way, you know where to find us. No pressure, just worth mentioning. Luna: Yeah, it keeps us ad-free and independent. Appreciate everyone who's part of that. Lucas: Alright — back to the data machine. So here's a number I love: when Netflix greenlit 'House of Cards' back in 2011, they already knew that subscribers who watched the original British version also watched films directed by David Fincher and starring Kevin Spacey. The intersection of those three data sets was big enough to justify spending one hundred million dollars on two seasons sight unseen. Luna: That's the famous story. But what's less known is how they've refined that approach since then. Lucas: Right. The early model was about matching talent to audience segments. Now it's much more granular. They track what they call 'starters' and 'completers.' A starter is someone who begins watching a show within the first week of release. A completer finishes the season. That ratio tells them not just how many people clicked, but how many actually engaged. Luna: So a show could have a huge opening weekend but a low completion rate, and that's a red flag. Lucas: Exactly. And they use that data to make renewal decisions. It's not just raw viewership — it's velocity and depth. Did people binge it in a weekend, or did it take them three weeks? Did they drop off after episode two? That's all feeding back into the greenlight process for future shows. Lucas: There's also the A/B testing layer. Netflix constantly tests different artwork, different trailers, different episode descriptions. For the same show, a subscriber in New York might see a thumbnail featuring the lead actor, while someone in Tokyo sees a scene with action. The algorithm picks the version most likely to get a click. Luna: I've read they test up to twelve different thumbnails per title. That's a lot of creative work for a single image. Lucas: It is, but the lift is real. They've said that a well-optimized thumbnail can increase viewership by double digits. When you're spending seventeen billion, small percentage gains are worth huge amounts of money. Luna: Let's talk about the sourcing side. How does data influence what they actually produce? Lucas: So they look at what's undersupplied. They'll ask: what genres have high demand but low supply in our catalog? That's how we got 'Stranger Things.' They noticed that eighties nostalgia, sci-fi, and coming of age stories had a passionate audience but not a lot of original content. So they commissioned a show that combined all three. Luna: And then they gave the Duffer Brothers creative freedom, but with a data-informed brief. Lucas: Exactly. The brief wasn't 'write whatever you want.' It was 'we think there's an audience for this specific blend — go make it.' And that blend paid off enormously. Lucas: But the data also has limits. Netflix has canceled plenty of critically acclaimed shows that had passionate but small audiences. 'The OA,' 'Tuca & Bertie,' 'One Day at a Time' — all got the axe despite strong reviews. The data said they weren't pulling in enough completers fast enough. Luna: And that creates tension. Creators argue that some shows need time to build an audience. But Netflix's model rewards immediate engagement. Lucas: Right. And that's a deliberate operational choice. They're optimizing for total hours watched per dollar spent. A show that costs fifty million and gets fifty million hours in a month is a better investment than a show that costs forty million and gets twenty million hours over six months. The time value of attention matters. Luna: So it's almost like they're running a portfolio. Some big bets, some niche plays, but everything has to justify its cost per engaged viewer. Lucas: Exactly. And they've gotten better at predicting that. They have models that estimate the expected viewing hours for a proposed show based on similar past titles, the talent attached, the genre mix. It's not perfect, but it gives them a baseline for what they're willing to spend. Lucas: There's also the operational side of production itself. Netflix pioneered what's called the 'data-informed production' approach. They'll analyze scripts for pacing, scene length, even dialogue density. They found that shows with shorter scenes and more plot twists per episode tend to have higher completion rates. Luna: So writers are getting feedback like 'your second act drags' based on data from similar shows? Lucas: In some cases, yes. They don't mandate changes, but they share the data with showrunners. Some embrace it, some push back. But the operational machinery is always there, measuring and feeding back. Luna: What about the recommendation engine? That's another layer of operations, right? Lucas: Huge. The recommendation system is essentially an operations pipeline that processes billions of data points every day. Every time you pause, rewind, or skip a scene, that's a signal. The algorithm uses that to personalize the entire experience — not just what titles you see, but the order, the artwork, even the episode descriptions. Lucas: Netflix has said that seventy-five percent of what people watch comes from recommendations, not search. So the operational quality of that pipeline directly drives retention. If you can't find something to watch, you churn. Luna: And churn is the enemy. Especially with competition from Disney Plus, HBO Max, Amazon — the streaming wars are all about keeping subscribers month to month. Lucas: Exactly. So the content operations model isn't just about making good shows. It's about making the right shows for the right audiences at the right time, and then surfacing them perfectly. That's a massive operational challenge. Luna: One critique I've heard is that this approach leads to homogeneity — shows that feel algorithmically constructed, like they're designed to hit certain beats rather than take creative risks. Lucas: It's a real concern. And Netflix has acknowledged it. They've tried to balance data with creative instinct by giving showrunners a lot of autonomy within the broad parameters. But the data always looms. When a show like 'The Irishman' cost a hundred and sixty million dollars and had middling completion rates, the data says that's not a repeatable model. Luna: So they're constantly calibrating between art and science. Lucas: Exactly. And I think that's the core operational lesson from Netflix: you can build a system that makes better decisions than any individual executive, but you have to be willing to let the data challenge your instincts. And you have to accept that some great art won't fit the model. Luna: So what's next? Do you think the model will shift as competition intensifies? Lucas: I think we're already seeing it. Netflix is investing more in live events, in gaming, in cheaper international productions. The core data engine is still there, but they're expanding what kinds of content they measure. The question is whether the operational model can adapt fast enough to keep them on top. Luna: And whether the audience will keep trusting the algorithm. Lucas: Right. Because at the end of the day, the data is only as good as the human decisions it informs. And that's a tension every operations-heavy business has to manage.