[Majromax]

> I guess it shows that novelty is probably what drives people to the platform and gets new subscribers.

It assumes that novelty is what drives people to the platform and increases/maintains the subscriber base.

However, Netflix's evident model is not capable of falsifying this assumption. By cancelling series as soon as they plateau (not even _decline_, given reporting (https://www.wired.com/story/why-netflix-keeps-canceling-show...) of its production cost escalator), they do not generate a large library of "complete series" for later viewing.

This also puzzles me, since it contradicts Netflix's willingness to pay a pretty penny for established series like Friends. Its current show-commissioning practices seem to be incapable of generating a new generational hit like that.

[skohan]

I agree, and I think it's a problem which can occur in data-driven systems: the system can stabilize at a non-optimal local maximum.

Ideally they should A/B test at least occasionally with different decision models to see if their assumptions are correct, but those are very expensive experiments to run when you're talking about the production of a TV series.

[Majromax]

It's also difficult because Friends is also a long-tail phenomenon. There were dozens of 90s sitcoms that failed to become generational hits.

Also, a counter-argument to my proposition above is that broadcast TV through the mid-90s had a captive audience in a way that Netflix doesn't. To a limited extent, broadcast TV could force audiences to become familiar with a show, pushing an originally-marginal show over a threshold of popularity.