Are you sure? I feel like Netflix is making more data-driven decisions now than ever. If they aren't producing a constant crop of "HBO-quality" show, I would guess that that's because the data is telling them not to.

Consider: television networks have been in Malthusian competition for decades. You'd think the quality of the median TV show would increase. Instead, it seems like the quality of the best TV shows (on every network) consistently increases, but the median show stays the same. Why?

I would personally hypothesize that it's down to a not-oft-mentioned part of consumption psychology: people don't actually want to consume an indefinite stream of high-quality content that they need to give their full attention to at all times, any more than they want an indefinite stream of world-class meals or a radio station that consists only of their favourite music. Nobody has the time, or emotional capacity, to devote to consuming it. People want "bland" content just as much (if not more!) than "rich" content.

And, as licensing deals fall through and Netflix loses both bland and rich content from their catalogue, they have to replace both of those with their own offerings to keep viewers satisfied.

(Have you ever looked in the fridge, and seen that you have the ingredients to make several great meals—but those meals, you are planning to have at specific times with your significant other later in the week; and then realized that you have "nothing to eat" because there are no non-great meals you can make? Consider the television equivalent. That's the problem Netflix is trying to avoid here.)

I think this is a brilliant point that immediately resonates. I frequently watch or rewatch pretty formulaic shows, particularly if I’m doing something else at the same time. It’s comforting somehow. At the same time, I actively seek out and consume with undivided attention the best shows that the genre offers. And that’s enjoyable but it’s also...exhausting? That’s too strong a word, but I basically don’t want to do it all the time, just like I don’t want Michelin 3 star restaurants for every meal.

And I’m not alone: a LOT of people I know either split their TV time between the best stuff and the junk food stuff, or only really watch the junk food stuff. It’s interesting, for sure.

It’s simple economics. AMC, as an example, doesn’t have to have a ton of great shows to be carried on cable companies and to get them to pay to keep AMC on. They only need the “Walking Dead”. Enough people would be very upset if they couldn’t get that one show via cable.

But beyond that, while I'm sure Netflix uses data to allocate money for genres and other types of planning (and help determine whether to keep shows on the air--maybe it's a small audience but they watch every episode 3 times), shows are ultimately creative processes. Casts don't click. Leads become toxic. The audience just doesn't warm to it. No matter what the data said.

Sure, but that's not what I'm talking about here. I'm not claiming that Netflix is doing what would look like "intentionally trying and failing to make good content"—because that would produce qualitatively different content than the kind I'm referring to.

An attempt at rich food that tries and fails isn't bland food; it's bad rich food. A bad lasagna or a bad steak isn't going to suddenly taste like nothing. It'll just taste weird. Whereas a bland food is, say, a cucumber sandwich on white bread. Purposefully nothing-y. It won't ever taste exciting or weird.

Netflix is making "Netflix Original" bland TV, purposefully. Daytime soaps, "ghost investigations", prison documentaries, and other low-budget schlock. The data says people want it, just as much as the data says people want rich TV like Black Mirror or OitNB or whatever-else.

What if it isn't solvable? To me it doesn't look like there is enough room to maneuver, at least not in things like movies and TV shows, I think in music there might be. Say a recommendation needs to fulfill the intersection of 2+ things the user likes to feel like a recommendation, for example "you watched space and romance movies, here's a romance in space" otherwise it's just "you watched some space movies, here are some more". I don't think there are enough movies in existence, especially once you apply common criteria like "isn't in a foreign language, isn't on another streaming service, isn't very old or low budget" for these intersections to be filled with more than one or two items. In the space romance example above, the sole example I'm aware of is Passengers, there may be one or two others but it's not enough that you aren't going to deplete the pool even watching casually.

That only leaves extrapolating completely new things I might like based on more general patterns, but that falls victim to the same issue. If you know that I liked Passengers (already difficult without explicit feedback, watched does not mean liked), you have no idea whether I liked it because it had Jennifer Lawrence in it, or because of the tone, or the presentation, or the theme, or any other facet of it. And I only watch tens or hundreds of movies, not the thousands or ideally millions necessary to isolate the fact that I like movies with one very particular tone very slightly more than other movies.

In hindsight, that's obvious, but it struck me at the time. I don't ever want to see my thoughts get read by a computer (personal choice), but until such technology exists, recommendation engines will continue to suck.

Paul Lamere has done a number of talks on the topic at SXSW and the like. I know him from Sun Labs but I think he's at Spotify these days.

https://www.coursera.org/lecture/matrix-factorization/recomm...

You'd think, with enough data, you could build causal models of purchases of product types.

Amazon certainly has associative models (people who buy this also buy that)—it seems like they would just need to add a temporal dimension to the input data.

This is false. I remember in 2007/2008 using Netflix a lot to get some really amazing and hard to find movies. The same was true of their streaming service in its infancy. Now they seem to have abandoned their movie selection altogether and are focusing on making their own stuff. I'm guessing it's because licensing became a huge headache for them.

> No one has come close to solving recommendation.

Try this app and we'll talk:

https://www.coollector.com/help.html#recommendations

Recommendation is a solved problem from a computer science point of view. The problem with recommendation systems is that their outputs always get overridden by political and marketing concerns.

That's why I love content recommendation systems which actually work. Though the downside is, that today's systems (looking at you, Spotify) close you in kind of a bubble.

The best I've met was Google play music, the recommendations were great and at the same time you got to know a lot of great content unlike anything you've heard before! I switched, because unfortunately it seemed like nobody was working on it anymore.

I work 40-45 hours a week. Even when I am doing contract work, I limit the time I’m working. I don’t equate the 8 hours I spend working out to lost wages either.

This seems like it would make you a workaholic / crazy, but in fact at least for me, it makes me value the time more and make a bigger effort to spend it well.

It also makes it easy to decide if buying something that's meant to save me time is worth it, cause it's a simple "time saved" vs cost calculation.

Time spent sleeping is special, because I know how bad I function and unhappy I am if I sleep too short for an extended period of time and it actually equates to days wasted.

I don’t like doing yard work and I will happily pay the going rate for someone else to do it. I don’t like going to the gym or working out outside - By the time it is all said and done, I’ll spend way more than the accumulated hours I’m saving by not driving to and from the local gym.