Comment by kridsdale3

I really appreciate the reply, thanks for sharing that.

> Every time all the usage metrics tanked.

What if that's exactly what people want? Less usage of Facebook (horrifying, I know -- it can't be true, right?), with a focus on friends etc. when they do use it? I know you'll dislike the analogy, but isn't all that different from smoking. You think usage metrics tanking implies the outcome is bad... why exactly? Is it that unthinkable that less quantity and more quality is better for people, and what they actually want?

> So people say they want this, like they say they want McDonalds to offer salads. Nobody orders salads at McDonalds.

You seem to be missing that the people who have the means to eat out wherever they want don't eat at McDonald's every few hours. They go in moderation. They actively want to avoid McDonald's most of the time. Once in a while they get a craving, or get super hungry and don't see other options, etc. and they cave in and go there. Of course the get the tasty unhealthy option when they go, but it's foolish to think they prefer to eat McDonald's all the time. (Do you seriously believe that??)

>Several times we ran A/B tests with many millions of people to try exactly this. Every time all the usage metrics tanked. Not just virality and doomscroll metrics, but how many likes, messages, comments, re-shares, and app-opens. We never even measured ad-related things on that team.

Well, yeah, but this has an implicit "engagement === good" assumption. Exactly the same thing that incentivizes unhealthy McDonald's food: they make more money when they sell food that still leaves you hungry. So, yeah, people probably did want this, and when they got it they started using Facebook in a healthy manner (no point opening it at every available moment to just scroll through 'new' trash), which tanked your metrics. If you're actually worrying about your users you should also consider that them using your product more might not actually be what they want or need.

Ironically enough, I think the same mistake (or rather, it's more of a mistake because there's not quite such a naked financial incentive to make this worse for the affected users) has happened with the youtube analytics dashboard: multiple youtubers have said that it's actively addicting and really bad for their mental health, but any change that feeds that probably looks really good in their metrics because, hey, creators are using it more, that must mean it's good, right?

I'm sure there's more that could be shared about how "wants" were determined, which would counter my off-the-cusp thought here, but anyways:

Yes... my ideal would be for facebook feed to be a once-a-week addiction (maybe a bit more) where I go, see what's new, and clearly hit an end point where I know I'm seeing things I've seen before. But I'm also part of the "problem" in that I post myself maybe twice/year now.

I'd suspect the current doomscroll-y feed like we have now/you were working on reduces my likelihood of "interacting" with friends' posts. "Do I make the effort of commenting, or lazily keep scrolling to the next-often-good 3rd party content?"

A year or two ago, I copied some greasemonkey type script off reddit, and that nuked all the non-friend content off my feed, but that stopped working a couple months later and I haven't been strongly enough motivated to find an updated approach. I have little enough friend activity that I'd easily notice when I hit old content.

The current doomscrolling feed of algo content sure does manage to hook me, so that's a nice indicator of the current team being successful :P

Did you consider that you are gaming your own setup rules of measurement?

It's like "look nobody is ragebaited anymore, that's very bad for clicks"

Guess what, you should not have used that as a means of measurement before, but it was the cheapest way to sell it to advertisers.

If you have incentive to create a shithole of engagement, it's what you will get in return.

There's an old saying: you can never get enough of what doesn't fill your need.

For example, when you need sleep, you can't eat enough to make you not tired, but you may well pound a lot of caffeine and sugar.

If true, this would accomodate the simultaneous truths that:

(a) users accurately report their preference chronological friend connection when they come to a social feed

(b) users spend more time engaging with a social feed when the need they come to fill has irregularly payoffs

That you can get more engagement by not giving them what they want/need (or giving them what they need irregularly) wouldn't mean that they are lying to you, it would simply mean that engagement and social payoff curves aren't the same, and the incentives to drive one might not optimize the other.

You're saying that users weren't using the app enough like it's a bad thing. Users saw the tool as useful and used it.