It seems like lots of consumer products relying on advertising as a revenue model fall prey to the same problem.

They aggressively optimize for a panoply of metrics that make them appealing to advertisers. Often this means hacking the algorithms governing the interactions between users (viewership, timeline visibility, feed) until the outcome yields satisfying metrics, meaningful or not.

At this point a complex web of chaotic feedback loops has formed and it becomes hard to change anything. The system's stateful complexity far exceeds the engineering/product team's understanding of it. It's like trying to debug a neural network.