There's something weird about cost-cutting metrics where they just aren't as impactful as they sound. FB has enough money floating around that many teams will write big systems inefficiently in the name of "moving fast" and then go back to optimize later. This is also a hack for performance reviews as you get to claim impact twice - once for making the thing, and again for saving XX million dollars (even though it's just cleaning up your own mess). Management is somewhat aware of this hack.

I would say at FB it's shockingly easy to find a 10 million dollar efficiency win if you're looking for one, and I certainly shipped multiple things in that range. This wasn't really the probability team's charter though so they wouldn't get fully rewarded for it.

The probability team had some genius engineers working hard on some very interesting long-term investments, but AFAIK they hadn't really shipped any of their core products so they were a good target for layoffs. I am shocked FB fired the engineers instead of just moving them around though...