I agree, but having a cap that is "the lesser of 15% of your base salary or $50K per year" means that people are going to start trading authorship of the rewards when they're anywhere close to this.

Like others have said, I think this will also encourage not thinking about optimising running costs when working on a feature - anything that does the job and doesn't seem outrageously expensive at first is fine. And then you wait until suddenly that service is costing a lot to run and pull the fix out of your back pocket.

Also, tying the reward to the savings quite so explicitly (and particularly having graduated levels) provides an incentive to ignore problems until it hits the next cost threshold. Even if you see the problem now, you might as well wait a year before reporting and fixing it, because then the bonus payout will be bigger. Also, then you're playing a game of chicken with other engineers who may also have noticed the problem but are also waiting for the best time.