I came up with this idea of artificially slowing down expensive models within a company to incentivize people to use cheaper ones. I'm curious - has anyone tried this at scale?