How do you define the measures of success for a change you are specifying?

Tie them to the business objective and capture a baseline before the change goes live. Name the measure, the current value, the target, and who owns it, and pick something that would visibly move if the change works and stay flat if it does not. Without a baseline you cannot prove benefit afterward, and that number is much harder to get after go live.

I write the measure, the current value, the target, and who owns it, and I get the baseline before anything goes live, because afterward it is much harder to get and someone will always argue about what it was. On a process automation piece the measure was average handling time for a case type, and we spent two days pulling six months of history first so the comparison later was not a debate. I also try to pick a measure that would visibly move if the change works and would stay flat if it does not, rather than something that drifts anyway. And I am honest about attribution, because usually two or three things change at once. If we also restructured the team that quarter, I will say the improvement is a combination rather than claim the whole thing, since overclaiming once costs you credibility on every business case after it.