The mistake we correct most often is optimising toward the lowest possible ACOS. You can always lower ACOS by cutting spend on anything less than perfectly efficient, and you end up with a small, tidy campaign making very little money. The target is total profit, not the tidiest percentage.
The second is treating launch and steady state as the same. During a launch, a deliberately high ACOS buys sales velocity, reviews and organic rank, all of which lower ACOS later. Judging a launch campaign on ACOS alone means killing the thing that was working.