Part of the problem is that p-values are not the best indicator in all applications. One question in my work is whether a process change affects the yield. A

Part of the problem is that p-values are not the best indicator in all applications. One question in my work is whether a process change affects the yield. A confidence interval of (-1%,+1%) is much different than (-20%,+20%), even though they would look the same if I was just interested in the p-value. We might also accept changes with a (-1%,+10%) confidence interval. We can't 'prove' that yields would increase, but there is significantly more upside than downside.