[RA_Fisher]

Using Bayesian methods and some relatively rarely used Frequentist methods you can achieve what you suggest. Maybe you'll take it from a statistician that you're being grossly naive about experimentation.