How do you handle experimentation when your sample size is limited to a small number of large enterprise accounts?
