Actually, I was looking for some methods which can ensure randomization on future data in A/B testing. Say if I am launching an experiment for a offer on a product, and I created two groups, control(A) and target(B). How do I make sure(or maybe, just ensure at best) that B does not consist of all non-buyers (or by random chance, all purchasers go into A). While preparing training/test data I can ensure this since its an offline process and I already have class distributions from each of groups.