Statistically and probabilistically speaking, we can make an estimate (with error bars) of the value of the treatment, even with a trial of only a few people.