Optimizing the Power of Action Learning