We provide an alternative way to model sequential decision processes, which is consistent with the random utility maximization hypothesis and the existence of a representative agent. Our result is stated on terms of a direct utility representation, and it does not depend on parametric assumptions.