Complex transmission concepts may enable high fuel efficiency but require much effort in controller development. This effort should only be spent if the potential of the concept if high, a potential which can be determined using optimization techniques. This paper examine the use of stochastic dynamic programming for transmission potential evaluation, applied on a wheel loader. The concepts evaluated is the present automatic gearbox and a multi-mode CVT (MM-CVT). A probabilistic driving cycle is created from a measurement including 34 loading cycles. Trajectory optimization is performed both against probabilistic and deterministic cycles. The paper shows that the introduction of a probabilistic load highly affect the application of optimization. It is also shown that the MM-CVT has approximately 20% lower minimum fuel requirement than the present transmission, and that this number is not sensitive to the quality of the prediction.