Resource allocation problems in which demand is splittable are usually solved using different solution methods from their unsplittable equivalents. Although splittable problem instances can be the easier of the two (for example, they might simply correspond to a linear relaxation of a discrete problem), there exist many problems, including routing problems, for which the converse is true. That is, the technology for solving unsplittable problems is mature, but the splittable counterpart is not. For such problems, one strategy that has recently shown potential is the use of an a priori splitting rule in which each customer's demand is split into smaller pieces in advance, which enables one to simply solve the splittable problem as an instance of the unsplittable version. An important factor to consider is the number of pieces that result after this splitting. A large numbers of pieces will allow more splitting patterns to be realizable, but will result in a larger problem instance. In this paper, we introduce a splitting rule that minimizes the number of pieces, subject to the constraint that all demand splitting patterns remain feasible. Computational experiments on benchmark instances for the vehicle routing problem and a time-windows extension show that the solution quality of our proposed splitting rule can match the performance of existing approaches.