Although our primary goal is to use molecular properties that can be computed for a single molecule some physical phenomena may not relate to these properties. In some cases it is necessary to prove that global properties can be predicted from single molecule quantities. Thus, the computations of ensemble properties will often be at least a first step in the model development.
So, e.g., when finally the interaction energies of similar molecules are found to be determined by the number of alkyl groups available for van der Waals interactions it is necessary to derive this hypothesis and to prove its applicability by explicit computations on a full ensemble of molecules first. If significantly different molecules have to be judged the interaction energies can not be estimated easily but only be computed explicitely.
The thorough analysis of the data from the ensemble calculations may then lead to a more general scheme for the molecular interactions that can be cast into an everyday applicable relation.
On the other hand the calculations allow to determine quantities that are hard to measure experimentally. This extended data basis will improve the quality of the mathematical relations and thus the reliability of the predictions.