Including employee preferences in a shift-scheduling scheme raises the question of how to aggregate employee satisfactions in a sensible manner. To do so, we first need to model the employees' preferences, which can be done using so-called utility functions and a multi-attribute approach like MACBETH. Even then, though, we still need to make sure that it is sensible to compare or to add two different utility functions that concern different attributes. To do so, an appropriate normalization of utility functions is required. In this paper, we first discuss a naive extreme value normalization (EVN), which has utility values ranging all along two predetermined values, e.g. between 0 and 5. Then, we propose an alternative normalization, called AMACSN, which relies on a description of the "typical outcomes" produced by a shift-scheduling scheme. We then give both the conceptual grounds and the computational results to argue that AMACSN is more relevant and more meaningful than EVN to address the comparison or addition of different utility functions. As a bonus, we observe that AMACSN induces greater fairness.