A method is developed to exploit data on complex materials behaviors that are impossible to tackle by conventional machine learning tools. A pairwise comparison algorithm is used to assess a particular property among a group of different alloys tested simultaneously in identical conditions. Even though such characteristics can be evaluated differently across teams, if a series of the same alloys are analyzed among two or more studies, it is feasible to infer an overall ranking among materials. The obtained ranking is later fitted with respect to the alloy’s composition by a Gaussian process. The predictive power of the method is demonstrated in the case of the resistance of metallic materials to molten salt corrosion and wear. In this case, the method is applied to the design of wear-resistant hard-facing alloys by also associating it with a combinatorial optimization of their composition by a multi-objective genetic algorithm. New alloys are selected and fabricated, and their experimental behavior is compared to that of concurrent materials. This generic method can therefore be applied to model other complex material properties—such as environmental resistance, contact properties, or processability—and to design alloys with improved performance.