Yield potential has major interests in agronomic modelling such as for the computation of fertilizer recommendations. Most precise estimates of this potential can be generated with crop modelling approaches. However, such potential is often theoretical because it has never been obtained in the field and requires lots of input data to be computed, e.g. soil characteristics, management practices. In this study, a decomposition of this theoretical yield potential into a series of intermediate and more realistic potentials, denoted here production potentials, is proposed. This work comes along with a simple method relying on historical high-resolution yield datasets to compute site-specific production potentials and production gaps, i.e. the differences between actual yield and production potentials. Application of the methodology to a real database of yield mapping has proved of interest to identify areas with similar production potentials but different production gaps. Advantages and concerns regarding the use of such production potentials are discussed at the end of the study.