Agroforestry systems (AFS) can provide multiple ecosystem services (ES), partly due to their high (agro)biodiversity. However, multi-criteria analyses studying trade-offs between multiple ES and exploring AFS management optimization paths are still scarce. Routine methods, such as regressions or weighted/non-weighted scorings, may reveal unsuitable because data collections hardly meet rigorous statistical designs and knowledge about ES can be limited in such complex systems. In this paper, we explore a novel approach based on algorithms identifying Pareto fronts to check for management schemes which favour the multi-functionality of complex agroecosystems. We based our study on the ground truth data from 58 cocoa-based AFS fields in Cameroon and chose to study three ES: cocoa production, aboveground tree carbon storage and natural pest control. The combination of expert knowledge and Pareto front algorithms enabled us to identify four clusters of increasing ES provision among the 58 plots: “bottom”, “low-yield intermediate”, “high-yield intermediate”, and “top”. Significant differences in associated tree communities and management strategies were identified across the four clusters. While highlighting clusters of AFS with common management strategies, the use of the Pareto front algorithm enabled us to draw significant lessons on cocoa-based AFS despite their high complexity. We believe that such an approach can be used to design suitable benchmarks for the study and improvement of multiple ES provision in complex agroecosystems.