As the complexity of information systems evolves, there is a growing interest in defining suitable process models than can overcome the limitations of traditional formalisms like Petri nets or related. Causal nets may be one of such promising process models, since important characteristics of their semantics deviate from the ones in the literature. Due to their novelty, very few discovery algorithms exist for Causal nets. Moreover, the existing ones offer very few guarantees regarding the outcome produced. This paper describes an algorithm that can be applied as a second step to any discovery technique to significantly improve the quality of the final Causal net derived. We have tested the technique in combination with the existing algorithms in the literature on several benchmarks, noticing a considerable improvement in all of them.