Even if immaterial capital represents an increasingly important part of the value of our enterprises, it’s not always possible to store, trace or capture knowledge and expertise, for instance in middle sized projects. Email it still widely used in professional projects especially among geographically distributed teams. In this paper we present a novel approach to detect zones inside business emails where elements of knowledge are likely to be found. We define an enhanced context taking into account not only the email content and metadata but also the competencies of the users and their roles. Also linguistic pragmatic analysis is added to usual NLP technics. After describing our model and method, we apply it to a real life corpus and evaluate the results based on machine learning experiments and filtering algorithm.