The Operational Risk Advanced Measurement Approach requires financial institutions to use scenarios to model these risks and to evaluate the pertaining capital charges. Considering that a banking group is composed of numerous entities (branches and subsidiaries), and that each one of them is represented by an Operational Risk Manager (ORM), we propose a novel scenario approach based on ORM expertise to collect information and create new data sets focusing on large losses, and the use of the Extreme Value Theory (EVT) to evaluate the corresponding capital allocation. In this paper, we highlight the importance to consider an a priori knowledge of the experts associated to a a posteriori backtesting based on collected incidents.