Most commercial models (PIMS, RPMS, PETRO) perform refinery planning under deterministic conditions, that is, they do not consider uncertainty in process, demands, refinery parameters, etc. and as a consequence, they are unable to perform risk management. Although risk management is attractive to refinery planning operators, its development has been considered hard because it entails the extension of these deterministic models, complex as they are already, to perform optimization under uncertainty and manage risk. The extension never posed conceptual problems, just possible computational problems (running time, memory, etc) and eventually business will to pursue this on the part of software vendors. A variety of methodologies for risk management in engineering decision have been already developed. We follow the approach presented by Barbaro and Bagajewicz [1], who used two-stage stochastic programming and where the reader can find all other approaches analyzed and discussed. All of them, including the one by Barbaro and Bagajewicz [1], presented computational challenges and if implemented commercially would require changes in the available commercial code. To deal with the aforementioned computational difficulties, Aseeri and Bagajewicz [2] proposed a methodology that is capable of performing risk management using a deterministic model repeatedly. The methodology is conceptually rigorous and practically sound. It removes the need to alter existing code. Only new code needs to be generated. The methodology proposed by Aseeri and Bagajewicz [2] was applied to refinery planning by Pongsadki et al. [3], who used a linear model as the core deterministic planning solver. In this work we implement the strategy outlined by the aforementioned previous work using a commercial planner. We use PIMS as engine to solve the stochastic model and write computational routines to do it and manage financial risk. The results show that the procedure found solutions with higher expected value than those suggested by the deterministic model. We present more details now and cover some computational issues later.