Implementation of an AI-based monitoring system to proactively identify and analyze performances decay of AI models In production environment, ML models deal with real time data that are influenced by external variables, not always identifiable during model development and training. In the worst case, they change over time bringing the model predictions to be no longer useful. Data drift breaks the fundamental assumption of machine learning models: data distribution is static and past data is representative of future data. Therefore, ML algorithms face with a constant stream of new, changing data, and the need to be regularly updated in order to maintain their predictive ability.