Retail Modeling and Analytics team in Scotiabank were tasked with solving the business problem of maximizing risk adjusted margins within Scotiabank's risk appetite. The team used new predictive models to solve the problem. The team experimented with a lot of machine learning techniques for predictive modeling with the aim to better discriminate risk for one of Scotiabank's retail lending portfolios. The criteria used were a combination of improved predictive power, stability and interpretability. The bank chose an ensemble method and modified it significantly to retain interpretability with minimal loss of predictive power. The bank augmented its sources of data to include some new data sources, and utilized the ensemble method in conjunction to achieve upto 70% incremental predictive power. Subsequently, the bank has been scaling this framework to other retail lending portfolios as well.