AI and ML promise to revolutionize how risk functions use data, operate, and interact with the business. We bring integrated data sources, advanced modelling techniques, and automation technologies that clients need to support this transformation.
Our experience has shown that many institutions have machine-learning capabilities that can identify sources of value; few, however, manage to fully realize this value. These companies need a more agile approach to integrating digital capabilities with advanced analytics. Risk Dynamics helps companies adapt their operating model to incorporate an agile approach that enables fast-cycle prototyping, test-and-learn exercises, and the implementation of minimum viable products.
Risk Dynamics works closely with our clients’ analytics teams to:
Working with the data science team at a leading regional bank in the US, we designed and implemented a machine-learning algorithm to pro-actively manage credit lines. The team quickly prototyped leading-edge algorithms that could predict the optimal line for each client using a machine-learning sandbox.
As part of this effort, we assessed over 1,300 inputs to create a profile of each account—including variables such as product holdings, customer behavior, demographics, and interactions. Our machine-learning approach generated simple yet hard-to-find decisions rules that can predict accounts’ profitability and potential for default. We were then able to develop segment-specific strategies for credit management, with specific actions for specific trigger events. This allowed the client to reduce losses by 10 percent and to save $72 million in charge-offs.
Risk Dynamics worked with a leading bank in Germany to develop a machine-learning-driven recommendation engine for the bank’s collections function. Using machine learning, we identified the optimal treatment and contact strategy for each delinquent account, integrated the solution into the existing collections workflow and environment, and trained collectors to use the system and to collect additional data to improve model performance.
Key to the project’s success was the integration of new data sources on client behavior. We also leveraged multiple machine learning models to identify features of high-risk accounts. The models now work seamlessly with the contact center interface and automated touchpoints. Efficiency in early collections has increased by approximately 30 percent, and in late collections by approximately 15 percent. Retail delinquency-related write-offs have dropped by about 10 percent.