WHY

A major Canadian bank was interested in developing artificial intelligence technologies that would enhance its understanding of its credit card business. The bank and Dessa collaborated on a machine learning model that aimed to learn who among their customers would be at the highest risk of not paying their credit cards back. Conceptually, the model would rank customers on their the likelihood of paying off their credit cards as well as identify whether a customer is due to pay their credit payment before it becomes late. The bank’s previous model was believed to take a longer time than what executives needed to arrive at to take action on.

HOW

After working closely with the bank’s business and technical teams, Dessa built a model using advanced AI techniques, notably stacked non-linear combinations, that incorporated more than 500 variables and attributes in the system. In order to generate accurate results, the team told the model what output it expected to generate and inputted more than a million customer data points. Once a pattern was observed, the level of accuracy was measured against the bank’s internal benchmark to determine whether a customer needed to be contacted.

RESULTS

After running the model through the bank’s credit card customer data within a six-month period, Dessa’s AI technology was significantly better than the bank’s earlier work while saving the bank more than three times the original cost estimate. Because of that, the bank switched its credit card risk repayment model over to the Dessa. Using Dessa’s technology, the bank has also improved its ability to track customers who are most at risk of being late or missing credit card payments, specifically those who forgot to pay or who are actively avoiding making a payment.