Case study
Improved loan decision making
Assisting lenders to make data-driven decisions for loans nearing default.
Challenge
Solution
Lenders have different options to manage loans nearing default, such as restructuring, forbearance, refinancing, and more.
Traditional machine learning methods often fall short of predicting the best loss mitigating strategy, producing inaccurate forecasts due to confounders.
Our goal was to develop a model that, based on available data of the borrower, recommends the best action to the lender.
Impact
Assessed and cleaned the available historical loan data.
Enriched the data with additional macroeconomic factors.
Used expert knowledge and data-driven approaches to identify the causal predictors for the loan loss.
Trained a Causal AI model on historical data and the causal predictors to accurately predict the outcomes of the lenders action on the loan loss.
Validated the trained model on hold-out data and on new data accumulated by the client during the project.
Based on the trained Causal AI model, available options to the lender, and data about the borrower, the loan officer is recommended the action predicted to incur the minimal loss.
Developed a causal model that more accurately forecasts the impact of decisions on loan loss than traditional machine learning techniques.
Delivered POC system that recommends best actions to the loan officer.