The challenge
The quality of underwriting is critical to a lending business. The client, an SME lending company, asked us to automate its underwriting process and challenged us to build machine learning models that would outperform the expert judgement of its underwriters.
The model had to be developed and tested on a limited set of loan applications, and external sources of creditworthiness information had to be tested and proposed.
Our solution
We developed an AI component that fully automates the underwriting process. It uses a machine learning approach to predict the probability of default for each potential borrower.
More than 800 features were extracted from 14 different data sources to feed the model.
The outcome
The model performed better than human underwriters and improved the health of the loan portfolio: fewer accepted customers went on to default, while the overall acceptance rate for new customers stayed the same.
Decisions became much faster, falling from a few hours or a day to a fraction of a minute, which also gave borrowers a better experience through quick answers on their applications.
