FICO leverages machine learning (ML) in solutions ranging from fraud detection to marketing. In credit scoring, we combine the power and speed of insights derived from Machine Learning with our 30+ years of domain expertise in building credit scores to help ensure that the resulting credit scoring models are reliable, predictive, accurate, robust, and transparent for all consumers, including underbanked and “unscorable” populations, and withstand regulatory and lender scrutiny.
FICO has long-supported financial service institutions in fulfilling their obligations to comply with applicable adverse action notice requirements under regulations such as the Equal Credit Opportunity Act (ECOA). FICO® Score models, developed using advanced computational methods, including insights from ML, are designed to provide lenders with explainable reason codes to share with consumers for their credit decisions.
This issue is at the forefront of the discussion on Machine Learning models as discussed in a circular issued by the CFPB on May 26, 2022, reminding credit score developers and users of ML models that every applicant and borrower has the right under ECOA to a specific explanation if their application for credit was denied, and that right is not diminished simply because a company uses a complex algorithm that it may or may not understand.
However, over the last decade, new credit score providers have developed scores entirely or predominantly with ML, which they touted as more innovative and effective for fair, inclusive credit decisions for lending, particularly for underbanked and “unscorable” populations. Machine Learning is simply another analytic technique; one that can help produce highly predictive credit scores which must also be explainable, with two important caveats:
- The use of Machine Learning must be balanced with deep domain expertise in credit risk modeling.
- Appropriate safeguards must be put in place to make sure the resulting models are fully palatable, explainable and intuitive.
As an industry, we must be clear-eyed about the benefits that ML can provide. A fundamental challenge for underbanked and “unscorable” consumers is the lack of data or credit history to allow lenders to conduct a robust credit assessment. ML does not create new data. In fact, no analytic modeling technique including ML can solve what is in essence a lack of data problem. To drive greater financial inclusion across underserved communities, FICO has developed innovative new credit scores that incorporate new and robust alternative data sources such as telecom payments.
FICO research on “unscorable” populations with sparse traditional credit scoring bureau data has found that any minor improvements in accuracy found in an unconstrained ML model are negated when accounting for necessary constraints to address palatability and intuitive score dynamics.
While FICO is very heavily invested in explainable artificial intelligence (AI), we utilize Machine Learning as a complementary tool to traditional scorecard technology for deriving fast insights into the data rather than as the underlying methodology for arriving at the final model weights. This FICO® Score model development approach guarantees that explainability and palatability remain paramount in lending decisions.
Without a doubt, we are going to continue to invest in industry-leading research aimed at finding new and more powerful ways to combine the benefits of ML with our unparalleled domain expertise in credit scoring. Our mission is clear: to expand access to credit responsibly for more consumers around the globe while meeting our high standards of palatability and explainability.
To learn more about FICO’s research in the use of artificial intelligence and machine learning for credit scoring models in the financial services industry, please refer to the following white paper: https://www.fico.com/en/latest-thinking/white-paper/machine-learning-and-fico-scores.