- Equifax recently debuted a machine learning credit scoring system that uses neural network modeling to help customers assess risk more accurately. The NeuroDecision Technology overcomes a previous challenge the company faced using neutral network applications — generating understandable and explainable outcomes — and will help customers and regulators better understand consumer and commercial risk models.
- Equifax is not the only credit bureau dabbling in AI solutions. FICO uses AI to first segment customers into more similar categories for credit risk models and then to generate more accurate scorecards with greater explainability of results.
- VantageScore uses ML algorithms to assess customer risk and assign scores, even to "credit invisible" consumers without recently updated credit files. Experian augmented its analytics tools with ML functionalities last year, using open-source ML and deep learning technology to drive deeper, on-demand insights.
As Equifax continues down its recovery path from a data breach that has already cost $114 million and counting, reestablishing trust in its products and services and keeping up with the latest in fintech is crucial.
Equifax's machine-learning application is unique because it is "regulatory-compliant," according to Peter Maynard, SVP of Equifax Global Analytics, in a statement to CIO Dive. Equifax's NDT framework generates reason codes with output, which improves model accuracy.
"While other companies use machine learning methodologies in portions of the model building process, the final score is not based on a machine-learning regulatory compliant model," said Maynard. "These techniques inherently lack the explain-ability and transparency that is required for compliance and regulatory purposes."
Equifax began researching AI and ML applications for credit scoring systems in 2015 and is looking to bring more modeling technologies to the market, said Maynard. The company's synthetic identity alert solution also uses ML technology to assess "user velocity and identity discrepancies."