Credit has always been a fundamental part of both formal and informal financial services. In the last three decades, however, the exponential growth in data and computing power has led to new ways of assessing creditworthiness.
Artificial intelligence (AI) and machine learning (ML) have opened up the possibility of scoring new and alternative data sources, either to complement or to replace more traditional lending methodologies. But how do financial services providers (FSPs) ensure these new systems are both efficient and fair? Amidst the backdrop of a rapidly changing credit landscape, this practical field guide walks executives and data scientists alike through recommendations for ensuring that revised and new credit scoring methods are not unintentionally excluding women.
This guide combines academic work on bias detection with practical experience analyzing administrative data from real lenders working to increase financial inclusion around the world. The diversity of institutions this report references offered a natural test for generalizability of a core set of easy-to-understand bias detection questions. Although our focus is on detecting gender biases, the same tools and principles can be applied to bias detection for any underrepresented group.
Detecting bias is not a superfluous exercise. For financial institutions, knowing where bias exists can serve as a way of identifying overlooked markets (as is the case with rejected applicants who are highly creditworthy); maximizing the value of current customers (for example, those who are not receiving sufficiently large loans); or proving alignment with regulatory or legal compliance (in demonstrating the likelihood of a credit offer among men versus women, for instance). For customers, an institution attuned to bias detection is more likely to provide equal opportunities for business growth for men- and women-owned businesses. For regulators or policymakers, bias detection processes that ensure fairness contribute to broader economic participation.
This report has three main sections. The first section is a primer on the fundamental concepts of bias and fairness that anyone working in lending should know. In the second section, for the more technical readers, we discuss the statistical foundations of bias audit. The last section offers three examples of bias detection from three different institution types, as well suggestions on potential bias mitigation interventions specific to the institutions’ context. This report is relevant for all lenders, even if most of our examples are from institutions using more automated and digital processes.
If you can’t see the forms, please disable your AdBlocker
