Socure, the leading provider of digital identity verification and fraud solutions, today at Money 20/20 announced the industry’s first Digital Identity Fairness and Inclusion Report. Based on its research with large financial institutions, the report revealed how ML and AI can dramatically improve financial inclusion, democratize access to services and benefits as well as eliminate unfair, bias-driven friction. The goal of the research was to evaluate how ML and AI used by Socure reduces the negative ethnic, racial, age, and socioeconomic bias that can easily creep into digital identity verification solutions and the risk scoring of certain fraud models.
Marketing Technology News: Weave Communications Files Registration Statement for Proposed Initial Public Offering
“Furthermore, Socure continuously monitors and evaluates production Sigma Identity Fraud models using clients’ feedback data ensuring these are not only the most accurate, but also the most inclusive.”
Report Key Findings and Takeaways:
- Socure ran a head-to-head accuracy and auto-acceptance rate comparison against a legacy identity verification provider evaluating hundreds of thousands of transactions. To objectively evaluate the increase that Socure’s advanced CIP solution would provide over the incumbent provider, Socure ran the test based on the exact CIP rules used by the incumbent’s solution. This established a fair comparative analysis of auto-approval rates across each of the key demographic segments.
- ML and AI used by Socure provided the following increases in auto-approval rates:
- <25 year-old: 99% increase
- Asian: 46% increase
- Hispanic: 36% increase
- Black: 28% increase
- If used exclusively across the country’s financial services industry, Socure’s technology could more than double the amount of automated approvals made in the US in any given year in the growing immigrant population, the 45 million+ group of credit invisibles, especially younger, thin-file 18-25 year olds, as well as minority populations who have historically been subject to unfair bias.
- For the ML fraud model portion of the analysis, Socure tested hundreds of thousands of transactions and evaluated Socure’s Sigma Identity top 200 predictors. All variables were tested across multiple demographics: race, ethnicity, sex, age, socioeconomic class, and immigration status. Socure used the variable neutralization methodology to identify negative effects of individual variables on the accuracy of the model output for each key demographic. Across all demographics tested, there were no statistically significant variations (e.g. bias) associated with any of the top 200 Sigma Identity predictors analyzed.
- Unfair bias occurs when a financial institution applies a practice (e.g. identity risk assessment) uniformly to all applicants while, at the same time, that practice has a discriminatory effect on certain segments of the population. Unfair bias can cause increased friction seen by younger and underserved populations in the exploding new account originations across digital channels that can result in abandonment rates of 50% or more, limiting that population’s ability to access some of the industry’s most beneficial financial products as easily and seamlessly as other socioeconomic classes.Marketing Technology News: MarTech Interview with Johann Wrede, CMO at Xactly
“No ML models at Socure get deployed into production without rigorous performance, sensitivity and fairness bias testing,” said Pablo Abreu, chief product and analytics officer at Socure. “Furthermore, Socure continuously monitors and evaluates production Sigma Identity Fraud models using clients’ feedback data ensuring these are not only the most accurate, but also the most inclusive.”
Detailing the Demographics
Credit invisible groups continue to be negatively impacted by traditional customer identification programs’ (CIP) primary reliance on credit header data. According to the Consumer Financial Protection Bureau (CFPB), approximately 45 million people within the US aged 18+ have limited credit history. This group of people are either not able to be scored or have no credit bureau files, rendering them credit invisible. As a result, this segment of the population is effectively shut out of the financial system that is available to the other 80% of the population. In addition, according to a Citi Global Perspectives and Solutions economic impact report, about 15 percent of Black and Hispanic populations are credit invisible (compared to 9 percent of Whites and Asians), and an additional 13 percent of the Black population and 12 percent of Hispanics have unscored records (compared to seven percent for White consumers). These differences are observed across all age groups, suggesting that they materialize early on in the adult lives of the consumer population, and persist thereafter.
Also, the explosive growth of US foreign-born populations will create further inequalities if entrenched players’ CIP technologies are not enhanced. The US foreign-born population reached a record of 44.8 million people in 2018. Beyond having reduced information footprints, these growing immigrant groups present an identity verification challenge for digital account openings because they more often use different surname and first name characteristics as well as hyphenated names and apostrophes.
Marketing Technology News: MarTech Interview with Michael O’Connell, VP Marketing at AnyRoad