Author: Raabiya Singh
The journey of artificial intelligence has been a long and intriguing one, punctuated by significant milestones that have reshaped industries. From the introduction of the first ATM in the banking sector to the gradual automation of various processes, AI has evolved into a powerful tool capable of handling tasks that were previously the domain of human expertise. As technology progressed, AI further manifested itself in the form of teller machines, enabling streamlined transactions and freeing up human resources for more intricate tasks.
AI in Lending: Enhancing Efficiency and Accuracy
Lending has witnessed a remarkable transformation with the infusion of AI. One of the key areas where AI has proven its worth is in the automation of lending processes. This automation streamlines the decision-making process, leading to faster approvals and an improved customer experience. Additionally, AI has the capability to assess thin-file cases and cater to the specific needs of underserved demographics who often possess unconventional and/or limited credit histories.
Mitigating Bias and Expanding Opportunities
As AI becomes an integral part of lending operations, compliance remains a top priority. The Consumer Financial Protection Bureau (CFPB) guidelines serve as a critical framework in ensuring ethical and lawful AI implementation. Adhering to these guidelines ensures that AI-powered credit decisions are free from any form of bias and uphold fairness in lending practices.
A recent circular from the CFPB that reiterated the importance of adhering to adverse action notice mandates within the ECOA Act. Following which many credit union leaders were left in a dilemma on how they would explain their credit decisions. However, modern AI solutions have progressed beyond mere explainability facets. They now facilitate equitable lending and have gone the extra step in mitigating inherent biases that might emerge within data-driven models.
How does AI in lending work to bridge this gap and reduce bias? One aspect involves moving beyond traditional data sources for credit risk analysis. AI Models focus on relevant or inclusive data, evaluating a member's creditworthiness even with limited credit history. Exploring FCRA compliant alt-credit data like, service payments, rent and property records can accurately assess an applicant's creditworthiness, especially when credit file is lacking.
The next step involves using AI to review past data for any hidden biases and examine new choices with the perspective of potential“Disparate Impact”. Disparate Impact Analysis (DIA) quantitatively gauges the adverse treatment of protected classes. Through an examination of member data, AI assists in distinguishing between unintended biases that might be present, and the genuine elements that should be employed to assess an applicant's creditworthiness in a fair and consistent manner.
Lets see how this works - Consider a scenario where a black male exhibits elevated utilization and a substantial number of inquiries. Under conventional credit scoring methodologies, he might face rejection. However, a closer examination of additional factors reveals a different picture. For instance, there are no derogatory marks on his record, he holds valuable assets, and he has maintained a stable address for the past 60 months. This comprehensive evaluation alters the perspective of his creditworthiness, furnishing solid grounds to endorse his loan application.
“We have been using the AI for over a year now. In that time we got an expanded field of membership. Now we serve the entire state of South Dakota so we are getting applications from all over which is new for us because smaller credit unions usually serve a small area and we are very familiar with our members. In South Dakota we have a very large native american population. A lot of those people have unconventional credit histories. But with AI we are very pleased that we are getting consistent answers on approvals. AI doesnt take into account human error/bias, it is all numbers driven, information driven. With the AI algorithms and FCRA compliant alternate data we are using more information than we were in the past. From a fair credit standpoint the decisions we get for application A,B and C and no human bias making distinction between a last name, or an address or anything like that.”
Floyd Rummel, III CEO Northern Hills Federal Credit Union
Regulatory Landscape: How AI Fits into the Compliance Puzzle
Scienaptic’s AI underwriting platform have gone beyond to integrate transparency and fairness into their models. The foundation of constructing transparent and unbiased models rests on three fundamental principles: : transparent adverse action reasons, comprehensive disparate impact analysis, and well-elucidated documentation of model/decision parameters.
Adverse Action Reason (AAR) and letter
Under the Equal Credit Opportunity Act (ECOA), creditors must give a written explanation with specific reasons to applicants who face adverse actions. This requirement is guided by Regulation B, stating that the explanation "must be specific and indicate the main reason(s) for the adverse action." Our AI platform has embraced this practice, ensuring that credit decisions always come with clear and primary reasons. Scienaptic AI is refining the language of these explanations to be simple yet complete, allowing consumers to easily understand and take meaningful steps towards adjusting their credit behavior in the future.
Adequate testing for fair lending
The Equal Credit Opportunity Act (ECOA) forbids creditors from discriminating against applicants based on race, color, religion, national origin, sex, marital status, age, income source, or the exercise of rights under the Consumer Credit Protection Act. Similarly, the Fair Housing Act bars discrimination in real estate transactions. AI platforms tackle historical training data biases by using thorough disparate impact analysis, which examines potential discriminatory effects of credit decisions. A disparate impact arises when seemingly neutral policies harm protected groups. Scienaptic AI meticulously reviews policy attributes, ensures unbiased model design, and conducts thorough tests to identify discriminatory effects on protected classes. Through continuous adaptive assessment, AI models correct inadvertent biases and establish equitable lending models.
Documentation of model/decision parameters
Model Risk Management and its documentation play a crucial role in credit risk decisions. The guidance that shapes this area primarily comes from the Federal Reserve and the Office of the Comptroller of the Currency (OCC) through Supervisory Guidance on Model Risk Management (SR11-7). This guidance is designed for both banking organizations and supervisors to evaluate how organizations manage model risk. It's relevant to all banking organizations under Federal Reserve supervision, considering their size, nature, complexity, and use of models.
Scienaptic AI employs a comprehensive approach to ensure the effectiveness of its model documentation. This involves conducting a rigorous assessment of data quality and relevance, as well as implementing tests to verify the proper functioning of various components within the model. Additionally, the platform focuses on evaluating the model's robustness and stability under different conditions. It also takes into consideration the limitations and assumptions inherent in the model, thereby ensuring a well-rounded and reliable AI solution.
“I am looking at AI for a reason to approve a loan. Show me those diamonds in the rough, those low FICO, low Vantage scores. If they paid their rent, cellphone bills, medical bills, let me find those people and find reasons to approve them. Once I have done that I am not worried about any reason codes from an adverse action point of view. I think where a lot of lenders get frightened, is when it comes to declines. Just because you have AI, doesnt mean you cant auto decline something, but you have to make sure you decline based on your FCRA compliant reason codes. The purpose behind using AI is to approve loans quickly and efficiently.”
George Sellito Chief Lending Officer WyHy Federal Credit Union
Compliance at the heart of AI powered credit underwriting
AI holds the potential to create a more inclusive lending landscape by reducing bias and expanding opportunities. By analyzing vast amounts of data, AI can identify patterns that human judgment might overlook, resulting in more informed credit decisions. This leads to increased approval rates, especially for individuals with thin credit files or those from underserved communities.
The impact of AI-based credit decisions is not just theoretical; it's tangible. Numerous success stories have emerged, highlighting the transformative power of AI in lending. Credit unions leveraging Scienaptic’s platform have seen impressive outcomes, such as higher approval rates and the reduction of racial bias in credit decisions with some of them experiencing upto 26% rise in approval rates for protected classes including Hispanics, Blacks, Asians and Senior citizens. These real-world examples underscore the potential of AI to bring about meaningful change in the financial industry.
The convergence of AI and compliance presents an exciting future for the banking and lending sectors. As AI continues to evolve, it has the potential to reshape the way lending decisions are made. By combining data-driven insights with regulatory compliance, AI can pave the way for more transparent, efficient, and equitable credit assessment processes.