Artificial Intelligence presents an opportunity to transform how we allocate credit and risk so that we can create more fair and inclusive systems of lending. Its ability to allow lenders to avoid (or have alternatives to) traditional credit underwriting and scoring systems that help perpetuate existing bias makes it a high value use case that can be both disruptive and lucrative. From a socio-economic standpoint, AI can help take racial bias out of commercial and business lending by assessing all of a lender’s application submissions at once versus relying on humans to implement biased processes to shortlist underwriting candidates. From a financial scalability standpoint, using an automated top-of-funnel process that eliminates the bias by using both traditional and alternative credit scoring to shrink the initial pipeline to optimal capacity is highly cost efficient and accretive to a lender’s bottom line.
According to a report by Lending Tree (TREE), examining HMDA 2020 data set, African American borrowers have the highest denial rates for mortgages at 17.4%, and non-Hispanic white Americans have the lowest at 7.9%. Large shares of Americans say there is at least some discrimination against several groups in the United States, including 80% who say there is a lot of or some discrimination against Black people, 76% who say this about Hispanic people and 70% who see discrimination against Asian people.
Generally, lenders assess credit risk based on a number of factors, including your credit/payment history, income, and overall financial situation. Also known as the “5 Cs”, lenders generally look for:
Credit history: Nearly all lenders look at credit scores and report because it gives them insight into how people/businesses manage borrowed money. A poor credit history indicates an increased risk of default
Capacity: Lenders will look at income and employment history to determine the ability to repay the loan
Collateral: If applying for a secured loan, such as an auto loan or mortgage, the lender will want to know the value of the collateral that can sold to repay the loan
Capital: Lenders will also consider capital when evaluating loan applications. Capital refers to the assets you have available that could be used to repay the loan if you were unable to make payments.
Conditions: Finally, lenders will consider the conditions surrounding the loan application, to include use of funds and current market conditions
Lending discrimination, however, occurs when lenders base credit decisions on factors other than the applicant’s creditworthiness such as race, color, sex, religion, familial status, nationality, age, receipt of public assistance or disability. While Fair Lending laws are meant to protect against and provide recourse for lending discrimination, it still occurs and there are enough headline cases that the government has mandated data collection to investigate unfair practices in business lending.
AI can underwrite commercial and business loans by using artificial intelligence and machine learning algorithms to assess loan applications and determine a borrower’s creditworthiness. This process can be faster and more objective than traditional underwriting methods, as it is based on data analysis rather than human judgment. More importantly, it can leverage alternative credit profiles that can help remove racial bias in lending by providing a more comprehensive view of a borrower’s creditworthiness.
Financial institutions use AI to improve predictive models, which can make data analysis and credit risk assessment more efficient as it allows large quantities of data to be analyzed quickly. AI can expand credit availability for those whom creditworthiness can be measured using nontraditional metrics.
In credit decisioning, AI allows the use of nontraditional and unstructured data, which enables lenders to utilize their predictive analytics to improve their credit qualification, limit assessment, and pricing process to better serve underbanked and unbanked customers. Previously, lenders have been using rule-based or logistic-regression models, which relies on a narrow set of criteria, to analyze credit bureau reports to determine if a customer qualifies for a particular type of loan, leading to unequal access to credit for consumers and SMEs who lack a formal credit history.
In recent years, however, AI has allowed leading financial institutions to develop complex models for analyzing structured and unstructured data, enabling lenders to examine vast data points collected from unstructured data such as social media, e-commerce expenditures, SMS, emails, browsing history, telecommunications usage data, and more. By leveraging optical character recognition (OCR) to extract data from both conventional and new sources of data, lenders can now automate both its credit qualification and limit assessment process. This automated decisioning process can be completed nearly instantaneously, enabling lenders to predict the likelihood of default and customers’ capacity to pay for unbanked and underbanked consumers and SMEs.
Furthermore, banks have generally offered highly standardized loan rates, with relationship managers having some flexibility to adjust rates within certain thresholds. Traditionally, borrowers with weak risk scores are at a disadvantage and end up with more expensive loans. AI-first lenders, however, have been able to offer competitive rates to this group of borrowers while keeping their risk costs low by using Natural-language processing (NLP) to analyze unstructured transcripts with sales representatives and collections personnel to determine customers’ credit worthiness.
On the subject of alternative credit profiles, while the “5Cs” provide a consolidated view of an applicant’s credit risk, these factors may not always accurately reflect a borrower’s ability to repay a loan. Alternative credit data can provide additional information about a borrower’s financial situation that may not be reflected in a traditional credit score. This can include information about rent payments, utility bills, and other non-traditional sources of credit that could decrease denial rates to otherwise low-risk borrowers.
According to a survey by the National Association of Federally-Insured Credit Unions, over half of credit unions use alternative data in their underwriting process. In addition, the Consumer Financial Protection Bureau has issued guidance encouraging lenders to use alternative data sources to expand access to credit.
It is difficult to predict when unbiased and fair lending will be achieved in the U.S. and globally. Our viewpoint is that AI can absolutely bring equity into the lending market, potentially in the next 7-10 years. To do so also requires a discussion in highlighting the mistakes in earlier iterations of automated claim processors, e.g., how learned bias and selection bias were baked within earlier models and incidentally reinforced lending stratifications - all of which will be discussed in Part 2 along with some of the Fintechs on the cutting edge of AI and alternative credit profile scoring.