Jin Han is the Chief Legal Officer of Pipe Technologies. The company, headquartered in San Francisco, California, is a fintech firm providing a capital platform for small and medium-sized enterprises.

Despite decades of efforts by industry and regulators to improve fair access to business financing, significant gaps persist, especially for businesses owned by women, minorities, and immigrants.

These gaps largely stem from structural biases in current risk rating systems, compounded by additional biases introduced in manual and algorithmic underwriting processes, whether AI-based or not. Increasing access to financing for underserved groups requires more than legal compliance; it also requires addressing the structural disadvantages these groups face.

Many arguments that AI and machine learning can make underwriting processes unbiased border on wishful thinking. No algorithmic process is immune to biases inherent in the data it analyzes, nor can it replace careful consideration at the underwriting policy level.

We urgently need more discussions on how to systematically achieve financial inclusion. Advances in AI and machine learning are often touted as a panacea for unequal access to financing. What we need is a plan that clarifies how these technologies can genuinely help, rather than take over, the work of increasing financial accessibility.

AI technology is an excellent tool for accurately ingesting and analyzing data, as well as for automating the implementation of fair and well-designed underwriting policies. We believe that combining the efficiency gains from recent technological advances with human careful selection of risk data that benefits underserved groups is the winning strategy. Here is our thinking on this issue:

Limitations of Traditional Underwriting Methods

Traditional risk assessment models, relying primarily on FICO or Vantage scores, offer clear benefits to capital providers. Positive repayment and credit history demonstrate a business owner's responsibility, management capability, and willingness to meet obligations. Relying on credit bureau data also tends to shorten underwriter review times and makes the underwriting process easier to automate. A good credit score is a reliable indicator of business responsibility, serving as a substitute for time-consuming research and interviews.

Despite these benefits, traditional underwriting models have long been criticized for perpetuating unequal access to financing opportunities. According to data from the Federal Reserve's Small Business Credit Survey, as of 2023, businesses owned by women and minorities are still disproportionately denied financing or are more likely to receive approved amounts lower than requested.

The reasons for this disparity are straightforward. Women, minorities, and recent immigrants are often newer entrepreneurs, typically operating brick-and-mortar retail businesses with thinner profit margins. Some new immigrants may be cautious about debt and place less emphasis on building repayment or credit histories.

These historical and cultural factors make credit-score-based underwriting unsuitable for increasing financial accessibility. While considering a business's operating history for risk underwriting appears value-neutral on the surface, its impact is more pronounced on newer business owners.

Addressing a system with clear predictive power and no obvious discriminatory intent is not easy. But unless the industry finds ways to reduce reliance on credit scores and operating history, it simply cannot solve the issue of financing accessibility.

This is a huge challenge. But magical thinking around AI is not the answer. There is a different and more systematic approach to tackling this issue. We have seen it work at Pipe, and I believe it will continue to increase financial accessibility.

A Different Approach to Increasing Financial Accessibility

Given the clear benefits of traditional factors, completely removing them from underwriting is neither wise nor practical for most fintech companies or banks. Even if underwriters consider "alternative" data, such as rent, utility payments, public records, bank data, and educational background, it is uncertain whether additional datasets can bring greater financial inclusion. Some alternative datasets inherently require long-term measurement, while datasets around education and social circles may adversely affect certain traditionally underrepresented groups.

Pipe's approach relies primarily on value-neutral alternative data, such as revenue, expenses, and cash flow trends for underwriting, while progressively shortening the length of business history required to obtain financing. Rather than pinning hopes on breakthrough AI underwriting processes, Pipe uses AI and machine learning technologies to ensure data integrity and reliability, as well as the predictability of its underwriting models.

After gaining confidence in its models, Pipe has also been able to avoid using credit scores as a primary underwriting factor. In fact, today 100% of our applicants are reviewed through revenue-based underwriting models. We only consider incorporating credit scores as a factor when doing so helps us fund more people overall, rather than fewer.

Pipe has also invested heavily in tools to automate the implementation of its credit policies, with nearly 100% of its risk decisions now applied through automated processes. Manual review is typically reserved for cases where automated compliance, identity verification, or fraud prevention systems trigger the need for applicants to provide additional information.

With these measures in place, Pipe has been able to provide meaningful financing to businesses with as little as three months of operating history. We have also built industry-specific risk models capable of underwriting merchants who might present higher-risk characteristics under traditional rating systems. This achievement has enabled Pipe to successfully meet the capital needs of small businesses with high concentrations of female and minority owners, including in the beauty, services, retail, and food and beverage industries.

Results Require Meticulous Work

The historical excitement over AI and other algorithmic technologies should not cause financial services companies to overlook the opportunities for achieving greater financial inclusion, which can only be realized through tedious, meticulous human thought. Given the current state of technology and data availability, AI and other algorithmic technologies are best suited to bringing efficiency to well-designed underwriting, compliance, and fraud prevention processes. In short, AI is used to process the results of critical human thinking.

Human underwriters must focus on systematically screening risk factors that have predictive power and benefit underserved groups. When technology eventually catches up, we are confident that the companies that have thought most deeply about achieving financial inclusion under current constraints will be the ones that most successfully leverage the technology.