A new regulation meant to prod banks to lend more to minority and women-owned small businesses will require some to completely rethink their entire small-business lending process, including the underlying technology.
Section 1071 of the Dodd-Frank Act is a fair lending law. Regulators are asking bankers to prove through reported numbers that they are lending to enough minority, women and LGBTQ-owned small businesses.
On the face of it, complying with the Consumer Financial Protection Bureau's 1071 rule, which requires small-business lenders to report demographic details about their small-business borrowers while keeping that information hidden from human or automated lending engines, is simple. It involves adding a few questions to the small-business loan application and putting the answers behind a firewall or partition. It's similar to the effort required for mortgage lenders to comply with the Home Mortgage Disclosure Act of 2011.
But in reality, meeting the 1071 rule requirements is "a complex effort for banks involving multiple data collection platforms," said a spokesman at Citizens Bank in Providence, Rhode Island. It will require changes to certain existing systems, he said, including customer onboarding applications and regulatory reporting. The bank is in the process of making these modifications.
At some banks, it will take more than changes to existing systems.
Traditionally, small-business loan officers have had autonomy and discretion to lend to whichever companies they choose. Some don't use lending software, and instead use paper loan applications and Excel sheets. The 1071 rule will force many banks to standardize their processes and underwriting criteria to make sure they lend to women- and minority-owned businesses at the same rate as businesses owned by white men. This means those that don't use lending software today will need to buy or build some.
Congress tried to challenge the new rule. The Senate and House passed a resolution that would put an end to 1071. This was supported by several bank and credit union associations that said 1071 is too far-reaching and would harm the relationship banking model banks have with businesses.
Late Tuesday night,
However, all compliance deadlines are temporarily stayed due to a preliminary injunction issued by a Texas district court in July. The court is hearing a lawsuit challenging the validity of the CFPB's 1071 rule. The court has granted preliminary relief to the plaintiffs — the Texas Bankers Association, the American Bankers Association and Rio Bank, McAllen, Texas — and their members.
Which banks will find this hardest
The level of difficulty the 1071 rule presents to a bank will depend partly on its size and the technology it uses for small-business lending today.
Large banks with modern small-business loan origination software and standard processes can simply add additional fields to the loan application and make sure the data entered into those fields is shielded from loan officers.
But for community banks that still have human loan officers making judgment calls and that use paper-based loan applications or older software, it may be tough. Not only do they have to prepare for data collection, they have to fundamentally alter the entire way they do business lending.
It's the banks that don't use software for small-business lending that will feel this the most, agreed Patrick Reily, co-founder of Uplinq.
"A large portion of small-business lending today is still done a lot like commercial lending," Reily said in an interview. "There are these manual processes that happen in branches, where people are having to fill out forms." Several small-business lending software vendors, including Uplinq, Amount and Baker Hill, have built 1071 data collection into their programs, or are doing so.
Banks considering new small-business lending software will need to decide if they want to use artificial intelligence. AI-based lending software providers say artificial intelligence will improve small-business lending and help improve fairness. For instance, AI-based models can take into account so-called alternative data such as a business's cash flow.
At TCF National Bank in Detroit and First National Bank of Omaha, early results from artificial intelligence pilot programs are strong.
Conventional wisdom has long held that AI models, such as machine learning models and large language models, perpetuate bias. They are often trained on incomplete datasets, which can lead to gaps in the understanding of borrowers. For instance, to a credit decision model trained on data in which women are underrepresented, women who left the workforce for several years to raise children look like deadbeats. And if models are trained on the outcomes of past decisions that were biased, they will look for the same patterns.
But some say it doesn't have to be this way.
"Models can be tasked to optimize around multiple, simultaneous objectives like shareholder value, fairness and economic development," Reily said. "Further, they can be regularly tested to assure satisfaction of each of these goals."
For banks that have very old core systems, collecting 1071 data could prove challenging.
"Without a modern origination platform, the ability to aggregate and then pass over that data becomes very difficult," said Jonathan Katz, vice president of Amount. "It becomes a compliance nightmare, especially if you're a really small bank and you now have to report this for any business under $5 million in revenue."
The fast path for banks stuck with legacy systems "is to put something in place that's doing the data capture that's modern and then figuring out how to squeeze it into the legacy," Reily said.
The relationship pricing many banks do may have to change. Sometimes loan rates are affected by a borrower's ability to, say, bring in $100,000 in deposits or other lines of credit at the bank.
A fair lending law
From the CFPB's perspective, small-business lending should be held to the same standard of fairness under the Equal Credit Opportunity Act as consumer lending. In that sense, the 1071 rule is not just about collecting more data.
"It is the opportunity to adjudicate fair lending," Reily said.
Fair lending seems like it should be a given, Reily said, but when human judgment is being relied on for loan decisions, "it's very difficult to avoid the reality if somebody says, well, you grew up in the town that I grew up in, you went to the college that I went to, we have some kind of a bond at some kind of a level, that connectivity creates bias," he said. This is not done out of evil intent, he noted, but it does create subtle biases.
To see how their small-business loan portfolios will look to regulators, banks can conduct Bayesian Improved Surname and Geocoding (BISG) testing, a methodology developed by the RAND Corporation that can help organizations produce estimates of racial and ethnic disparities within datasets and find areas for improvement.
"I would 100% recommend any institution do [BISG testing] because the likelihood that they're going to find a problem is high," Reily said.
One of the challenges of fairness in small-business lending is that many banks rely heavily on the FICO score of the small-business owner in credit decisions.
The CFPB will be looking for "good-faith efforts" to implement 1071, said Alan Ellison, small-business lending program manager at the CFPB, at American Banker's Small Business Banking conference in November.
"That means integration in your compliance management systems at your financial institution in terms of policies and procedures, and validating the data is correct," he said. "We're also going to be looking to see that you have effective programs for training your bankers to be able to deal with the demographic data collected."
