Artificial intelligence (AI) is woven throughout everyone’s daily life. It is so prevalent that most people don’t think about AI as our navigation systems propose new routes based on real-time traffic, offer grammar and spelling suggestions as we write or create hilarious—and sometimes embarrassing—texts with autocorrect.
As technology has evolved, the opportunities for AI adoption to increase efficiency and productivity in all industries have grown, including the financial industry.
“One of the most obvious benefits of AI is the ability to automate underwriting to support credit decisions,” said NAFCU’s Director of Regulatory Compliance Nick St. John, NCCO, NCBSO. Predicting borrowers’ likelihood to repay the loan is a critical component of underwriting. “While credit scores based on reports from the different credit bureaus have been used to quantify that likelihood, AI-based underwriting programs claim to be better able to pinpoint, or more accurately predict, borrower repayment behavior,” he said. “AI can expand the underwriting process to include a number of data sources, far more than a human can handle in a timely manner, which could make the process more efficient and less risky.”
Streamlining the underwriting process to speed up decisions and provide more automatic approvals without increasing risk was the goal of Donn Mende, loan manager at HFS Federal Credit Union in Hilo, Hawaii, when his organization added AI assistance to their underwriting processes. “AI has been a buzzword for a while, but we spoke to other credit unions about their experience with it in their loan origination systems, and everyone said it was a good tool,” said Mende. “We use AI for personal, credit card and vehicle loans.”
Mende and his organization worked on the evaluation and testing of the program for a year before implementation. “During the testing phase, we ran the program with information from loans we previously made, then compared the results of the AI program’s decisions against the decisions made by our employees,” he explained. The results of these tests were used to tweak the program to reflect needs and requirements that are unique to HFS Federal.
HFS Federal’s goal is to increase automated approvals by 25%, to provide a quicker response to members and to allow loan officers to focus on more complex decisions, said Mende. “This is a cultural shift for our loan officers, so we had to reassure them that they did not need to go back and doublecheck every automated decision.” The combination of testing results before implementation and experience will lead to more confidence in the system, he added. “We have always conducted internal audits on lending decisions, and we’ll closely monitor the decisions as well as delinquencies related to the automated approvals.”
Although HFS Federal, like many credit unions, opted to use a third-party partner with expertise and experience in credit union lending for the software, the credit union is still responsible for regulatory compliance, said St. John. “Regulators are consistent in their guidance for all AI products. They want human beings involved in the process,” he said. For example, Regulation B requires an adverse action notice that ensures there is no discrimination in the determination. “If an AI program is approving or denying loans, regulators want an employee to review all denials and provide the applicant with the specific reasons for the denial. Simply saying that the algorithm denied the application is not enough.”
Another caution when introducing AI in the loan origination process is the potential presence of flawed data and a lack of understanding as to how the AI model can result in unintentional discrimination. A process that not only reviews denials, but also randomly audits decisions can identify potential changes needed to avoid discriminatory decisions.
Other AI Uses
“AI can also be helpful in generating fraud alerts based on algorithms that identify an individual’s pattern of behavior,” said St. John. The alerts may help credit unions ensure compliance with Bank Secrecy Act requirements and generation of Suspicious Activity Reports.
Chatbots or automated messages on websites and in credit union apps are a convenience for members who want to ask questions outside traditional business hours or during the day when they are unable to make a phone call to customer service. While the service can help better serve members by providing timely answers to questions, there are compliance issues that can arise.
Although some chatbots are limited to a set menu of questions or categories of information, others can be more complex and mimic conversation. These more “free-flowing” interactions between a member and chatbot can create potential for compliance risks.
“If a member using a chatbot notifies the credit union about a transaction error with a debit card or a service such as Zelle, Regulation E requires that the credit union begin an investigation within 10 days of the notice of error, even if it is in a chatbot,” said St. John. To ensure compliance, organizations must set up a monitoring program for the chatbot that captures specific words or phrases such as “unknown transaction” or “error” and sends the information to a person who can review the conversation and take appropriate action. The other issue that the Consumer Financial Protection Bureau has reported are complaints from consumers about chatbot “doom loops” that send consumers in circles without resolving a problem, answering a question or offering to connect them to a human. “It is always a good idea when using technology to communicate with members to offer them the ability to call a human being,” said St. John. Not only does the pathway to a person help the credit union avoid creating a negative image of its service to members, but it reinforces the industry’s commitment to “people helping people.”
