For Credit Unions The Time For Automation is Now
July 9, 2020
Credit unions that do a significant amount of manual underwriting in their retail banking operations are facing a difficult juggling act. Leaders at credit unions tell us they are frustrated when they can’t deliver shorter turn-around times for consumers seeking everyday products like mortgage refis, credit cards and personal loans. This is all despite a sharp decline in loan applications during the COVID-19 period. The CFPB recently reported consumer loan applications in most categories declined between 30-50%. COVID-19 has highlighted the need for greater automation, so credit unions make the best loans possible since there are so many fewer loans to be made.
Why haven’t credit unions increased automation already?
Credit unions take pride in serving their members. Often held in contrast to big banks that operate based on rigid rules, credit unions offer a more customer-centric approach that takes into account the specific and personal history of each member, especially when they apply for a credit card or loan. Credit unions recognized early that fully-automated underwriting decisions, determined by rigid rules based on a credit score and just a handful of attributes, would result in denying applications from too many members and for the wrong reasons.
For many, the gains from automation don’t outweigh denying those borrowers. Simple rules can’t take into account other important information, like members’ banking history, asset information, and responsible behavior. Credit unions have expert underwriters who know this additional information can be used to more accurately assess an applicant’s creditworthiness and often use it to make lending decisions. Manual underwriting enables credit unions to make lending decisions that consider additional objective details about a borrower’s banking history that credit scores miss.
“Instant decisions allow us to approve more members and focus our team on the personalized white-glove service our members have come to expect."
Unlike finance companies, credit unions have a data advantage and offer multiple banking products, including checking and savings accounts, credit cards, auto and RV loans, personal loans, and mortgages. Credit unions also serve a specific membership which allows them to tailor their products and policies to their membership base. A broad-market credit score doesn’t reflect all this data, nor does it reflect the specific character of a credit union’s membership.
Too much manual underwriting is not sustainable
Unfortunately, high-touch experiences come at a cost, as has been highlighted by COVID-19, where manual processing has stretched call centers and customer service agents to their limits. Internet and mobile banking experiences, pioneered by fintechs and large banks, have forever changed consumer expectations. Members expect to be able to make transactions and apply for loans from the convenience of their mobile phone and get an instant decision, not wait hours or days for a call back.
Learning from this difficult period, smart institutions are preparing for the inevitable recovery of consumer demand for loans, where greater automation will be required to ensure lenders can meet increased demand and make profitable loans without degrading the member experience or ballooning costs.
Machine learning can help credit unions deliver a better member experience while also increasing automation
What if it were possible to increase automation and provide instant credit decisions, while also safely increasing approval rates? It is, and that’s just what the innovative credit unions like VyStar have done. Working with Zest, VyStar became the first credit union to adopt a custom, AI-powered credit risk model. VyStar sought cost-savings associated with greater automation, and the opportunity to offer instant decisions, better pricing, and personalized service to its 675,000 members. As VyStar Chief Lending Officer Jenny Vipperman said in her Credit Union Journal op-ed, “Instant decisions allow us to approve more members and focus our team on the personalized white-glove service our members have come to expect."
Machine learning offers the ability to consider the specific attributes of each customer to make a more informed, fair and transparent credit decision, without requiring manual review.
With AI-based decisioning, says All In Credit Union SVP of Sales and Lending Todd Peeples, "we now have consistent decisions being made across all the branches. That's allowing us to approve more loans while also decreasing our risk.
Machine learning models incorporate the wisdom and expertise of your underwriters without wasting their time.
Machine learning models can take into consideration additional factors missed by credit scores and simple approve/deny rules. Data like checking account balance history, asset information, and tenure with the institution can be used when generating a risk score. ML models consider how distinct factors like total debt and total income can be combined into ratios to produce a more accurate prediction. Unlike traditional methods, machine learning can consider all combinations of inputs.
With machine learning, models consider when variables like income, inquiries, and credit utilization indicate higher or lower risk. Unlike traditional methods, machine learning models are able to handle situations where bigger isn’t always better. For instance: expert underwriters know that more income does not always indicate lower risk. Too high an income can indicate fraud. Modern machine learning methods can detect subtle patterns like these to provide more accurate predictions.
Just a few years ago, machine learning was only practical for the largest banks. Now it's accessible to financial institutions of all sizes, including regional banks and credit unions, looking to increase their automation.
Credit unions can increase automation safely with deliberate change process
More accurate predictions allow for greater automation because you can be more certain of the outcome. When a machine learning model says the loan will go bad, you can be more certain the model’s prediction is right. The same goes for good loans. Machine learning models better rank-order the risk of loan applications. When you can trust a risk score to be more accurate, and you know that it incorporates all of the attributes your expert underwriters would use, there is less need for manual review. Zest’s customers have used machine learning models to increase auto-decisioning to more than 95% of all loan applications.
Zest’s customers have used machine learning models to increase auto-decisioning to over 95% of all loan applications.
Lenders often implement a deliberate change-management process to gradually increase automation over time, based on early indicators of the performance of loan vintages originated using a machine learning risk score. By increasing the percentage of loans auto-approved and auto-denied at a deliberate pace, and carefully observing repayment behavior, lenders can achieve greater automation safely and responsibly.
The above diagram shows how a more accurate risk score can help automate the underwriting process. As a result, underwriters can focus on a smaller number of truly marginal applications. When lenders observe the ML-powered risk score is more accurate than traditional methods, they can adjust auto-approve and auto-deny thresholds so that fewer loans get routed to underwriters for manual review. That means more members experience an instant decision, and because the ML model uses more data about each applicant, more good borrowers can be automatically identified and approved. This results in a better member experience.
Recent times have stretched credit union operations to the max. Increased automation, enabled by explainable and transparent machine learning credit risk models, can help credit unions safely and responsibly increase auto-decisioning, allowing staff to focus their attention where it is most needed. Says All In's Peeples, "Automation allows your staff more free time to have some additional conversations with a member to be able to provide additional services, cross sell new opportunities, see what else may be available there and for us to be able to help the member."
Increasing automation now will lead to stronger futures
Credit unions that increase automation can originate more good loans without requiring manual review. They will be able to handle the larger and more profitable loan books likely to occur once the recovery gets fully underway. Increased automation also helps credit unions offer a better experience to members, increasing longer-term profits without making more demands on already overworked staff or significantly increasing costs. With greater automation, underwriters and loan officers can spend more time on the customers that really warrant their time and attention. More good loans can be automatically approved and more bad loans automatically denied. Recent times have stretched credit union operations to the max. Increased automation, enabled by explainable and transparent machine learning credit risk models, can help credit unions safely and responsibly increase auto-decisioning, allowing staff to focus their attention where it is most needed.
Zest AI team
June 2, 2021