First Financial created its own AI/machine learning short-term loan system in three steps that involved extensive cross-enterprise commitment
Application abandonment has dropped along with the number of apps that have to go to manual underwriting.
The system will be expanded to other loan products soon.
First Financial of Maryland FCU ($1.3B, Sparks, MD) is nearly two years into a real-life experiment in using its own internal smarts to create and fine-tune a short-term loan product that relies on artificial intelligence and machine learning.
CU QUICK FACTS
FIRST FINANCIAL OF MARYLAND FCU
DATA AS OF 06.30.22
HQ: Sparks, MD
12-MO SHARE GROWTH: 5.73%
12-MO LOAN GROWTH: 11.45%
As currently structured, the credit union’s AnyTime Express Loans can be for $1,000, $1,500, or $2,000, with a fixed rate of 5.99% for a 12-month note. Loans are applied for through online banking or the First Financial mobile app and are funded instantly upon approval.
AnyTime Express is an end-to-end loan application, underwriting, and fulfillment product that’s part of the credit union’s lineup of unsecured Anytime Loans. Members complete a mostly pre-filled application either online or through the credit union’s app.
Then, says Michael Powers, the Maryland cooperative’s chief innovation and strategy officer, “The AI/ML decision engine decides to approve the application and fund the loan, or pass it off for a second look to one of our human underwriters. Approved members are presented with loan documents, and upon digital signing the loan records are created on the core and the members have their funds available immediately.”
While the product provides an alternative to payday loans and other short-term lenders, the instantaneous nature of the process appeals to any First Financial member interested in quickly getting access to the money they’ve applied to borrow.
“Our vision for AnyTime Express is to deliver a low-dollar consumer loan that could be started at the time a member gets out of their car at a retail location and be finished with proceeds in a checking account in less time than it takes to get to the front door of the store,” Powers says.
Our vision for AnyTime Express is to deliver a low-dollar consumer loan that could be started at the time a member gets out of their car at a retail location and be finished with proceeds in a checking account in less time than it takes to get to the front door of the store.
Seizing The Opportunity
Here’s how Powers describes the opportunity his credit union seized: “Fundamentally we saw that while we have an effective rules-based automatic approval engine in our LOS, many applications and especially subprime credit applications were not approved automatically but were very frequently approved in manual review. We saw the opportunity for machine learning to replicate this behavior and deliver more instant loan decisions and instant funding across the credit spectrum, from prime credit to subprime credit borrowers.”
Since its deployment on Oct. 15, 2020, the system has processed about 3,000 applications, with about 40% of them received outside of business hours.
Powers says the approval rate moves between 60% and 80% because of fluctuations in the credit quality of the applicants with those who don’t make that cut sent to manual review. He says the abandonment rate for instantly approved applications is only about 3%, a fourth that of the abandonment rate for manually approved applications.
A Lift In Sales And Member Satisfaction
Powers says management wasn’t necessarily seeking a meaningful reduction of approved application abandonment, but the difference is striking and persistent, he says, calling it the clearest quantitative example of member satisfaction around getting instant answers and instant funding.
“We all know the time value of money; our instant-approval results show the time value of timing,” Powers says. “Delaying a response for manual review by even a short interval can lose someone’s attention and discount the value of your offer to zero.”
They’ve also seen a sales lift and internal efficiencies boost from the work, Powers says.
Before AnyTime Express we had seasonal personal loan promotions in the summer and winter. Participation in those pre-approved programs had been waning for years, he says. “We replaced those pre-approvals with an invitation to apply for an AnyTime Express loan at a rate discount, and volumes have recovered to healthy levels last seen years before. Additionally, AnyTime Express is lower cost and requires less processing labor than the previous pre-approval program.”
We all know the time value of money; our instant-approval results show the time value of timing. Delaying a response for manual review by even a short interval can lose someone’s attention and discount the value of your offer to zero.
How It Was Created
Creating the AnyTime Express project was a three-stage process that began with a prototype of an AI/ML underwriting decision engine run in a controlled test environment. From there, the elements were built for real-time integration into the cooperative’s core and digital banking platforms.
Then, Powers says, employees submitted loan applications in a real-time, controlled setting. “We worked out a number of issues with this beta test and concluded that we were prepared to release an improved version to production,” he says.
The final result: AnyTime Express is fully integrated with one part in our core banking system, one part in our digital banking system, and a third part, the decision engine, running as a web service on-premises. The three parts communicate with each other and their respective back-ends to make it work, Powers says.
Further Refinements And Rollouts
Only, it’s not final. AI/ML technology is all about continuous improvement, and First Financial’s system is no exception.
“There has been additional data collected on approvals, manual reviews, and charge-offs that we can incorporate into re-training the AI/ML system,” Powers says. “We found that some notable improvements came from adding rules to the engine to handle some cases.”
Those refinements will help further reduce the number of applications that can meet credit union approval without being sent for time-consuming manual review, and better prepare the system to handle the differences as the instant fulfillment process is expanded to other loans at the credit union.
Teamwork Makes This Dream Work
Powers himself brings a unique skill set to the job. He has a Ph.D. in electrical engineering from the University of Maryland and joined the credit union’s staff in 2017 after 11 years as a member of its board of directors while he held robotics and imaging research posts with General Dynamics and then the U.S. Army.
But Powers is quick to point out that enterprise-wide involvement was critical to the instant approval system’s creation and continued evolution.
His Innovation Strategy team built the core AI/ML decision engine, the IT department constructed the core extensions, online banking widgets, and server resources, and the lending department configured the loan product while determining credit risk and underwriting performance parameters.
“We also had consistent support from our compliance department to ensure fair lending standards were observed from the start and followed throughout. Marketing, meanwhile, created the graphics that gave it a visual appeal consistent with our branding and banking app appearance,” he says.
Management Buy-In, Hiring Insight, Doing Something Original And Significant
Powers adds that management buy-in was crucial across every department and functional area, including, he says, those closely involved in the project and others not involved but interested in the outcome.
“Our step-by-step approach to development let everyone see and build confidence in the results of each increment as it built toward an ever-more practical system in an increasingly realistic environment,” Powers says.
He gave a particular shout-out to the lending team. “I give them a lot of credit. This was a very big step into the unknown and they met it with impressive optimism, courage, and care,” Powers says.
And altogether, he says, “Despite the technical demands and considerable unknowns of the work, the team came together with strong collaboration and common purpose.”
And, in true AI/machine learning fashion, the learnings here include how to hire for an enterprise that wants to innovate like this now and going forward. First Financial looks at a variety of characteristics when it considers at a candidate, and Powers says his favorite is tolerates ambiguity.
“The people who will be the most productive and the most valuable to a technology development in your organization will be those who can make progress toward the overarching intent of the work despite and sometimes because of limited information, conflicting priorities, unclear feasibility, and undefined methods,” he says. “These conditions are always present to some degree when something original and significant is to be done.”