Role of Artificial Intelligence (AI) in Auto Lending

Authors | Subbu Venkataramanan, Madhuri Prabhakar

The credit underwriting process has been significantly improved with the use of Machine Learning and Artificial Intelligence tools. Lenders are now able to combine internal data such as application, relationship, customer interaction, digital footprint, with alternate data sources such as LexisNexis, Clarity etc. along with tradeline level Credit Bureau data along with AI (Neural Nets) to make more accurate predictions about whether an applicant would turn delinquent and thus improve the approval process.

This is particularly useful in case of Auto Loan Underwriting, where alternate sources like the Kelley Blue Book, as well as key features of the car itself can be leveraged to make the risk model more predictive. For instance, the primary purpose of a car, whether for leisure use or for jobs such as construction or farm work can provide insight into the financial standing of the customer as well as the potential depreciation of the car.

The key characteristics, including car make and model, year of manufacture, used/new, number of previous owners, miles driven and primary purpose of the car, can help set the Loan to Value (LTV) risk cuts. They are also instrumental in calculating the difference between estimated value and actual sale price.

A new loan approval process

The turnaround time in decisioning an auto loan is critical, since the customer has many choices and the dealership wants to sell the car as soon as possible. Customers should ideally be able to walk in to a car dealership with a pre-approved loan. The way that this process would work is as illustrated below.

The customer logs onto their bank’s auto loan website and fills in details, including expected car details and loan amount. This input is then combined with the customer’s background information and repayment history from the bank’s internal database and additional information from an external source like Kelley Blue Book. Using explainable Artificial Intelligence (Neural Net) risk model to provide a decision about approval of the loan, as well as the approximate amount that can be disbursed. All this can be done in a matter of minutes, from the comfort of the customer’s home.

The risk models should be capable of providing scores based on partial information and if that score is above a certain threshold, approval decisions can be done instantaneously without waiting for features from all data sources.

Armed with this information, the customer can approach the dealership and negotiate a deal for the car. Once this has been finalized, they can reach out to the bank, either online or in person to complete the formalities and finalize the terms and conditions of the loan. This would greatly simplify the process and make it more efficient.

The Dreaded Depreciation

All participants in the auto loan market, whether borrower or lender, buyer or seller, are well aware of the term ‘depreciation’. As the saying goes, a car begins to lose its value the second it leaves the showroom. The average rate of depreciation is about 20% of the original value within the first year, after which it stabilizes and plateaus at about 15% every consecutive year. Of course, this is a ballpark figure, and the actual value depends on factors like mileage, model and region.

This means that typically, in a five-year loan for a new car, the loan outstanding tends to fall below the car value around the three-year mark. The exact timing depends on the car’s depreciation curve.

From a lender’s perspective, beyond this point, the risk is much lower due to two reasons:

• The customer would have been re-paying the loan for almost three years by then, demonstrating credit worthiness and
• In case of non-payment, the repossessed car would have good value compared to loan outstanding.

For used cars, the timing can be anywhere from two years on, depending on the car’s depreciation curve.

Kelley Blue Book provides depreciation curves for make, model and vintage. This helps in estimating the loss i.e. the value that the lender can recoup in case of repossession vs. the amount of loan still outstanding. These depreciation curves can be made more accurate by leveraging Artificial Intelligence (AI) techniques, using actual resale data on the loans in the bank’s book, region (heavy snow areas vs areas with more temperate weather) and purpose of the car. Models can be built to assist with customer retention and induce repeat loans based on depreciation curves and analytics.

‘Upside Down’ in a vehicle

In 2017, around 33% of the customers who traded in their cars for new ones at dealerships were ‘upside down’ on their loans. This means that they owed more on the loan than what the car was worth, due to depreciation, long term of repayment and high APRs.

Advanced analytics can be employed to identify customers who could potentially turn ‘upside down’. At that point, the dealer could suggest multiple remedies to the situation including exchange offers for a new car purchase or a new loan. This could be beneficial for both the lender, from a risk perspective and for the borrower, to stay above water.

The Road Ahead

Since the great recession of 2009, the industry has seen a shift in approach to auto lending by banks. They have loosened their standards, lending more to sub-prime borrowers than earlier. This has resulted in a boom in auto lending, at a near record pace, with almost $1.1 Trillion in outstanding loans by the end of 2017.

However, lenders might have let borrowers take on more debt than they can repay, leading to a burgeoning number of delinquencies. Economists fear that if the economy dips into a recession, there would be a very large number of people on the verge of losing their cars to repossession, due to default on their loans. Additionally, driverless cars look to be ushering in a new era in car ownership and leasing.

All factors considered, Artificial Intelligence based nuanced risk models are paving the way to the future in auto loan underwriting.