AI for Auto Lending – Improving Deal Flow and Risk Reduction with Data

Dylan Azulay

Dylan is Senior Analyst of Financial Services at Emerj, conducting research on AI use-cases across banking, insurance, and wealth management.

AI for Auto Lending - Improving Deal Flow and Risk Reduction with Data

Ten years into the longest economic expansion on record, auto lenders are looking for ways to leverage new opportunities for growth and risk reduction.

Machine learning is poised to transform the auto lending industry in particular as much as FICO and other traditional credit scores did before it. Auto lenders that are able to adopt artificial intelligence might be able to win more customers and take on less risk at the same time.

According to our report, banks see a real fit for machine learning when it comes to optimizing financing and loan processes, such as credit scoring, risk management, and compliance reporting.

This is evidenced by the amount of funding financing and loan products have generated: $438 million; financing and loans as a category ranked third for highest amount raised among banking functions. Funding amounts are a good proxy for AI traction in a space because banks and VC firms are unlikely to fund AI startups unless they’re sure their products will generate an ROI. 

In addition, 40% of the companies that offer AI products for financing and loans to banks scored above a 3 (out of 4) on our Evidence of ROI Score, indicating a relatively strong vendor ecosystem and existing traction. This indicates that there are several AI vendors in the lending space that are already generating returns for banks and financial services firms. 

What this means is that machine learning is already automating key lending processes in banking and finance, and auto lenders stand to benefit.

That said, the automation discussion in finance is in large part focused on robotic process automation (RPA). Although RPA might be able to automate credit scoring at scale, it in most cases only serves to improve efficiencies because it streamlines existing processes.

Machine learning, on the other hand, may require a more intensive integration, but it can improve deal flow and reduce risk by allowing auto lenders to approve more ROI-positive loan applicants and deny riskier applicants.

To learn more about where machine learning is impacting the auto lending industry, we spoke to Jay Budzik, CTO at Zest AI, about how it can drive value for auto lenders. Interested readers may want to read how Zest AI helped Prestige increase loan approvals by 14% using AI, read Zest’s case study.

In this article, we discuss how auto lenders can leverage new data sources to approve more applicants and take on less risk.

New Data Sources for Auto Lending Decisions

Auto lenders are accustomed to using logistic regression models that factor in one to two dozen variables when determining whether or not to provide an applicant a loan. In general, they take an applicant’s credit history and current usage, income, employer, downpayment, and other financial and credit indicators into consideration when deciding whether to approve an application or not.

Machine learning algorithms, however, can process hundreds or thousands of variables and look at a near infinite number of relationships among those variables to assess borrower risk.

There are a variety of data sources that might inform a machine learning-based lending model, many of which are overlooked by most traditional models.

AI and machine learning models, such as those built with Zest AI’s software, allow lenders to draw accurate credit risk insights based on the interactions among hundreds or thousands of new data variables previously unavailable to traditional lending models, such as application data or CRM data a lender already has on that customer. 

Widening the data set, for example, allows a lender to factor in not just whether an applicant had a bankruptcy but when the bankruptcy occurred, what type of bankruptcy it was, and how severe.

A machine learning model can also factor in the applicant’s customer call center history, history at a residence, or whether the applicant has any open court cases that, depending on how they resolve, may impact the financial standing of the applicant.

Auto lenders can exploit this vastly expanded menu of credit variables by using factors that are specific to the car the applicant is applying for a loan to buy. Data such as the make and model of a car are valuable to consider. According to Budzik:

We work with auto lenders of all sizes – and sometimes we’re able to look at the models they have in place before using machine learning. We’ll often see 10 to 15 variables used in the model to make a decision about whether or not to lend to a person – including: Percentage down on the car, the loan amount, etc… and then make thier decision around rules based on those factors.

Machine learning allows lenders to consider hundreds or even thousands of variables, giving them a more accurate view of that consumer… This gives lenders an advantage to approve folks who they might have overlooked otherwise.

Auto companies with enough digitized historical transaction data may be able to feed that data into an algorithm and gain insight into who their specific dealerships should approve for loans for specific cars. This was possible with previous statistical methods but is taken to a greater degree of granularity with new data sources and machine learning. 

For example, it may be that across the US on average, people who buy BMWs are less likely to pay back their loans than people who buy Toyotas. However, a specific  BMW dealership may use this data and algorithm to find their customer base is above average for loan repayment. 

As a result of the fine-tuning that machine learning allows, the dealership could approve more loans than they otherwise would using national data from dealerships and lenders across the country. In addition, they can take on less risky applicants that may have been approved using traditional credit scores.

Using AI to Win Customers Without Credit

New data sources can be used to offer loans to customers many lenders may not feel comfortable lending to at present because either they don’t have adequate credit histories or because they have a blemish on their credit record. According to Budzik:

This is particularly true for young people these days who have taken out a debit card, don’t have credit cards, do most of their transacting and using their checking account online, and have a very limited credit history. It’s also the case for underrepresented segments of the population that might not have the means or setup to get that history. 

Credit scores are often the best objective indicator that a customer will pay back their loan, and lenders use them as a way both to decide accurately on whether to lend to the customer and to remain in compliance with the FDIC’s fair lending laws.

That said, many people either don’t have credit scores (26 million, according to the US Consumer Financial Protection Bureau, or one in ten adults) or their credit scores aren’t an accurate reflection of their borrowing history. Younger borrowers are used to buying with their debit cards and checking accounts online, and many don’t have credit cards.

In addition, young people in particular have difficulty acquiring loans due to a lack of credit history or poor credit. According to TransUnion, 43% of millennials have poor credit. Their lack of credit history in part is due to their age, but many millennials also report not knowing how to build their credit. As a result, younger loan applicants often lack the data that traditional credit models use to determine whether or not a lender should give them a loan.

Lower-income segments of the population face especially acute issues due to thin or mixed credit histories. As they are the most likely to need a car to get to work, auto lenders that can serve them first, before their competitors, may be poised to win customers that will stick with them throughout the course of their lives.

Traditional lending models may make it difficult for both groups to get approved for loans because they factor in fewer variables and fewer indicators of their creditworthiness. AI and machine learning could make this easier for auto lenders, allowing them to offer auto loans to underserved markets or to decent consumers who would have been passed up by previous models. According to Budzik:

If you have had a credit issue in the past or don’t have a spotless record, you can get denied. The folks who might have a blemish on their record – those are the folks for which having access to a car can be most meaningful – allowing them to go to work and provide for their family.

In addition, AI and machine learning could reveal certain customers with blemishes on their credit histories may not be as much of a risk as traditional lending models assume they are. As such, auto lenders could approve more people for loans (and therefore make more money) using machine learning. On the other hand, machine learning could reveal that an applicant that would have just barely been approved using a traditional credit score may be riskier than their credit score might indicate; auto lenders could then deny their loan, mitigating their risk.

Zest AI claims to have helped Prestige Financial Services increase the number of applicants it was approving for loans using machine learning software. The company’s software purportedly used 2,700 variables to determine if Prestige should lend to applicants or not. As a result, Zest AI claims Prestige was able to approve 14% more applicants without taking on additional risk.

Prescriptive Analytics for Setting Interest Rates

Machine learning may also allow auto lenders to set interest rates on an individual basis. Similar to how some prescriptive analytics models recommend claim payouts for insurance adjusters, a prescriptive analytics model could recommend to an underwriter at an auto lending company how much interest an applicant should pay on their loan. 

While data volume is an advantage in applying machine learning, lending algorithms don’t need enterprise-size data volumes to create more accurate, targeted predictions, and Zest AI claims that smaller lenders who have been in business for years often have enough data to apply ML. This trend of using smaller data volumes will likely only continue as algorithms and ML approaches improve in the years ahead.

Compliance Challenges to Adopting AI

The “Black Box” of Machine Learning

Traditional models allow underwriters to explain to loan applicants how they arrived at the decision of whether to approve an applicant for a loan or not. This is more difficult with machine learning. 

Machine learning has what is referred to colloquially as a “black box” problem, meaning it’s sometimes impossible to tell how an algorithm comes to the conclusion it does. This is one of the reasons it’s a challenge to adopt AI in finance; not being able to explain how a model rejected someone’s application for an auto loan could land a lender in serious violation of laws and regulations. Zest AI claims its software builds models with explainability and transparency, giving auto lenders a correct and consistent answer for why an applicant was approved or denied. 

Avoiding Discriminatory Data

A lot of auto lenders enter into the decision to use ML with concerns about regulation. There are a number of factors auto lenders can’t legally build into their traditional lending models, and as such, they can’t build them into their machine learning models either. These factors include race, gender, and age, and potentially other personal markers that may be added to the list in the coming years as more non-discrimination laws take effect across the United States and Europe. 

Complying with fair lending laws when using machine learning models can be a bit trickier because the models are sifting through an infinite number of interrelationships within the data and can connect two seemingly innocuous variables together to make a discriminatory factor. One of these variables is location or where the applicant lives. According to Budzik:

The location of the applicant can be a pretty good predictor of their race and ethnicity, and so you have to be pretty careful when you’re considering attributes like the location of someone. We’ve seen even at the state level when you associate the state in which a person lives with other attributes like the mileage on the car, you can end up with pretty perfect predictors of their race. So it’s important to be able to do a full inspection of the model to make sure it’s not doing the wrong thing from a discrimination perspective, that it isn’t making decisions that are biased and unfair.

The Bottom Line for Auto Lenders

The bottom line is that machine learning could allow auto lenders to price competitively, offering more tailored loans and interest rates to a larger customer base without added risk. This is the goal, at least. Although adoption is slow, financial institutions are already investing in AI for lending. 

Machine learning may prove necessary for auto lenders to stay competitive in the coming years. So long as vendor solutions allow for transparency and interpretability (ensuring compliance and avoiding unwanted bias), adoption in lending will only increase as more data sources become available.


This article was sponsored by Zest AI and was written, edited and published in alignment with our transparent Emerj sponsored content guidelines. Learn more about reaching our AI-focused executive audience on our Emerj advertising page.

Header image credit:

Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the 'AI Advantage' newsletter: