There are several key elements to any successful lending transaction. Most companies in the lending ecosystem specialize in supplying or doing one or more of these key elements well.
- Capital: the most obvious element, a lender must be able to source risk capital at attractive interest rates
- Enforcement: there must be a way to enforce the terms of the lending agreement in compliance with business regulations
- Servicing: efficiency in monitoring and ease of use in transacting the terms of the lending agreement such as regular payments within an ever-evolving payments infrastructure is a key capability
- Valuing collateral: for collateralized loans tracking and valuing collateral is a key element
- Fraud management: the ability to guard against fraud is critical especially in online transactions
- Creditworthiness / Underwriting: this is a core capability needed to assess whether the loan itself will be worth making
- Upselling / personalization: customers are expensive to acquire
AI can have its biggest impact in areas related to fraud, underwriting, and upselling / personalization. This is because AI and Machine Learning technologies and algorithms are fundamentally about extracting the most value from the information that we have, and excelling in these three areas depends on solving information asymmetries well.
The asymmetries revolve around the differences between what the lender knows about the customer and the true nature of the customer’s intent (with respect to fraud), financial state (related to underwriting), and aspirations and needs (personalization). The better the lender can use the information they have to accurately judge these for each and every customer, the more profitable and efficient their business will be.
AI can help in the other areas as well, but in this article we focus on how AI can help in solving the information related problems.
Fraud
Online fraud is a very challenging problem. In general, so many electronic transactions happen, but only a few are likely to be fraudulent, and decisions have to be made quickly. So it is truly a problem of finding needles in haystacks.
Another challenging element is that the fraud adversary is not a stationary target. Fraudsters are smart and may themselves be proficient in AI technologies. So fraud prevention is also a kind of arms race. Fraudsters live off of generated information asymmetries, and AI in fraud prevention is designed to dispel them.
A number of companies bring AI technology to this challenging problem.
Creditworthiness / Underwriting
Imagine what would go through your mind if someone asked you for a loan. You might think of the whole history of interactions you may have had with that person, perhaps how you have seen the person behave in various contexts, you might think about what you may have heard about that person from others in social contexts, you might even recall what you understand about how they think about money matters, you might think about any number of highly detailed information related to the person.
Clearly a lender does not have this much information and (thankfully) in many cases there are regulatory limits on the kind of information a lender can use. But the lender seeks to make critical business decisions optimally from limited information.
This is the kind of setting where AI and ML can make a big difference. For example, a number of companies operate in this space such as Zest.ai and Upstart.com. These companies are essentially building their own credit scoring models or custom credit scoring models for lenders, that are distinct from traditional approaches based on industry standards like FICO scores.
These companies find that AI allows them to not only reduce default rates, but, crucially, also improve access to credit by identifying creditworthy borrowers who would otherwise be denied by traditional underwriting that does not use AI.
In this way, AI truly delivers on the promise of making better decisions by helping to solve legacy problems of information asymmetry but making use of many additional data points at scale and without biased human judgement. These success stories also happen in the context of full regulatory compliance.
Upselling / Personalization
In another article on AI in Lead Optimization, we discuss at length the problem of lead optimization. This is because acquiring customers in a competitive online fintech marketplace is extremely expensive. As a result, once a company has acquired a customer with hard-fought and expensive effort, it is critical that the company makes the most of the opportunity by developing the relationship with the customer and using the interactions to offer other services that will benefit the customer.
As we alluded to in our article on AI in Lead Optimization, this kind of challenging problem is not very different from the challenges faced by any other online retailers such, e.g., Amazon which offers other things a customer might be interested in or even Netflix which is continually offering new content that it hopes the customer will enjoy.
These giant companies invest billions in AI technology and talent to maintain the highly profitable competitive advantages they have. Usually, the AI technology that is relevant here is known as “recommendation systems”. A few companies have found success in applying those technologies in the fintech space, including some of the largest financial services companies in the world such as Chase Bank.
