Artificial Intelligence (AI) has had a significant impact on the whole of the fintech sector.
Continuous development and application of Machine Learning (ML) have become a fundamental part of innovation, streamlining processes, and cutting costs.
According to Statista, the market share of AI within fintech is projected to triple in value by 2026, reaching $ 26.67 billion.
Know Your Customer (KYC), and Anti Money Laundering (AML) are areas where AI has been particularly successful — critical areas of Credit Underwriting.
As part of Fintech Nexus USA 2022, Trust Science sponsored a track focused on the intersection of AI and Credit Underwriting. Panels tackled the intricacies of the new technology, an area that regulators have only approached.
“I think the way to think about AI is not to worry so much about the artificial portion. Think more about the intelligence,” said Pankaj Kulshreshtha, CEO of Scienaptic AI.
“When you define intelligence in the context of financial services organization, I think there are many smart and intelligent things that financial institutions have been doing already. Financial institutions are ahead in this game of doing clever stuff with data for a long period of time.”
Data and alternative data were the buzzwords of the track. Many panelists recognized the limitations of the current credit underwriting process and the data considered, calling for alternative data usage to better serve customers with credit options.
Humanising AI to improve growth in credit industry
In many cases, AI has been implemented to assist in decision-making. Based on data points and quantifiable facts, it avoids the messy world of human emotions and prejudice (discounting prejudice in historical data bias).
Despite this, some call for an increasingly human approach when adapting AI for future applications.
“I think the change we have to make in our thinking is to start with the customer,” continued Kulshrestha.
Many echo this idea. Murli Buluswar, Head of Analytics at Citi, agreed, “An absolutely critical path for us to go from here to where we want to go as an industry over the next decade is to have this ability to have a 360 view of the customer.”
“A single record that stitches together every interaction and transaction at a customer level that helps you understand their needs, their latent needs, and helps you think through more intelligent ways in which you could build relevance and engagement with your customers.”
Typically, automated credit scores focus on historical data showing payment behavior, a tiny part of the customer’s life. Many factors make up a customer’s ability to pay back the credit, resulting in undue rejection of customers and loss of business opportunity.
To open up services to as many consumers as possible and provide certain demographics with more specific credit lines tailored to meet their needs, alternative data consideration is seen as essential.
“We know there’s a lot of other data out there that could be used to make a better decision,” said John Witterschein, Vice President, Consumer Credit at Bethpage. “In fact, for existing members, we look at the positive data as part of our manual decisioning if it comes to a manual decision.”
“So why can’t these decisions be automated, leveraging all of this data that will speak to the credit worthiness of a borrower? If you got a good education, a home, and a bank account with good cash flow, those are things that will demonstrate that you’re going to have the ability to pay.”
The added facet of alternative data would bring a holistic view into the automation of credit decisions, creating a picture of the customer which looks beyond their payment history.
“You need sharper underwriting tools which use more data, better scores, more scores, to actually pinpoint which other cohorts you can still lend to. You have to be smart about making sure that you are writing while you control some of the riskier loans. You continue to get more good loans on your book,” said Kulshrestha.
Cash flow data could be the answer
Fin Reg Lab has conducted investigations on cash flow data since 2019 further to understand the potential of available alternative data sets.
“We wanted to understand if cash flow data offered insights that could help lenders actually extend credit to populations who otherwise would have been turned down and even populations who don’t have a credit score at all, to begin with,” said Melissa Koide, CEO and Co-founder of FinRegLab.
“There, we found that the data was independently offering additional lift or insights over and beyond what the bureau data would tell us.”
“There are really important questions in AI research around whether the data create disparities between protected class and non-protected class populations. From the particular analytics we undertook, we found that the data was not creating that disparity that would generate a potential disparate impact risk.”
Underwriting, at its core, assesses whether a customer can repay a loan. Many experts agree that cash flow data, as well as information on assets, paints a broader picture of consumer credit worthiness.
“Cash flow data gives us the ability to fill in much more of that overall financial picture,” said Jason Gross, CEO, and Co-founder of Petal and Prism data.
“If you think about traditional data, or what’s in your credit report that drives your credit score, it’s the liability side of your balance sheet. But it’s missing your assets, cash flows, and income.”
“If we thought of consumers as we think of businesses, it’s really hard to understand what’s going on with the business by looking only at their liabilities. The key elements that drive our cash score, and the type of underwriting that we’ve been doing, have to do with those core economic fundamentals in the cash flow and the asset profile of the consumer.”
Five points to consider when applying AI
Although, in many cases, AI improves underwriting processes, it seems the complete reduction of human input may not be the best solution for all. Automation based on a credit score could lack the broader view that may be necessary when assessing applicants’ access to credit, unduly turning away customers.
Murli Buluswar found that although there are high levels of technical and operative talent in AI, at times, a creative outlook is lacking. He felt this approach was essential, especially when deciding how to underwrite in alternative ways.
Buluswar explained the proper application of AI to automate underwriting can be assessed when taking into account the following five points:
- Is there a significant customer or regulatory exposure to that problem or opportunity?
- Are there manual decisions being made on scale?
One of the key reasons behind the implementation of AI is to simplify, streamline and automate repetitive processes on a large scale. If a lot of manual decisions are made, it might be suitable for automation.
- As a consequence of those manual decisions. Do we have partial coverage in really solving that particular problem?
As a result of manual processes, it may mean there is no capacity to address a problem thoroughly. The problem would benefit from AI and ML to assess more data points efficiently and accurately.
- Is there a significant data exhaust that has been created by that process that isn’t being tapped into?
The issue with alternative data is that it can be spread over different formats, which need to be examined to create suitable input data. For this reason, some areas are not yet enhanced to their full capacity. This data, if processed, could result in much more efficient solutions and the creation of new credit products.
- What are the financial implications of fully automating that process and taking the human out of the loop?
The process of implementing AI requires significant investment. Although, on average, AI can cut costs, as with any investment, deciding whether to apply AI also involves assessing the value it creates.
