AI/ML use in banking & finance: will it kill capital advisory in the future?

For eons, financial institutions across the world have approved both personal and business loans using what can be described as relatively simple due diligence and vetting practices.

As a child, I can recall applying for my first personal loan as if it were yesterday. Aside from a cursory application being filled out, my loan was approved by the bank manager on the long-standing banking relationship the bank had with my parents.

Thus, in my case my loan approval hinged more on the “intangible” aspects of my personal character (or more precisely my family’s), versus my ability to debt service the loan, my beacon score or my steady employment history (I was making $40/month as a paperboy in Winnipeg at the time).

In Malcolm Gladwell’s latest book, “Talking to Strangers” he posits that human beings, by nature, will see another person for what we think and know that person to be (the “truth”), often eschewing signals (often very slight and not at all obvious) that point to a different character.

Now while Mr. Gladwell’s book is not about finance, the book’s main theory, coined the “default to truth” is how I believe my first loan application was approved. That is, I was approved based on someone’s view of who they thought I was and what the past said about how I would act in the future (i.e. say in a loan default situation). Any signal, however slight, that I might’ve shown that I was an irresponsible youth, without steady financial means to repay said loan, were likely quickly dispelled by the bank manager. He defaulted to truth.

So what does this have to do with using AI/ML in banking, finance and capital advisory?

Well lots actually.    

Typically, credit information to approve a loan application is provided by the borrower and might include a paper questionnaire (aghast!) being filled out, including the borrower providing other supporting documentation.

Information on hand, and with access to a borrower’s credit score (in the case of personal borrowings or personal guarantees), the data is crunched by the lender using multiple regression, among other quantitative techniques, to assess the credit risk of the customer and then either provide an approved or declined credit decision.

Yet despite all this technical quantitative analysis, something very personal and subtle is happening behind the scenes. That is, loan and credit officers are making judgements about a borrower’s propensity to default based on personal biases. This default to truth, or more simply what the lender sees in a borrower while speaking and interacting with them (i.e. the “character” of the borrower) often weighs heavily in the loan approval process.

Yet despite the sophistication of the financial sector, we as lenders and advisors continue to be duped by phony borrowers who appear to be legitimate.

To combat this, increasingly financial institutions are using artificial intelligence (AI) and machine learning (ML) techniques to analyze multi-dimensional data. The combined use of AI/ML models to support credit decisioning from different digital platforms (e.g. mining public sites for negative news about a borrower that might act as an early warning signal) and applications, in turn improving the accuracy, scalability, and efficiency of the credit decisioning process, has stimulated the development of advanced credit scoring approaches.

The question thus becomes if AI/ML will take over credit decisioning, thereby negating the need to bring one’s “gut instinct” to the credit decisioning table, where is the human touch in this equation?

When I founded Kaeros, I believed that it would be a long way off before AI/ML renders me and my firm’s services obsolete. I still believe that multivariate analysis, specifically in M&A and cash flow financing, will still require lenders, financiers and advisors to draw on years of experience to properly assess credit risk.

While behavioral scoring models, often used in personal and SME (small medium enterprise) credit adjudication appear to properly capture, and correctly analyze, the multitude of data points that determine credit risk and default risk, in more complex transactions a formal credit committee of people still makes the final decision.

Few can dispute that AI/ML, quantum computing, credit analytics using machine learning are the way of the future in loan adjudication.

The question thus becomes how fast will the adoption be, who will adopt first and what involvement with humans have in the adjudication process of the future, if any?

Food for thought.

Trevor Palmquist

Founder & Managing Director