Adaptive vs Static

Mar 27, 2019

The Business Case for Classical Machine Learning in Credit Decisioning

About two years ago, I had a meeting with the Chief Risk Officer of a large sub prime lender in Toronto. Business was booming, delinquencies were high but contained within a range and the CRO was disinterested in using Machine Learning for Credit Decisioning. He felt that his algorithms were time tested, adaptive in his mind didn’t mean much and he just saw this as incremental costs for his business. He had built a large army of thirty people just for underwriting and didn’t see why he needed to change the approach. We just agreed to disagree.

So, is there a business case for Machine Learning in Credit Decisioning? Why are banks and lenders investing in Machine Learning?

To answer this, lets flash forward to today, two years after that meeting. I was at a dinner and ran into an executive who dealt with this lender helping them with their collections. During dinner, he was telling me about his lender client, the sharp increase in their delinquencies and write offs (more collections work for his company), totally unexpected by the client (but a windfall for his collections company). Result – this lender had to cut their budgets and spends and were laying off personnel. The immediate cost is in the many millions. Longer term, the company has tightened its credit criteria which constraints growth and further accentuates the problem.

And therein lies the tale. Algorithms are static. They tell you what happened in the past but do not capture whats happening on a day to day basis. And there is a huge implicit cost to staying static. This lender is not alone. There are tens and thousands of lenders still dependant on algorithmic models.

Time for Change. Adaptive IN. Static OUT.