How To Calculate The Value Of AI Lending
We've written about how to compare the statistical outperformance of AI-driven credit underwriting models over traditional credit models and scorecards.
Now to answer a more fun question: how much more revenue and profit can AI produce for my financial institution?
To answer that, you need to do a swap-set analysis, which compares a hypothetical machine learning-scored portfolio (ML) to a benchmark portfolio. Most loan applicants are likely to be approved or denied just the same by both types of models. The swap-sets are the more influential applicants who were approved or denied by only one model but not the other. Typically we see a good 15-25% moving from in to out depending on how lenders adjust their risk thresholds and booking rates. By counting up the number of applicants who move (and analyzing their risk tiers) you learn what you need to know about how the ML model would have performed if it was in production during the test period. (Zest software automates this analysis so it's instantly apparent what's going on.)
In theory, since the ML models do a better job at replacing high-risk borrowers (swap-outs) with low-risk ones (swap-ins), you can and should be approving more borrowers without increasing risk. The chart below shows how an ML-scored credit card portfolio would have achieved significant growth across all credit tiers by generating more swap-in borrowers than the swap-outs removed for risk concerns. (The K-S score on the left axis is a measure of statistical accuracy. A higher bar is better at predicting which borrowers will go 60-days delinquent within the first two years.)
From there, it’s not that difficult to apply these higher approval rates in your internal economic models to arrive at a new, higher originations amount across your product lines, plus any impact from lower charge-offs. From there, you may want to tweak your assumptions around booking and funding rates. ML’s ability to price across risk tiers more precisely and effectively allows you to bring out those risk-based pricing programs you've always wanted to use. More accuracy also means more instant approvals and faster decisions, which can lead to higher booking rates.
Whatever economic gains you arrive at, remember to factor in the value of timely decision-making. One credit union customer of ours ran a financial analysis that showed an increase of $1 million in incremental profit -- per month -- from switching to an ML model. You can use that kind of immediate ROI to fund a host of other digital transformation initiatives. Why wait?