How accurate are scoring models anyways? In the 1990’s and early 2000’s, scoring models were coming on the scene within financial institutions. At that time, business leaders with decades of experience were skeptical and had much cynicism towards these types of predictive models—there’s a certain pride that comes with years of business experience after all, and how could a machine replace years of experience! Notwithstanding, nothing competes with empirically derived models, where machine learning techniques are used to train a model based on historical data with known outcomes.
That was a difficult period for us data scientists who were faced with all this skepticism and adversity—middle managers everywhere, and even some executives felt threatened and were overreacting, thinking they had just lost much of their value to the enterprise. But you can’t argue with statistics, and eventually a new era was ushered in. Today these models are not only powerful, but sexy! Adoption really started to grow over the last decade when Hollywood (and corporate marketers) learned how to sell more stuff, touting the virtual clairvoyancy of AI and ML models. Today you see AI and ML in all forms of advertising, and it is difficult to sift through all the hype.
Based on the experience of this humble author, here is an oversimplified chart of the type of predictive accuracy one might expect from different types of decision models:
(1) Not recommended
(2) Far better than a coin-flip, but can't compete with empirically trained models
(3) Same output format as empirical models, but not empirically trained (humans set weights using experience and intuition)
(4) Logistic regression models - excellent power, easy to explain predictions
(5) Artificial neural networks - slight lift over logistic regression models, but difficult to explain predictions
Lendisoft supports the first 4 types, and even comes out-of-the-box with Expert Scoring Models so you’ll hit the ground running with our highly advanced Risk Grades for actives, and Collectability Grades for charge-offs. From there we’ll work together on building Empirically Derived Scoring Models, which will boost predictive accuracy significantly. The “net net” of all this means you’ll be able to create risk-based strategies that help you generate more collection revenues, and with fewer staff—and that’s having your cake and eating it too! For more information, schedule a demo today!