Modelling Mismatch In Predictive Analytics: A Case Study Illustration And Possible Remedy

Predictive analytics (PA) is a popular approach to support managerial decision making. The core of a PA-based decision aids consists of an empirical prediction model. It is common to create such models using statistical standard methods. We argue that this approach suffers from a partial modelling mismatch (PMM) because the internal (statistical) objective of the decision support model and the business objective of the decision maker differ. We explore the severity and consequences of PMM in the context of resale price prediction, which is an important decision support task in the car leasing industry. The key hypothesis of the paper is that PMM decreases the quality of decision support. To test this, we develop a modelling methodology that creates predictive decision support models in a way so as to account for business objectives. Empirical experiments on real-world data confirm the effectiveness of our approach. In particular, we find evidence that i) PMM can substantially hurt decision quality, that ii) the new modelling approach is a suitable remedy, and that iii) it is generally important to consider the business objectives of decision makers when devising a predictive decision support model. With standard prediction methods falling short of this requirement, the paper makes a first step toward more business orientation in PA.