Does selection bias blind performance diagnostics of business decision models? A case study in salesforce optimization

Modern business decision models are often very complicated due to a deluge of information. Evaluation and diagnostics of such decision models is extremely challenging due to many factors, including the complexity and volume of data. In addition, since there is no ideal data sample to construct a control group for comparison studies, performance evaluation and diagnostics of business actions can easily be distorted by selection bias. In this paper, we design a framework to analyze this sample bias issue under a practical business scenario. In particular, we focus on: a) identification of the key factors which drive selection bias during the business decision; b) evaluation of the performance of business actions with consideration of the identified selection bias. We evaluate baseline analytics tools on the worldwide sales-force data of a large global corporation and clearly demonstrate that the selection bias issue makes the usual evaluation very unstable and not trustable. However, by removing such detected sample bias, our framework can generate reasonable diagnostics results across different dimensions. The implemented analysis tool was applied to a worldwide business opportunity dataset of a multinational Fortune 500 corporation; the analytics results clearly show the significance of such a bias detection-based evaluation framework for sales-force optimization.