Software Escalation Prediction with Data Mining

One of the most severe manifestations of poor quality of software products occurs when a customer “escalates” a defect: an escalation is triggered when a defect significantly impacts a customer's operations. Escalated defects are then quickly resolved, at a high cost, outside of the general product release engineering cycle. While the software vendor and its customers often detect and report defects before they are escalated it is not always possible to quickly and accurately prioritize reported defects for resolution. As a result, even previously known defects, in addition to newly discovered defects, are often escalated by customers. Labor cost of escalations from known defects to a software vendor can amount to millions of dollars per year. The total costs to the vendor are even greater, including loss of reputation, satisfaction, loyalty, and repeat revenue. The objective of Escalation Prediction (EP) is to avoid escalations from known product defects by predicting and proactively resolving those known defects that have the highest escalation risk. This short paper outlines the business case for EP, an analysis of the business problem, the solution architecture, and some preliminary validation results on the effectiveness of EP.