Firms employ a variety of passive listening platforms (e.g., customer comment cards, help lines, social media) that elicit a variety of complaints. Since these complaints are self-selected, it isn’t obvious if complaints are widespread or genuinely indicative of systemic service failures. In other words, how can can a firm know when complaints can be ignored and when they need to addressed? This issue is related to a larger class of problems where the goal is to make inference about the size of a population from a non-random sample (e.g., the German tank problem). In this paper, we develop a model that uses a variety of observed customer complaints on Twitter to make inference about the severity of service failures.
For general details on GitHub usage, project organization, and project workflow, see Research Assistant Training.