Untangling Human Interaction Patterns: Learning from Automated Emotion Detection in the Consumer Pension Context

Artificial intelligence (AI) increasingly enters multiple aspects of our lives, improving the efficiency and effectiveness of day-to-day operations (AI100 2016). Specifically pension service providers may profit from this development. Despite huge investments in call center labor and software, providers only manage to resolve about half of customer requests in a satisfactory manner (Upbin 2013). Emotions are critical to successful service interactions in general (Mattila and Enz 2002) and consumer pension decisions in particular (Bruine de Bruin 2016). A failure to recognize and appropriately respond to emotions can be detrimental for providers and customers alike (Grandey 2003, Henkel et al. 2017b).
The proposed research will answer to repeated calls for investigating the characteristics and processes of emotion cycles in service interactions (cf. Hareli and Rafaeli 2008). It will examine how the emotions of one party influence the emotions of another party. For instance, a customer may approach the pension provider expressing concern and fear or anger. Will the agent pick it up? How will it influence the agent and her response? What is the ideal response? To this end, a prototypical technology will be developed at the BISS institute to detect emotions in speech conversations of APG and PGGM with their customers. The technology will use single- and multi-label machine learning models (Bromuri et al. 2014), combined with event based alerting (Bromuri et al. 2016) for the agents.