Artificial intelligence in customer service interactions: From multi-layered information to organizational insights

Services shape our daily lives, from ordering coffee and streaming our favorite TV show to navigating technical issues and accessing healthcare. These customer service interactions generate vast amounts of information every day. However, organizations struggle to systematically analyze and learn from these interactions due to their volume and complexity. This dissertation examines how artificial intelligence (AI) can automatically extract meaningful patterns and insights from customer service interactions, transforming them into actionable information for organizations. Central to this approach is the recognition that service interactions contain multiple layers of information: explicit content (what is actually said), implicit signals (emotions, dissatisfaction, and call quality), and broader interaction-level patterns. Each signal requires specialized analytical techniques to unlock its value.

The dissertation contains six interconnected studies, starting with a systematic literature review that establishes how technology can augment human service agents. From a customer-centric perspective, a comparative analysis of neural network approaches for speech emotion recognition shows that simpler models perform as well as more complex ones, while being more efficient to deploy. Building on this audio-based foundation, subsequent research investigates customer dissatisfaction detection from both text and audio, demonstrating that multimodal approaches significantly outperform single-channel methods. Shifting to a service agent-focused perspective, AI can automatically identify different response strategies that companies use in social media service interactions, with custom-trained models outperforming general-purpose pretrained models. The research then explores how computational approaches can assist human evaluators in call quality monitoring while highlighting the continued importance of human judgment. Finally, a curiosity-driven approach for aggregated knowledge extraction demonstrates how AI can dynamically structure collections of service-related documents into ontologies.

Based on these chapters, this dissertation offers several key insights. First, service interactions contain various layers of information that can be extracted automatically. When aggregated across interactions, this information enables organizations to move from reactive problem-solving for individual interactions to developing strategic insights over multiple interactions. Second, the optimal AI approach varies depending on the specific task and context. The results show that simpler, domain-specific models often outperform complex, general, or pretrained models. Third, the findings suggest that the most effective applications emerge when AI technologies are combined with human expertise, rather than replacing human judgment.

This dissertation establishes customer service interactions as underutilized sources of competitive advantage. It demonstrates that AI can systematically extract value across multiple information layers of service interactions. Rather than automating service delivery, these findings suggest that AI systems can provide analytical support that enhances human understanding of service interactions while preserving the essential human elements of customer service. The studies offer both a theoretical understanding of how AI can process interaction data and practical guidance for organizations that are seeking to implement these technologies effectively, contributing to knowledge at the multidisciplinary intersection of service management, information systems, and artificial intelligence research.