Current customer service chatbots often struggle to meet customer expectations. One reason is that despite advances in artificial intelligence (AI), the natural language understanding (NLU) capabilities of chatbots are often far from perfect. In order to improve them, chatbot managers need to make informed decisions and continuously adapt the chatbot’s NLU model to the specific topics and expressions used by customers. Customer-chatbot interaction data is an excellent source of information for these adjustments because customer messages contain specific topics and linguistic expressions representing the domain of the customer service chatbot. However, extracting insights from such data to improve the chatbot’s NLU, its architecture, and ultimately the conversational experience requires appropriate systems and methods, which are currently lacking. Therefore, we conduct a design science research project to develop a novel arti-fact based on chatbot interaction data that supports NLU improvement.