Although conversational agents are highly popular and profitable systems in a number of diverse settings, research on the human-like capabilities of these agents is still in its infancy. Another issue that impedes progress on these systems is the tangible discrepancy between output of the scientific community and its implementations in industry. Our project aims to tackle the lack of connotative commonsense knowledge (viz. emotions) in text-based dialogue systems. We will therefore (i) implement novel machine-learning models to detect explicit and implicit fine-grained emotion trajectories, and (ii) generate appropriate response strategies for the detected emotions. To bridge the gap between research and industry, we focus on domain-specific conversations in Dutch and English customer service. Our final system will be validated in terms of portability and utility in various use cases.