Abstract
This short-paper examines the tensions between user experience (UX) designers and data scientists in requirements engineering for machine learning (ML) systems. Through a case study at Marcomp, a maritime service provider, we investigated how an image recognition ML project revealed fundamental differences in approaches to requirements elicitation. Data scientists adopted a model-first trajectory, focusing on the model's accuracy and expecting users to adapt to its constraints. In contrast, UX designers prioritized user workflows, questioning whether the solution satisfied user needs. Our findings demonstrate that while both groups work iteratively, they conceptualize requirements differently: data scientists iterate to improve model performance, whereas UX designers iterate to enhance user value. This disparity, amplified by management's push to implement AI technologies, resulted in a system that improved administrative efficiency but disrupted users' workflows. In this short-paper, we contribute a descriptive story to deepen our understanding of how the perspectives of software development practices and data science methods differ in requirements engineering processes for AI systems.