Lessons learned to improve the UX practices in agile projects involving data science and process automation

User-Centered Design (UCD) and Agile methodologies focus on human issues. Nevertheless, agile methodologies focus on contact with contracting customers and generating value for them. Usually, the communication between end users (they use the software and have low decision power) and the agile team is mediated by customers (they have high decision power but do not use the software). However, they do not know the actual problems that end users (may) face in their routine, and they may not be directly affected by software shortcomings. In this context, UX issues are typically identified only after the implementation, during user testing and validation.Aiming to improve the understanding and definition of the problem in agile projects, this research investigates the practices and difficulties experienced by agile teams during the development of data science and process automation projects. Also, we analyze the benefits and the teams’ perceptions regarding user participation in these projects.We collected data from four agile teams, in the context of an academia and industry collaboration focusing on delivering data science and process automation solutions. Therefore, we applied a carefully designed questionnaire answered by developers, scrum masters, and UX designers. In total, 18 subjects answered the questionnaire.From the results, we identify practices used by the teams to define and understand the problem and to represent the solution. The practices most often used are prototypes and meetings with stakeholders. Another practice that helped the team to understand the problem was using Lean Inception (LI) ideation workshops. Also, our results present some specific issues regarding data science projects.We observed that end-user participation can be critical to understanding and defining the problem. They help to define elements of the domain and barriers in the implementation. We identified a need for approaches that facilitate user-team communication in data science projects to understand the data and its value to the users’ routine. We also identified insights about the need of more detailed requirements representations to support the development of data science solutions.