Advances in machine learning (ML) open up possibilities for better supporting the decision making that occurs in high-stakes domains such as air traffic management (ATM). The success of such decision-making systems highly depends upon end users’ involvement in their development process. However, most designers face challenges with finding appropriate ways of doing this. This paper presents our ongoing work to investigate design practices by reporting lessons learned from user involvement in the development of an ML-infused ATM decision support system. To explore if and how UX design methods need to be refined when working with ML as a design material, we conducted an online study with domain experts consisting of three iterations. The paper reports the main challenges we faced and our actions to overcome them. Our results can be useful to other designers working with ML-infused systems.