Machine learning-based clinical decision support in the intensive care unit has the potential to improve physician decision-making and patient outcomes. The gap between the development of models and their deployment at the patient's bedside compromises these results and has multifactorial causes. The article focuses on the selection of a clinical problem as a critical phase that contributes to the gap. Problem selection requires the involvement of a multidisciplinary team, but a practical framework for evaluating candidates is lacking. The authors propose questions based on information value chain theory, focused on complexity and actionability. They operationalized these questions into a CAPE checklist to assess a problem's readiness for deployment or need for reformulation. Going forward, they suggest optimizing care performance in parallel with decision support for maximum value and scalable improvements in patient outcomes.