This paper presents the Define-ML framework, which extends the existing Lean Inception technique to address ML-specific challenges such as data dependency, technical feasibility, and alignment between business goals and probabilistic system behaviors in the ML product conception phase. Define-ML systematically integrates data and technical constraints from the early stage by adding activities such as data source mapping, feature-to-data source mapping, and ML mapping. Static and dynamic validations following the technology transfer model are performed through toy problems and real industrial case studies, and usability, ease of use, and adoption willingness are evaluated through surveys and qualitative feedback. As a result, Define-ML is shown to be effective in clarifying data-related issues, aligning ML capabilities with business goals, and promoting cross-departmental collaboration.
