Toward Greener AI-Based Smart Services

As digitalization accelerates, AI-based Smart Services have become increasingly vital, but their energy consumption is emerging as a critical concern. Despite widespread discourse on Green AI, a structured methodology to assess energy efficiency throughout the development lifecycle of Smart Services remains underdeveloped. This paper introduces a practical framework that maps energy efficiency measurements to the CRISP-ML(Q) lifecycle. Parameters influencing both training and inference phases are identified, categorized, and evaluated. The study proposes prioritization rules based on service type and usage patterns and contrasts hardware-and software-based measurement tools. We present structured visualizations and tools-based comparisons to guide the measurement of AI energy demands. The outcome is a lifecycle-aware model that facilitates more precise, comparative, and operationally relevant AI efficiency assessments, providing a foundation for sustainable AI deployment in service innovation.