Waste of prepared but unconsumed meals in collective nutrition systems generates material losses and nutritional deficits. Data from the Military Academy of Belgrade show that in 2023, 36,284 meals (5.81% of total output) were left unconsumed, corresponding to a direct ingredient cost of 5,401,579.22 RSD. This paper presents a conceptual architecture for an intelligent Decision Support System (DSS) for menu adaptation that transforms static, expert-driven menu planning into a predictive, data-driven process. The system consists of four modules: a data module, a predictive module (machine learning), an evaluation and ranking module (DEA + BWM + R'MAIRCA), and a menu adaptation module. The implemented evaluation and ranking module was validated on Military Academy data: 10 out of 31 analysed dishes were rated as efficient, and replacement of one highly unconsumed dish was confirmed by a pilot test (218/218 portions consumed; sensory score 9.6 +/- 0.7; estimated annual saving ~50,000 RSD). The predictive module and coupling mechanism are proposed and constitute the research agenda. The system is applicable in military, healthcare, and educational institutional food service settings.