This study is carried out to develop a smart ore blending methodology for high carbon ferromanganese production units. Geometallurgical characterisation of the ores collected from 10 different mines has been carried out to estimate variation in their aptness for the alloy production process. An evolutionary algorithm-based methodology has been adopted for ore blending to maximise the total manganese content of the ore blend under applied physico-chemical constraints. Analysis provides various blends of the same chemical composition but of different geometallurgical ranks. An artificial neural network model has been developed to predict the operational parameters such as slag and metal composition. This method allows the operator to visualise the impact of combinations of different ore blends on slag–metal composition as well as on electric power and coke required for smelting reduction of ores in the submerged arc furnace. It was observed that best blends could reduce power and coke consumption by 100 kWh ton−1 and 30 kg ton−1, respectively, but optimum values can be established in long run considering conservation of natural resources.