Energy consumption represents a considerable concern for industries in terms of cost and the need to adopt future green energy sources. As a result, several recurring research initiatives seek to reduce it. In this work, we propose a data-based approach to optimize the overall process of fiberglass manufacturing, aiming to reduce the energy consumption of an industrial boiler. Initially, the process parameters with the most significant impact on the plant’s energy consumption were identified. A machine-learning model was developed using the historical factory data collected for these parameters. Then, we formally relate these parameters to the energy consumption of the analyzed system. With this, an approach based on genetic algorithms was used to search for the ideal values of the model’s input parameters that optimize energy consumption for particular production demands. As a result, we have formulated an operational approach that effectively guides the daily functioning of the modeled system, contributing to the reduction of energy consumption per unit of production. We show that simple learning techniques, such as decision trees, can efficiently model this class of problems. This is a crucial design requirement as the tool can be deployed on IoT devices with limited processing resources. The application of our study made it possible to save thousands of dollars in boiler energy consumption costs. The results show that monthly savings can reach $8440.