While large-scale industrial applications often rely on big data to train generic models on standardized equipment, it’s a different story for use cases related to sustainable energy production and consumption. Diversity in equipment and situations leads to scarcity in data, and generic approaches don’t typically work. To address this issue, we propose a two-fold approach. First, we combine traditional machine learning techniques with embedded business knowledge to develop bespoke models. Second, we use a data-centric approach to curate data and train models at scale. We have used this approach to industrialize applications that increase operational efficiency and energy efficiency. In the first case, we developed digital platforms for the predictive maintenance of equipment in thermal and renewable power plants. In the second one, we productized carbon footprint measurement and reduction strategies for large campuses and multinational organizations. As we head into more distributed energy generation systems, we need to evolve our AI toolkit to address challenges in a decentralized, fragmented ecosystem.