As a result of the electric grid’s growing penetration of variable renewable energy (VRE), supply-side management is more constrained, heightening the challenge of balancing electricity supply and demand. In turn, electricity providers have focused on demand-side management (DSM), a group of strategies that reduce electricity demand at a specific time, or shift demand from one time to another. In addition to improving grid reliability, these strategies can also provide value through decreased emissions for grids with high fractions of renewable energy. Throughout the day, there is variation in which resources are dispatched in response to changes in demand. Thus, any DSM strategy that shifts load from one time of day to another changes the total emissions associated with this load. Marginal emissions factors (MEFs), which describe the changes in emissions of a specific pollutant per unit of change in demand, are one method of quantifying the changes in emissions associated with a change in demand that occurs in a specific region at a specific time. In this study, we introduce a bottlenecked multi-layer perceptron model that uses temporal and electric grid features to predict CO2, NOx, and SOx MEFs at an hourly resolution for the grid overseen by the California Independent System Operator (CAISO), a high solar-penetration grid. We then apply the social costs of these pollutants to determine the potential value of DSM on CAISO’s grid. Our model predicts these MEFs day-ahead, making it possible to plan load-shifting strategies ahead of time and better incorporate the social cost of these gasses in DSM planning, and can also be used to inform users real-time about the environmental impacts of their consumption behaviors. As grids across the US transition to higher fractions of VRE, this framework for estimating the social value of DSM will become increasingly relevant for consumer-based programs.