In response to the urgent need to mitigate climate change, significant efforts have been made to implement grid-interactive heating, ventilation, and air-conditioning (HVAC) controls, such as model predictive control (MPC), in real-world buildings to reduce peak demand and energy consumption. However, despite their potential benefits, conflicts can arise between the end-user and the MPC due to its supervisory control nature, leading to control action overrides. To address this issue, we investigated end-user thermostat usage behavior and improved the MPC structure to incorporate override actions in our demonstration project for K-12 school buildings. Through exploratory data analysis of the baseline data, we found that override behaviors are typically due to habit, rather than thermal discomfort. While the MPC can still achieve its goals under override actions, unnecessary habitual overrides could reduce the potential benefits of the MPC. To ensure optimal performance of the MPC, effective communication efforts are crucial to ensure that the end-user understands the project. In this research, we present our efforts and experiences in the project, including the improved MPC structure and our recommendations for effective education and communication strategies. As a result, the MPC reduced peak demand by at least 25% in seven school buildings by controlling 15 rooftop units (RTUs), while allowing the end-user to retain controllability over thermostats.