.webp)
Eliminating Manual Overrides First, Saving Energy Second: How Microsoft Built the Foundation for Future Energy Projects
Three months in, Microsoft's machine-learning HVAC system had netted roughly $3,500 in energy savings across roughly 50 buildings. Modest, by any measure.
April Yi, Director of Digital Engineering at Microsoft, presented those numbers at NexusCon 2025 without dressing them up. The dollar savings weren't the headline she wanted to lead with anyway. "If you ask my team, one of the things I hate is manual processes," she said. "Whatever we can automate, if we know and recognize that it's a repeated process that we're handling manually every time, let's automate."
What the system actually replaced was the daily ritual of manual schedule overrides. Static schedules across a 50-building campus rarely align with how the buildings are used. Schedules get nudged for after-hours events, moved meetings, or holiday weeks. Setpoints get overridden as a quick fix to a temporary problem. Every nudge is human time and a chance for the schedule and setpoints to drift further out of sync with the building.
The machine learning system took that ritual off the team's plate and replaced it with predictive schedules that adapt daily. Room meets comfort setpoints when occupants actually arrive, not on a schedule someone set last quarter. The team gets that time back.
This serves as an example of implications for energy programs bigger than the single-dollar figure. A reliable, trusted, and automated path from prediction to BAS is the operational foundation. With that foundation in place, additional energy work (sequence optimization, demand-side flexibility, deeper FDD integration) sits on top of something that already runs without intervention and isn't littered with setpoint overrides of unknown origin.
Register for the next Nexus Labs event.
Sign up for the newsletter to get 5 stories like this per week:
Three months in, Microsoft's machine-learning HVAC system had netted roughly $3,500 in energy savings across roughly 50 buildings. Modest, by any measure.
April Yi, Director of Digital Engineering at Microsoft, presented those numbers at NexusCon 2025 without dressing them up. The dollar savings weren't the headline she wanted to lead with anyway. "If you ask my team, one of the things I hate is manual processes," she said. "Whatever we can automate, if we know and recognize that it's a repeated process that we're handling manually every time, let's automate."
What the system actually replaced was the daily ritual of manual schedule overrides. Static schedules across a 50-building campus rarely align with how the buildings are used. Schedules get nudged for after-hours events, moved meetings, or holiday weeks. Setpoints get overridden as a quick fix to a temporary problem. Every nudge is human time and a chance for the schedule and setpoints to drift further out of sync with the building.
The machine learning system took that ritual off the team's plate and replaced it with predictive schedules that adapt daily. Room meets comfort setpoints when occupants actually arrive, not on a schedule someone set last quarter. The team gets that time back.
This serves as an example of implications for energy programs bigger than the single-dollar figure. A reliable, trusted, and automated path from prediction to BAS is the operational foundation. With that foundation in place, additional energy work (sequence optimization, demand-side flexibility, deeper FDD integration) sits on top of something that already runs without intervention and isn't littered with setpoint overrides of unknown origin.
Register for the next Nexus Labs event.
Sign up for the newsletter to get 5 stories like this per week:
This is a great piece!
I agree.