Considering Gen AI in Building Operations: Key Questions to Ask

We put our software and operation engineering heads together and came up with a list of 5 questions and use cases to help you evaluate whether gen AI is "fit for purpose" in plant operations.

To help you envision how generative AI could fit in your operations workflows, here are some questions and use cases to consider, as well as some specific drawbacks to be aware of.

1. Is your AI purpose predictable and repeatable? In other words, do you have the data and systems necessary to predict a building operation’s outcome? Is your AI purpose at least probabilistic?

Issues to Consider

Of course, the more predictable a need is, the better the fit for automation by a computer.

However, it’s critical to keep those probabilities in mind. Every building is a unique blend of probabilistic systems, and no occurrence of events can be perfectly predicted. There is ALWAYS a degree of error attached to any prediction.

That said, here are two places where AI might make sense in the operation of large facilities and campuses:

•  Occupancy predicting for demand/peak tonnage analysis and temperature control
•  Fault analysis for building mechanicals (AHUs, etc.)

2.  Does your building have the right sensors for your AI purpose, and can you accurately deliver the required data to your AI engine on time? Are your current building devices and systems (e.g., the BMS) ready and able to provide high-quality data?

Issues to Consider

• What happens if your data is bad (and we mean really bad)? For example, what if a sensor is out of calibration or in a failed state? How will you catch the problem before everything goes haywire?

3. Are your sensors accurate and well maintained? Is there bad data within your sensory data?

Issues to Consider

• Your site doesn’t have physical power meters, so usage has to be estimated based on fan/pump laws.
• So many meters are out of calibration that it’s difficult to monitor efficiencies on a per equipment basis.

If your building is in either of these situations, you are not a good candidate for gen AI.

4. Do you track important building changes in devices, equipment, occupancy, sensors, etc., to keep the AI models accurate over time?

Issues to Consider

• If some-to-all of your submeters are manually read, it will be very difficult to keep the AI holistically informed.
• Property managers will need to stay on top of changing tenant situations. For example, say a tenant moved to a different location within a building. Without proper updates about occupancy, your system may show two separate instances of this tenant and incorrectly adjust heating and cooling in the old and/or the new space.

5. Will building AI make life easier for your building operations staff? Whose job will it be to analyze and diagnose building-wide and building systems–wide issues? And how will the other operational responsibilities be divided up?

Issues to Consider

• Building engineers typically don’t have a second to spare. Many times these issues are left to the BMS guy, who isn’t likely to have experience working with gen AI. Similarly, the building AI cloud may detect building issues, but it doesn’t have the local skills, resources, or knowledge to fix the issue.
• Implementing and validating ECMs is highly collaborative and involves people from the plant engineer to the vice president. Like any new implementation, adding gen AI will require good and frequent communication if the effort is to be effective in the long term. 
• Gen AI’s impact on the commissioning and recommissioning of buildings is unclear. How will the AI participate in this process. Will it require its own recommissioning? Imagining that interaction is a headache for another article.