By Rhyne Brown, executive vice president of NAI Global

You have likely already heard about how artificial intelligence (AI) is changing one industry or another.

In fact, most of the advancements in analytics people are ascribing to AI are actually done using machine learning (which goes by the much less well-known moniker of ML).

Machine learning is a computer’s ability to receive data points and, based on pre-programming, enable a computer to decide if a value is within preset limits. ML programs can then make a decision to identify and record data points as an anomaly and, in some cases, then create an instruction to take an action intended to return the naughty variance to its proper place.

Here is how that would look for buildings: Let’s say that the sun rises on Tuesday at 07:15, and the weather is sunny and bright. This fact results in the building’s eastern face having increased heat load, and interior space warms up. The building has thermostats that soon register air temp above 76°F on the building’s exposed side.

A pre-programmed software connected to the HVAC system instructs that more chilled air is needed when an interior space reaches 76°F. To achieve the goal of a cooler building, the chillers get a message to crank up the production of cool air and send more conditioned air to the hot spots, perhaps doing this by also opening dampers in the ducts to achieve the needed result.

Chillers and fans continue this work until the temperature returns to another pre-programmed level, say 73°F. This one-dimensional action taking is machine learning. The key here is that the schedule for the action is preset.

Artificial intelligence is a computer technology that starts with a pre-programmed ML landscape, but it has a more profound capability. AI emulates the process of human thinking. People think in multiple dimensions that result in what we term ‘reasoning.’ Reasoning, as defined by Oxford Press is: “the action of thinking about something in a logical, sensible way.” Fully-evolved AI considers and processes complex decisions that result in simulating human intelligence.

Take the situation above. Let’s now assume that the AI program has access to more than one source of data, more than just thermostats. For example, what if this program has a headcount and distribution data of humans in the building? What if the program has data on outside air temperature that suggests the chiller can still bring in outside air, resulting in less costly cool air production? What if the program knows that there is a high likelihood that every Tuesday the conference room in the overheated space has a pattern of use that will fill it with sales agents at 10:00. What if the cost of electricity is lower if consumed before 09:00?

These items are multi-variants. Facts that AI can and will consider in order to refine a more complex but better solution. This reasoning capability can potentially save a lot of resources.

Buildings that use AI can set a goal to find the best solution based on data. Calculations can now be completed hyper-fast on an ongoing, real-time basis. The result is the ability to operate property at the highest efficiency level possible. Do not just pump in chilled air until the thermostats are happy. AI also records and learns this solution, building a deeper database. It becomes the basis for future problem-solving as more data is ingested into its memory bank. Said differently: quality AI continues to drive efficiencies.

In some buildings, over a million data points per day are exchanged between building data sensors and a comprehensive AI program that operates the properties. This can help lead to some very important real-time adjustments. When an 85 000 square foot building north of L.A. was directly downwind of a significant wildfire early in 2020 the building was able to scrape data not just from mechanical systems in the building but also from outside data sources, including, in this case, Open Weather. Part of the data readout from Open Weather is local air quality. As the fire raged ten miles to the east, outdoor air quality started dropping like a rock.

Within twenty minutes, smoke was so thick tenants could not see the other side of the building parking lot, and the outside air was very hazardous to humans. As programmed, the AI undertook a number of actions to protect the indoor air quality for occupants. The program shut down the outside air intake as its initial corrective action. Using daily logic of tenant patterns inside the building, the program altered the flow of interior air to spaces occupied by people.

This particular property has a part of the HVAC systems known as an economiser. This part of the HVAC system is controlled by local sensors that collect air temperature and humidity data. The sensors also trigger a process that can control a damper that opens or closes on demand. For example, if outside air is cooler than recirculated interior air, the system can draw this cooler air into the building’s HVAC airstream. The mixture lowers the air temperature being chilled and, in turn, again reduces the energy cost to produce cold air. However, the air quality was the issue and the AI thus closed the outside damper.

AI’s ability to make precise alterations to the internal air distribution system was a function of two other aspects of a proactive AI system: knowing how many people were in the building and knowing the air quality data both inside and outside. In this case, the property owner had installed a smart people counting platform that uses ‘cameralytics’ throughout the building.

During the fire, the AI would continue to adjust airflow based on need. As people left the building in the afternoon and occupancy densities dropped in some areas, the system continued to redirect inside airflow. If the inside air quality dropped to an unhealthy level inside, the system would alarm, and the property manager and tenants remaining in the building would be advised to leave.

This was a rather extreme case but even without hazard levels of smoke AI is able to balance human wellness with reduced operating costs. For a 100 000 square foot mid-sized office building in Southern California, one landlord’s sole focus was to improve his building’s balance sheet, like many other building owners. The key to his achieving that goal was reducing power consumption. The owner wanted to know how to accomplish this goal without significant new capital investment for an aging mechanical system.

This building’s AI system was designed to connect all of the following systems: cooling, heating, elevating, lighting, water, grounds irrigation, existing BMS systems, and other electric power users into a single data hub located in the building. This hub is known as an edge controller, and it, in turn, is linked directly to a cloud-based artificial intelligence program. The goal is to find and maintain harmony and balance between the owner’s desire for savings and tenants’ needs for comfort and wellness. 

This is a condensed version of this article, the full version can be found on