These startups are using AI to cut energy waste from buildings

“Artificial intelligence” has captured the imagination of the world with its ability to churn out uncanny facsimiles of human-made writing, art and even computer code. To produce these outputs, AI systems suck up not only mind-boggling amounts of data and intellectual property, but incredible amounts of electricity as well.

But AI can also be a vital tool in saving energy — and cutting carbon emissions. That’s particularly true in cases where humans lack the capacity to comprehend the flood of data needed to make impactful, real-time decisions. In fact, the technology could be key to helping buildings reduce the stunning amounts of energy they waste from day to day — equivalent to about 30 percent of their total energy use, the Environmental Protection Agency reports.

Buildings use nearly one-third of the world’s energy and account for about one-quarter of global greenhouse gas emissions, according to the International Energy Agency, so cutting that waste isn’t just a money-saver, but a boon to climate goals.

Efficiency changes such as attempting to better align heating and cooling systems with actual energy use and carbon savings can make a major dent in this problem of wasted energy. But even seemingly simple tasks like this are a lot more complicated to enact than they sound — particularly when the people who are inside those buildings are messing with the controls.

At a minimum, each potential efficiency change comes with scheduling challenges and uncertainty around the financial payoff. But things get even more complex for building owners contemplating the array of HVAC improvements, building-envelope retrofits and clean energy options available for reducing their energy and carbon impacts.

Determining which options will yield the deepest reductions — and at what cost — requires analyzing historical utility billing and future energy-pricing data, understanding the shifting cost of equipment, materials and installation, and calculating all those variables against their targets and goals, whether those are set on their own or via mandates like New York City’s Local Law 97 and similar building-performance standards.

That means building efficiency is the kind of data-rich world in which AI can thrive. A growing group of property owners, venture-backed startups and building-equipment vendors are applying the latest advances in the field to solving these kinds of problems. There’s a particular focus on the office buildings among this group, since those buildings are used by big companies that are being held responsible for the carbon emissions from their energy use.

Many buildings — particularly older structures with outdated equipment — will require significant material upgrades to achieve the efficiency improvements set by mandate or internal sustainability pressures. But there’s a good deal of ​“wiggle room” for software, sensors and digital controls to make a difference, said Mike Kazmierczak, vice president of the Digital Energy Decarbonization Office within the digital energy division of Schneider Electric, a global player in building HVAC and electrical controls.

“There’s a learning approach of understanding where your energy consumption is today and what you want to tackle first,” he said. His office released a report last month conducted with global design firm WSP that found ​“digitalization” of large office buildings’ HVAC controls can yield carbon emissions reductions of up to 42 percent with minimal impact on building operations, yielding a payback on investment of less than three years.

“You don’t necessarily need to start with the most difficult things,” like replacing windows, building facades and fossil-fueled boilers, he said. ​“These are disruptive, and they also cost a lot.” Given that many commercial property owners only hold on to buildings for five to 10 years before selling them, ​“unless you’re going to hold on to the asset for many years, you want to look at things that are shorter-term.”

Here are three startups among the many that aim to harness AI for building efficiency and decarbonization.

BrainBox AI: Optimizing heating and cooling for buildings in flux 

Sam Ramadori, CEO of Montreal, Canada–based BrainBox AI, said his company’s artificial-intelligence approach to HVAC optimization stemmed from the decades of experience of ​“our inventor-slash-founder,” CTO Jean-Simon Venne, in ​“doing optimization the old-fashioned way.”

As a longtime executive at energy services companies, Venne ​“lived through the challenges of long development cycles and deployment cycles” for retrofitting buildings with new HVAC systems, Ramadori said — ​“and the building drift after the project was done.”

“Drift” is the term for a well-known problem with building efficiency-optimization projects. Simply put, the work of HVAC engineers to fine-tune a building’s energy performance can quickly fade as tenants move in and out, occupants demand or manually make changes to temperature and ventilation settings, and equipment such as fans and vents starts to malfunction or operate at less than optimal efficiency. Research from Texas A&M University and the Lawrence Berkeley National Laboratory has shown that this drift can lead to deterioration of 10 to 30 percent of whatever efficiency gains a project initially yielded within one to two years.

The first step to combating drift is knowing it’s happening, Ramadori said — and that requires sensor data. Low-cost sensors are becoming increasingly popular among major energy services companies like Schneider, Siemens, Honeywell and Johnson Controls, as well as building-efficiency startups like Redaptive and Carbon Lighthouse.

But knowing what to do with all that sensor data — and how to cross-reference it with all the other data points that determine a building’s energy use — is beyond the grasp of most facility managers, he said.

“The weather outside is constantly changing, and the electricity prices are changing throughout the day, the carbon-intensity is changing, the occupancy is changing,” he said. ​“But today’s systems only know what the thermostat on the wall is telling them.”

BrainBox AI’s system combines 5-minute-interval building management system data and sensor data with ​“as much outside information as we can get” to better inform room-by-room ​“deep-learning models” of how buildings use energy, he said. ​“Once it’s done learning, it can forecast what’s going to happen in each of those rooms over the next 6 hours with 99 percent accuracy.”

“That allows all kinds of optimization opportunities,” he said. But once again, facility managers can’t be expected to act from moment to moment to reset building operational controls to react to all those variables — which is where the ​“autonomous decision-making” comes in.

“Even a 30-story office building will have 300, 400, 500 pieces of equipment,” he said. ​“Autonomous AI can decide whether ​“to slow down a fan, move a damper a bit more open or closed, change a temperature setting.”

BrainBox AI is seeing reductions in energy use of roughly 15 to 25 percent from its implementations across more than 700 buildings to date, Ramadori said. Trane, a global HVAC manufacturer and energy services company, recently released a study citing the company’s performance in a project with an unnamed ​“industry-leading customer in the collection and processing of life-saving biological material,” which yielded ​“significant energy performance improvements” and carbon emissions reductions of greater than 30 percent across more than 100 facilities.

As importantly, BrainBox AI’s earliest installations indicate that its improvements are holding over time, Ramadori said. ​“When conditions change — for example, you just sold the building owner a window upgrade, and boom, the behavior changes — the AI will pick up on that behavior change, with the new windows, and start learning again.”

In April, BrainBox AI raised $20 million from investors including ABB, another major energy equipment and services provider. As part of that deal, the startup acquired ABB’s multisite retail energy management system integrator business — a potential inroad into deploying its technology in smaller buildings.

Community Energy Labs: Helping keep classrooms cool and saving cash

If you think the centralized HVAC systems in big office buildings are hard to keep running efficiently, just try optimizing the rooftop air conditioning units in an elementary school.

That’s the challenge that Tanya Barham, CEO of Portland, Oregon–based Community Energy Labs, is taking on. So far, the results indicate that her startup’s AI-based ​“model-predictive control” approach is working — and at a price that schools can afford.

Take these results from a pilot at Sonora Elementary School in Southern California involving six classrooms in two buildings, each with its own rooftop AC system controlled by a wirelessly connected smart thermostat. It’s one of a growing number of pilot installations that Community Energy Labs (CEL) is undertaking in California, Oregon and Washington state.

Instead of tasking a facility manager with ​“futzing with thermostats” via a building management system that could cost millions of dollars for a school to deploy, CEL installed its software in one day, then allowed it to learn the relationship between temperatures in each classroom and the minute-by-minute demands placed on each rooftop AC unit over about two weeks.

At that point, ​“we give it a test,” Barham said. ​“We say, ​‘OK, for the next week, please tell me exactly what the indoor air temperature will be in every single room in this building, and how much energy it will use.’ And then we look at what actually happened. And if its predictions were correct 90 percent of the time, we let it control the building.”

The results for Sonora Elementary were significant reductions in energy use and demand charges — the all-important extra charges that bigger buildings incur when their electricity use spikes during any 15-minute period during a month of utility billing. CEL’s software accomplished that by avoiding having too many AC units running at once.

CEL’s system cut 24 percent of total peak power and 30 percent of HVAC peak power. It also shifted 16% of the cooling load from an on-peak price period to a low price period. Those savings yielded a payback on the cost of installation in less than two months, according to pilot data.

But more importantly, they did so without triggering any complaints from teachers or students that classrooms were too hot — the key measure of success for any energy-focused HVAC control platform, she said.

“If you’re not doing both comfort and energy cost, someone’s always losing,” she said. But for traditional building-automation systems, keeping track of both of those variables on the fly is ​“a constant game of whack-a-mole…no human being can do this.”

Giving schools a low-cost way to control HVAC energy costs, which typically make up nearly half of their total energy consumption, has become even more important as schools seek to reduce their carbon impact, she said. In California, utility tariffs that charge far more during hours when the state’s grid is struggling to meet peak air-conditioning load are driving up school electricity bills, she noted.

“At the same time, they’re being told they need to cut emissions,” she said. Software that can effectively reduce air-conditioning demand during those peak times could allow schools to play a much more active role in that grid-balancing task — and in California, where fossil-gas-fired power plants are the primary source of peak grid power, that’s a significant carbon-cutting benefit as well.

VadiMAP: Optimizing efficiency and on-site energy across thousands of buildings

It’s hard to optimize the HVAC system of one building. What’s even harder is figuring out how to do that for hundreds or thousands of buildings across different climate zones and utility service territories — and adding rooftop solar and batteries to the mix.

But that’s the challenge vadiMAP is taking on. Since its 2017 founding, the Montreal, Canada–based startup has built a business on optimizing these variables for properties of companies including McDonald’s, Caterpillar and NAPA Auto Parts’ parent company UAP. At its heart is a subscription-based service powered by its AI-enabled NanoGrid Simulation Engine.

VadiMAP starts by asking a building manager to provide a relatively sparse set of data via an online questionnaire — ​”You can take a few pictures here and there” of the equipment within a building, along with past utility bills, said Dan Boucher, the company’s founder and chairman.

It’s the data that vadiMAP is collecting behind the scenes ​“that is the secret sauce,” he said. ​“We already have the market data, the weather data, the efficiency incentives, the utility tariffs” and other key influences on how much energy different buildings need and how much they’ll pay for it.

Then, vadiMAP asks building owners to rank the importance of achieving three primary outcomes: reducing energy costs, reducing carbon footprint and providing ​“autonomy and resiliency.” That last part is where the concept of a ​“nanogrid” — a building that can provide its own power during grid outages or supply excess power to the grid from a combination of backup batteries and on-site solar panels or other generating resources — comes in.

A growing number of commercial customers are looking for ways to reduce their exposure to increasingly frequent and long-lasting power outages, which have been blamed for economic losses that range from the hundreds of millions to billions of dollars from business closures, food spoiled due to lack of refrigeration and other impacts. But backup generators and batteries are expensive, and property owners need data to inform their decision on whether they’re worth the cost, Boucher said.

Also, solar panels and batteries aren’t just useful for backup power. They can store up self-generated energy when utility rates are cheap for use when rates are expensive. Those price spikes may come during summer afternoons and evenings in California, or summer midday hours in Canada’s Ontario province, and they can change from season to season. What’s more, commercial customers have an increasing choice of a variety of rates, all of which need to be plugged into a property owner’s calculations of costs and benefits.

Calculating all of these variables to come up with the most cost-effective option for a single building is complicated but doable with traditional software tools, he said. But those software platforms ​“are not necessarily running thousands of iterations” to match the number of properties vadiMAP is analyzing. At that scale, ​“we need advanced algorithms to determine the optimal configuration.”

Correction: This article originally misspelled Tanya Barham’s surname. We regret the error.