AI-driven HVAC control: why reliable end-use data makes all the difference
The Foobot x Smart Impulse collaboration for building energy performance
At Foobot, we deploy AI agents capable of continuously optimising HVAC systems in commercial buildings and healthcare facilities. Our technology acts as an energy specialist who knows your building inside out. But for this AI to perform at its best, it needs one essential ingredient: reliable, granular energy consumption data broken down by end use.
This is precisely what Smart Impulse provides us. Their metering solution enables, using a smart meter and AI-based algorithms, an analysis of a building's energy consumption broken down by end use (lighting, IT, heating, etc.). This data granularity enables our AI to operate with unmatched precision.
1) The challenge of intelligent HVAC control
HVAC systems account for an average of 50% of energy consumption in commercial buildings across Europe. Facing regulatory pressures (such as the French Tertiary Decree and BACS requirements), building managers are seeking solutions to reduce this consumption without compromising occupant comfort.
The challenge? Existing Building Management Systems (BMS) are configured once, then rarely optimised. Settings drift, conditions change (weather, occupancy, usage patterns), and energy consumption creeps up silently.
Foobot addresses this with a fundamentally different approach: our AI trains on a calibrated digital twin of the building, then controls key BMS parameters in real time. It anticipates, adjusts, and automatically corrects drift.
2) Smart Impulse: the data that powers our AI
To train our AI and measure its performance, we need energy data that is accurate, continuous, and broken down by end use. Smart Impulse provides exactly that.
a) A reliable baseline for learning
Through the Smart Impulse API, we retrieve HVAC consumption data isolated from other end uses. This allows us to build a robust reference baseline for calibrating our digital twin and validating our models according to the ASHRAE G14 standard.
b) Transparent savings reporting
Once the AI is deployed, Smart Impulse data enables us to precisely quantify savings achieved on HVAC, building by building, season by season. No more ambiguity about "what was really saved": the proof is in the data.
3) Our technology: Deep Reinforcement Learning and digital twins
Unlike static rule-based systems or MPC approaches used by our competitors, Foobot uses Deep Reinforcement Learning (DRL) - the same technology DeepMind used to beat world champions at Go. Our AI trains in a virtual environment on the equivalent of over 800 years of data, developing a unique ability to adapt to unexpected conditions.
In practice, Foobot AI optimises:
- heating and cooling curves (water temperature setpoints),
- air handling unit control,
- equipment start-up and shutdown schedules,
- or other parameters affecting consumption.
Key point: Foobot does not replace the BMS. It connects as an intelligent overlay that is interoperable and fully reversible (can be switched off at the push of a button).
4) Real-world case: office buildings, measurement + AI control
A representative example of our approach involves office buildings totalling approximately 13,000 m², where Smart Impulse and Foobot were deployed together.
The process:
1. Smart Impulse installation: rapid metering plan deployment with no outage, enabling precise isolation of HVAC consumption.
2. Foobot digital twin creation: building modelling, calibration per ASHRAE G14, AI training.
3. BMS connection: AI deployment via API or dedicated gateway.
4. Continuous control: adjustments every 15 minutes, optimisation based on real conditions.
5. Consolidated reporting: end-use savings tracking powered by Smart Impulse data.
Measured results:
27% savings on HVAC compared to the 2022 reference year, representing 170 MWh saved and approximately €60,000 in gains - all achieved without compromising thermal comfort.
5) Fast payback and frugal AI
The Smart Impulse + Foobot combination delivers accelerated return on investment:
- Foobot: typical payback of 12 months or less depending on building size and configuration.
- Smart Impulse: enables action prioritisation from day one and objective savings tracking throughout the project.
Our AI also stands out for its exemplary carbon footprint: 1 tonne of CO₂ emitted for every 700 tonnes saved. Unlike energy-hungry LLMs, Deep Reinforcement Learning is a frugal, robust, and deterministic technology.
Conclusion: better measurement for smarter control
The Foobot x Smart Impulse collaboration reflects a core belief: AI-driven energy control can only excel when built on quality data. By combining Smart Impulse's end-use measurement with Foobot's continuous optimisation, we offer building managers a complete solution to:
- Understand precisely where energy goes
- Optimise HVAC continuously, with no construction work
- Prove the savings achieved
- Maintain performance over time
Article written in collaboration with the Smart Impulse team. Thanks to Sabine Dorgan for her contribution.
FAQ - Definitions
HVAC: what does it mean?
HVAC stands for Heating, Ventilation, and Air Conditioning. It refers to the systems that provide thermal comfort and indoor air quality in a building, and it is often one of the most impactful areas to optimise in operations to reduce a building's energy consumption.
Deep Reinforcement Learning: what is it?
Deep Reinforcement Learning (DRL) is a branch of artificial intelligence where an agent learns to make optimal decisions through trial and error in a simulated environment. It is the technology DeepMind used to beat world champions at Go. Unlike generative AI (LLMs), DRL is deterministic, reliable, and suited for industrial control.
Digital twin: what is it?
A digital twin is a virtual replica of the building that faithfully reproduces its thermal and energy behaviour. Calibrated according to the ASHRAE G14 standard, it enables AI training and prediction of achievable savings before any deployment.
Tertiary Decree: reminder
The Tertiary Decree (Éco Énergie Tertiaire in France) mandates progressive energy consumption reduction in commercial buildings over 1,000 m²: -40% by 2030, -50% by 2040, -60% by 2050. Building managers must report their consumption annually on the OPERAT platform.