How buildings are using AI and data to evolve from static assets to active participants in the energy ecosystem

A new era is unfolding for the built environment, driven by the twin imperatives of smarter building systems and the need to respond dynamically to an increasingly renewable grid. Around the world, regulatory frameworks and rating tools like Green Star are encouraging buildings to shift from passive structures to active participants in a digital energy ecosystem. In a future of demand response and grid flexibility, innovations in data and machine learning are starting to shape how buildings track, estimate, and ultimately are rewarded for their measurable contributions to the energy transition. As energy grids become more volatile and carbon-conscious, buildings need to evolve to respond to dynamic real-time carbon and energy price signals.  A well-controlled, data-informed and tuned building will save both carbon and money.

State of practice: Data collection, integration and ontologies

The foundation for this transformation is robust data. Across the industry, buildings are now collecting data; however, the structure and quality are variable. The industry is starting the process guided by formal ontologies and frameworks that define concepts, relationships, and standards. Addressing the persistent challenge of poorly structured or incomplete datasets is a priority. By developing shared ontologies and open standards, stakeholders are overcoming data silos and enabling interoperability, ensuring the right data is available for both defect detection and performance monitoring. A European framework for classifying smart steps in buildings is under development. The first principle in this framework is to enable this shift toward high-quality, structured and validated data.

Model explainability, transferability and transparency

As AI tools are increasingly embedded in building systems, the focus is on model explainability and transferability. The aim: to dispel “black box” fears and enable engineers, designers, and clients to understand and trust the algorithms shaping critical infrastructure. Data insights are twinned with clear communication, rooted in outcomes, agency, and adaptability, which is vital as the sector moves towards AI-driven decision-making. This leads to more rigorous audits and follow-up, which is especially prevalent in growth in remote fault detection and facility management. What is harder to evaluate is when algorithms take control of the building systems or make tuning recommendations. Arup has been researching how virtual sandboxes can be used to benchmark AI-driven building logic control systems, before letting them loose on a building.

Advanced applications: Optimisation and control

While current practice largely centres on integrating data for defect management, the frontier is real-time optimisation. Industry leaders are pioneering physics-based synthetic data models to simulate, predict, and control building performance at scale. These best-in-class, adaptive systems are under development, promising responsive operations that align building activity with renewable energy supply, occupant needs, and broader sustainability goals.

Three examples of how machine learning (ML) has guided smart building practice at Arup were presented recently at the AIRAH intelligent building forum. The following examples discuss how ML processes have been utilised to create new insights and tools for existing buildings.

1.    Estimating occupancy profiles from existing BMCS environmental data (people-based efficiency metrics)

Arup is training ML models to infer occupancy patterns using data that many buildings already collect (e.g., environmental sensors such as CO₂ and other monitored signals in BMCS). This tackles the post‑COVID reality that occupancy is no longer a predictable 9-to-5 schedule. By reconstructing zone-by-zone, time-varying occupancy profiles without relying on privacy‑sensitive methods (like cameras, swipe cards, or wearables), teams can create synthetic occupancy datasets that enable new insights and controls. A key outcome is shifting building efficiency KPIs from kWh/m² to kWh/person, improving fairness when comparing buildings with different densities and helping connect energy use to how space is actually used.

2.    Physics-based synthetic data calibrated to real energy data (mini energy auditing/model tuning)

Arup is using building domain expertise to constrain uncertain model parameters (envelope performance, lighting usage, plant efficiencies, etc.), then generating many physics-based simulation runs (e.g., EnergyPlus) to produce outputs that resemble metered data. These synthetic datasets are calibrated against actual building energy/BMCS measurements, allowing the team to work backwards and tune the model so it better reflects real operational behaviour. This calibrated “digital twin”-style approach supports rapid diagnostic insight (“mini energy audits”), helping identify plausible drivers of energy use and improving confidence versus relying only on early-design assumptions that aren’t updated after handover.

3.    Estimating building thermal capacity for grid flexibility (load shifting readiness)

To support grid-interactive buildings, Arup is using measured data plus calibrated modelling to better estimate a building’s thermal dynamics and effective thermal capacity by considering how the building temperature responds over time when HVAC output changes. Characterising behaviour during scenarios like HVAC shut-off events, the approach helps predict how much load can shift, for how long, and under what weather conditions (hot/cold extremes). This turns buildings from passive energy consumers into assets that can reduce peak demand and provide flexibility, while still maintaining acceptable indoor conditions.

Challenges and outlook

Underlying all these advances is the importance of collaboration, open standards, and collective learning. The industry’s shift toward shared ontologies and data frameworks is both technical and cultural. By working together, FM and operations experts, designers, and clients are creating the strategies and tools needed to deliver meaningful change across projects and portfolios.

Despite the optimism, hurdles remain.  Legacy assets, fragmented datasets, skill gaps, and ongoing concerns around privacy and transparency in AI. The willingness to experiment, learn, and adapt suggests the sector is up to the challenge; in fact, it is imperative. Intelligent buildings are not merely about technology; they signal a fundamental rethink of how buildings serve people, communities, and the planet. The future is being shaped now, with data, AI, and human ingenuity working in concert to build smarter, carbon-efficient, more resilient cities.