Modern buildings already collect extensive data. This article argues for using existing operational data to compare design intent, commissioning settings and real behaviour, helping owners detect misconfigurations, avoid new data silos and improve performance before major optimisation or renovation.
The building sector has invested heavily in smart technologies, sensors, automation, and digital platforms. This development has created new possibilities for energy efficiency, comfort, and operational control. Yet many modern buildings already collect large amounts of useful data through their building automation systems. A central challenge now lies in turning this existing data into evidence: evidence that planned functions have been implemented correctly, that systems operate as intended, and that deviations are detected early during the first months and years of operation.
This article argues for a shift from technology deployment towards data-informed verification. Before adding further devices or starting major optimisation projects, owners and operators can gain substantial value by analysing the data already available in building automation systems. The key task is to compare design intent, functional descriptions, commissioning settings, and real operation over time.
Discussions about smart buildings often focus on algorithmic innovation, artificial intelligence (AI), new sensors and internet-connected devices. These technologies matter. They can support better forecasting, predictive control and fault detection. At the same time, many larger buildings already operate with extensive building automation infrastructure. They already record temperatures, setpoints, operating modes, schedules, actuator positions, alarms, pump states, valve positions, air-handling unit states, and energy-related measurements.
This data is valuable because it captures how the building behaves as a system. It shows how control logic, equipment, hydraulic systems, schedules and user needs interact in practice. In many cases, the first step towards better energy performance is therefore to understand whether the installed systems are configured and operating as originally intended.
The performance gap between predicted and actual building operation has been studied for many years. Causes can occur across design, construction and operation phases, and they often interact with one another. Commissioning - and monitoring- based approaches have repeatedly been identified as important mechanisms for reducing this gap and maintaining performance over time [1–3].
A practical implication follows: building owners should treat automation data as an operational evidence layer. This evidence layer can help answer a simple but demanding question: Does the building do what the project team expected it to do?
Modern building projects involve many actors: owners, architects, engineering consultants, HVAC planners, electrical planners, building automation specialists, system integrators, contractors, commissioning teams, facility managers and future users. Each actor contributes to the final technical outcome. Each also works within specific contractual, time and budget constraints.
This creates a structural dilemma. The functional description may define one intended behaviour. The installed equipment may support another set of possibilities. The automation contractor may translate the intended logic into programmable sequences under time pressure. The commissioning team may check defined functions during a limited handover window. The facility management team then inherits a complex system and must operate it under real weather, occupancy and comfort conditions.
Later users of the building experience comfort, air quality, reliability and energy cost. Their needs unfold over time. However, the contractual intensity of the construction phase is highest before long-term operational behaviour is fully visible. This timing mismatch weakens the incentive to verify building automation performance during the period when the system first meets real use.
A manual acceptance procedure remains important. It confirms that systems are installed, connected and functionally tested. Yet the behaviour of a complex building can only be fully understood across operating conditions. Heating season, cooling season, occupancy patterns, holidays, partial loads, simultaneous demands and unexpected interactions all reveal information that a single acceptance moment can only partially capture.
Data-informed commissioning extends this process. It turns the first months and years of operation into a structured verification phase. The aim is to confirm correct behaviour, identify deviations, and document corrective actions with traceable evidence.
The value of operational building data depends on access, context and continuity. A building automation system may contain the required data, while the owner may still be unable to use it effectively. Data may sit inside proprietary platforms, vendor-specific databases, local trend logs or isolated dashboards. Naming conventions may vary across systems. Equipment metadata may be incomplete. Trend intervals may be inconsistent. Functional descriptions may be stored separately from the automation data.
This leads to fragmentation: a building can be technically connected while analytically disconnected.
The alternative is to plan the data architecture from the beginning of the project. Data access should be treated as part of the building’s long-term operational capability. Owners and project teams can define which data points will be available, how they will be named, which metadata will describe them, how trend data will be exported, which interfaces will be maintained, and how future analytics can be performed without creating a new silo.
This also matters for digital sovereignty and competition. When operational insight depends on a single vendor-specific environment, owners can lose flexibility. When data structures are open, documented and exportable, owners can work with different service providers, auditors, analytics teams and facility managers over the building lifecycle.
An instructive example is Google’s Digital Buildings project. The publicly accessible project describes an open-source, Apache-licensed ontology and toolset for representing structured information about buildings and building-installed equipment. Google states that a version of the ontology and toolset is used internally to manage buildings in its own portfolio [4].
The Digital Buildings ontology defines semantic data primitives and concrete constructions to model physical spaces and equipment. It also provides configuration approaches for mapping concrete building assets to the abstract model [4]. The wider research literature on building ontologies shows that semantic models such as Brick Schema, Project Haystack, RealEstateCore and Digital Buildings address a central problem: data-driven analytics require a shared understanding of what data points mean, where they belong, and how they relate to equipment and spaces [3,5].
The important lesson is architectural. Advanced analytics starts long before the algorithm. It starts with consistent data structures, well-described equipment, clear relationships, and a disciplined approach to modelling building systems. This is closely aligned with good Building Information Modelling practice: the digital representation of the building should support design, construction, commissioning and operation.
For building owners, the implication is practical. A data model should help answer operational questions. Which air handling unit serves which zones? Which valve belongs to which heating group? Which setpoint governs which loop? Which sensor reading is used for which control action? Which schedule drives which operating mode? Which data points confirm that a function behaves as specified?
When these relationships are explicit, automated analysis becomes more reliable. When they are implicit, every diagnostic task begins with reconstruction work.
The industry can move from manual acceptance towards continued confirmation of building behaviour. This does not replace professional judgement. It strengthens it with time-series evidence.
A useful workflow can be structured in six steps.
This workflow creates a feedback loop. It allows the building team to check whether corrective measures improve actual behaviour. It also supports collaboration between owners, planners, integrators and operators because discussions can be grounded in shared evidence.
Data science in buildings should not be reduced to advanced machine learning. In many cases, the most useful work is disciplined engineering analytics: cleaning time-series data, aligning data sources, creating meaningful features, comparing expected and observed behaviour, detecting patterns, quantifying deviations and presenting results in a form that technical teams can act upon.
Modern fault detection and diagnostics research shows increasing interest in data-driven approaches, supported by the wider adoption of building automation systems and advances in sensing and machine learning [6]. However, field value often begins with more basic questions. Are schedules correct? Are setpoints plausible? Are pumps and valves behaving consistently? Are heating and cooling functions active at the same time? Are controllers stable? Are systems responding to outdoor temperature and occupancy as expected?
These questions can often be answered with existing data. They require access, context, domain knowledge and reproducible analysis. They also require a culture in which commissioning continues beyond handover.
The building sector faces substantial pressure to reduce energy use and improve indoor environmental quality. Globally, regulators increasingly recognise the role of technical building systems, automation and monitoring in achieving these goals, yet Europe has a leading role with the Energy Performance of Buildings Directive (EPBD). Renovation and optimisation will remain essential. Yet optimisation should start from a verified baseline.
A building is a complex system. Its performance emerges from many interacting components and decisions. Operational data can make these interactions visible. Used well, it allows owners and operators to move from assumptions to evidence, from handover to lifecycle verification, and from isolated interventions to continuous technical learning.
Before replacing equipment, adding sensors or deploying new control algorithms, owners can often gain value by analysing whether existing systems are configured as planned. Correct settings, coherent schedules, reliable trend data, clear naming, accessible interfaces and documented control sequences are foundational. They make later optimisation cheaper, more transparent and more durable.
The next step in smart buildings may therefore be less spectacular than many technology narratives suggest, but more immediately useful: use the data already collected, compare plan and implementation, correct misconfigurations, and establish a continued check of building behaviour over time.
[1] Mills, E. (2011). Building commissioning: a golden opportunity for reducing energy use and greenhouse gas emissions in the United States. Energy Efficiency, 4, 145–173.
https://link.springer.com/article/10.1007/s12053-011-9116-8
[2] Crowe, E., Mills, E., Poeling, T., Curtin, C., Bjørnskov, D., Fischer, L., Granderson, J., & Mathew, P. (2020). Building commissioning costs and savings across three decades and 1,500 North American buildings. Energy and Buildings, 227, 110408.
https://www.sciencedirect.com/science/article/abs/pii/S0378778820309961
[3] Jradi, M. (2021). Automated auditing and continuous commissioning of next generation building management systems. Energy Informatics, 4, 1.
https://energyinformatics.springeropen.com/articles/10.1186/s42162-020-00136-2
[4] Google. Digital Buildings Project. GitHub repository.
https://github.com/google/digitalbuildings
[5] Fierro, G., Koh, J., Agarwal, Y., Gupta, R. K., & Culler, D. E. (2019). Beyond a house of sticks: Formalizing metadata tags with Brick. BuildSys ’19.
https://dl.acm.org/doi/10.1145/3360322.3360842
[6] Chen, Z., O’Neill, Z., Wen, J., Pradhan, O., Yang, T., Lu, X., Lin, G., Miyata, S., Lee, S. H., Shen, C., Li, P., Li, Y., Wang, T., & Hu, M. (2023). A review of data-driven fault detection and diagnostics for building HVAC systems. Applied Energy, 339, 121030.
https://www.sciencedirect.com/science/article/abs/pii/S0306261923003948
