Ask a design team how much energy their new office building will use, and they’ll show you a model. Ask the facilities manager a year after opening, and they’ll show you a utility bill. Those two numbers often disagree, sometimes by a wide margin. In the UK this is widely called the performance gap. In the US, the same problem shows up when buildings that were modelled as high performers land in the middle of the pack once their energy use is benchmarked.
Closing it means connecting models to real data. Combining meter readings with energy models produces tools that can actually predict savings, guide retrofits, and stand up to scrutiny from investors, regulators, and auditors.
What energy models are used for
Building energy models simulate how a building uses energy hour by hour across a year, based on its geometry, construction, systems, occupancy, and weather. The US Department of Energy describes building energy modelling as a multipurpose tool used for designing new buildings and retrofits, checking code compliance, qualifying for tax credits and utility incentives, supporting real-time control, and informing energy codes and policy.
Most of those uses involve predicting the future. That’s why the gap between predicted and actual energy use matters so much. A retrofit justified by modelled savings that never appear damages trust in the whole process.
Why models and meters disagree
The reasons are usually mundane. Design models often assume standard occupancy schedules, but real buildings run longer hours, host evening events, or leave equipment on overnight. Plug loads from computers, servers, kitchens, and lifts are hard to predict and often underestimated. Systems may not be commissioned properly, so controls don’t work as intended. Construction quality may fall short of the design, with air leaks or thermal bridges the model doesn’t capture. And the weather in any given year differs from the typical weather files models use.
Often the design model was built to show code compliance, which is a different question from predicting real operation.
Calibration: tuning the model to the meter
For existing buildings, the most reliable way to close the gap is calibration. The modeller builds a model of the building as it actually operates, then adjusts uncertain inputs, such as schedules, plug loads, infiltration, and set points, until the model’s results match measured energy data.
In the US, ASHRAE Guideline 14 is the standard reference for how close is close enough. It sets statistical limits for calibrated models, commonly quoted as a coefficient of variation of the root mean square error (CV(RMSE)) within 15% and a normalised mean bias error (NMBE) within 5% when using monthly data, with looser limits of 30% and 10% for hourly data. In the UK, CIBSE TM54 gives guidance on estimating operational energy use at the design stage, encouraging design teams to model realistic operation rather than just compliance scenarios.
Calibration needs good data. Monthly utility bills are the minimum. Interval data from smart meters, typically every 15 or 30 minutes, is far better because it shows how energy use changes between day and night, weekdays and weekends. Sub-meters for lighting, heating, cooling, and plug loads, plus trend data from the building management system, make it easier to pin down where the model is wrong.
Weather data from the actual period being modelled, rather than a typical year, is essential for a fair comparison.
What a calibrated model makes possible
Once a model matches reality, it becomes much more useful.
Retrofit planning improves, because the savings from measures such as new heat pumps, better controls, or insulation are estimated from a realistic starting point. Measurement and verification gets easier, since the calibrated model can serve as a baseline to show how much energy a project actually saved, adjusted for weather and occupancy changes. Many energy performance contracts rely on this approach.
Compliance planning becomes more reliable too. Building performance standards are spreading in the US, with cities such as New York setting emissions limits for large buildings under Local Law 97, and benchmarking requirements in many other cities and states. In the UK, public buildings have long displayed operational energy ratings through Display Energy Certificates. Owners facing these rules need to know how future changes will affect measured performance, and a calibrated model can show them.
Data-driven approaches alongside physics models
Not every question needs a full physics-based simulation. Statistical and machine learning models trained on meter data can forecast energy use, detect faults, and flag unusual consumption quickly and cheaply. Simple regression models relating energy use to outdoor temperature can reveal whether heating and cooling are behaving sensibly.
These approaches work well for monitoring and fault detection. Physics-based models are still better for testing changes the building has never experienced, such as a new HVAC system or a different envelope. Many teams use both: data-driven models to watch the building day to day, and calibrated simulations to plan major changes.
Getting started with your own buildings
Owners and managers can take a few practical steps. Make sure interval meter data is collected and stored, ideally with sub-metering of major systems. Keep records of occupancy, operating hours, and system changes, because the model will need them. Ask design teams to model realistic operation as well as compliance. And for major retrofits, ask whether savings estimates come from a model calibrated to the building’s real data.
Models built on real data are easier to trust, and decisions based on them are far less likely to disappoint.
