Energy Modeling in Design: What It Predicts and What It Misses

Somebody has handed you an energy model. It is a PDF with a bar chart, a percentage saving against a baseline you have never seen, and an energy use intensity figure carried to one decimal place. You are being asked to make a decision on it, or to sign something because of it, and nobody in the room has told you how much of that number to believe.

The honest answer is that the number is real, the decimal place is not, and the model is far more useful than the single figure printed on the cover suggests. A building energy model is a comparison engine. It is very good at telling you that option B is fourteen percent better than option A, and considerably worse at telling you what either one will cost to run in 2029. Understanding the difference is the whole of the skill.

This article explains what a model contains, what it can genuinely resolve, what it structurally cannot, and how to commission one that earns its fee. It is written for owners, developers and project managers rather than for modellers. Budlong builds these models as part of MEP engineering services on California projects, and much of what follows comes from comparing predictions against metered data after occupancy. You can see the range of that work on our expertise overview.

1. What Is Building Energy Modeling?

Building energy modeling is the practice of representing a building and its systems in software precisely enough to simulate their energy behaviour over a full year. The result is a whole building energy model: a digital description of geometry, construction assemblies, glazing, internal loads, occupancy patterns, HVAC systems, service water heating, lighting, controls and, increasingly, on site generation and storage.

It exists because the alternative is worse. Without a model, the only way to compare a curtain wall against punched windows, or a variable refrigerant flow system against a four pipe fan coil scheme, is judgement and rules of thumb. Those work adequately for conventional buildings in familiar climates and fail badly the moment a project has an unusual shape, an ambitious facade, a high internal load or a target that sits below code.

Two things a model is not. It is not a load calculation. Design load calculations size equipment against extreme percentile weather conditions and are the basis of every duct, pipe and unit on the drawings. The energy model runs typical weather and is the wrong instrument for sizing. The relationship between the two is explained in more depth in our guide to HVAC load calculation methods. And it is not a forecast. It is a controlled experiment in which everything except the variable under test is held fixed.

2. How a Whole Building Energy Model Actually Works

The mechanism matters, because most misunderstandings about model accuracy come from not knowing which layer produced the number.

The load calculation layer

For each hour of the year, and often at shorter internal time steps, the engine solves a heat balance for every thermal zone. It accounts for conduction through walls, roof and glazing, solar gain through fenestration adjusted for shading and orientation, thermal mass storing and releasing heat with a time lag, heat from people, lights and equipment, and air exchange through infiltration and ventilation. The output of this layer is a heating or cooling load for each zone in each hour.

The systems layer

Those zone loads then meet the air side systems. The engine works out supply air temperatures and flow rates, economizer position, reheat, fan energy at the resulting airflow, and how the control sequences respond. This is the layer where HVAC system selection shows up in the results, and it is the layer most sensitive to how carefully the control sequences were entered.

The plant layer

The systems layer produces a heating and cooling demand on the central plant. Chillers, boilers, heat pumps, cooling towers and pumps are then simulated against performance curves that describe their efficiency at varying part load ratios and entering conditions. Fuel and electricity consumption falls out here.

The weather file

Underneath all three layers sits a weather file containing 8,760 hourly records of dry bulb temperature, humidity, solar radiation, wind speed and direction. For voluntary analysis this is normally a typical meteorological year assembled from decades of observations. For California compliance work it is the fixed weather file that the Energy Commission publishes for each of the state’s sixteen climate zones.

3. Compliance Model, Design Model, Calibrated Model

Most disputes about energy models are actually disputes about which of three quite different deliverables was ordered. They share software and geometry and almost nothing else.

The compliance model

A compliance model exists to prove that a design meets a code or standard. Its defining feature is that many of its inputs are not yours to choose. Occupancy schedules, plug load densities, thermostat setpoints, hot water draw profiles and operating hours are fixed by the rule set and applied identically to the proposed design and to the baseline. That is deliberate. It means the only difference between the two runs is the regulated features, which is exactly what the code wants to test. It also means the annual energy figure the model prints is not a prediction of your building’s consumption. It is a score in a controlled comparison.

The design model

A design model exists to inform decisions. Inputs are chosen to reflect how the building will actually be used: the real tenancy hours, the real equipment density, the real setpoint policy, the real number of people. Nothing is fixed by a rule set, so the model can answer questions the compliance model cannot, such as whether a shading strategy pays for itself or how much a twenty four hour tenant changes the plant selection. Its results are more relevant and less defensible, because the assumptions are now arguable.

The calibrated operational model

A calibrated model is a design model that has been tuned until its output matches metered reality. This is the only kind of model that has demonstrated accuracy rather than claimed it. ASHRAE Guideline 14 sets the widely used acceptance thresholds and requires at least twelve months of measured utility data. It is used for retrofit analysis, measurement and verification of savings, and for buildings running against an ongoing performance target.

The failure we see most often is a compliance model asked to do a design model’s job. An owner is shown a Title 24 performance run, reads the annual energy figure as an operating budget, and is then surprised eighteen months later. The model was not wrong. It was answering a different question.

4. The Four Reasons a Model Gets Commissioned

Almost every energy model on a California project is being paid for by one of four motives, and knowing which one applies tells you how the model will be built.

Code compliance

The Title 24 Part 6 performance path requires a model in approved software. This is the most common trigger and the one with the least flexibility. See our Title 24 compliance guide for how the documentation package assembles.

Green rating certification

LEED energy performance credits use the ASHRAE Standard 90.1 Appendix G performance rating method, which has its own baseline construction rules and its own reporting requirements. The rating body, the U.S. Green Building Council, is asking a different question from the Energy Commission and will not accept a Title 24 run as a substitute. Our LEED MEP services plan both submissions from one geometry rather than two.

Design decision support

This is where the model earns its keep and where it is most often skipped for fee reasons. Testing orientation, glazing ratio, shading, envelope performance, system topology and control strategy while those decisions are still cheap to change is the highest return use of simulation. It is central to any credible net zero building design, because a net zero target is a load reduction problem before it is a photovoltaic design problem.

Utility incentive applications

California whole building incentive programmes pay against modelled performance above the code baseline, with the incentive scaled to the savings claimed. That makes the model a financial instrument, which changes the review it receives. Programme reviewers scrutinise assumptions in a way plan checkers rarely do, and an aggressive schedule assumption that would sail through a permit set will be challenged when it is attached to a cheque.

5. What the Model Asks of Each Building System

A model is only as good as the description it is given. Each discipline owes it specific information, and vague answers become invented assumptions.

Envelope

Assembly U factors, glazing U factor and solar heat gain coefficient, visible transmittance, frame and spandrel treatment, thermal bridging, external shading geometry and an infiltration rate. Envelope inputs dominate the results in perimeter dominated buildings and are usually settled before anyone thinks to model them.

Mechanical

System topology, zoning, supply air temperature and reset strategy, minimum outdoor air rates, economizer type, fan and pump power, heat recovery, plant equipment with part load curves, and the control sequences in enough detail to be coded. Our HVAC design services team writes sequences with the model in mind, because a sequence that cannot be described cannot be simulated.

Electrical and lighting

Lighting power density by space type, daylight and occupancy control strategies, receptacle load density, elevator and vertical transport loads, and any process or specialty power. Coordination with electrical engineering services matters because the connected load on the panel schedule and the diversified load in the model are different numbers, and confusing them inflates results substantially.

Plumbing and service water heating

Hot water demand profile, storage volume, distribution losses, recirculation control and equipment efficiency. Domestic hot water is a small line in an office model and a very large one in a hotel or a hospital, which is why the plumbing engineering input needs to arrive early on those building types.

Lighting design

Where a project has a genuine architectural lighting design concept, the model needs the actual design rather than the allowance. Decorative and accent lighting is frequently omitted from early models and then appears in the metered data as an unexplained baseload.

6. Title 24 Part 6, CEC Software and Long Term System Cost

California places more weight on the energy model than any other jurisdiction in the country, so the California specifics deserve their own treatment.

The performance path

Part 6 offers a prescriptive checklist and a performance path. The performance path builds a standard design of the same geometry, orientation and occupancy as the proposed design but with every regulated feature set at the prescriptive minimum, then compares the two. If the proposed design scores at or better than the standard design, it complies. This is the route that any architecturally ambitious project takes, because it is the only route that permits trade offs between systems.

Approved compliance software

Compliance runs must be produced in software approved by the California Energy Commission for the applicable code cycle. For permit applications submitted on or after 1 January 2026, the approved 2025 compliance software list comprises the public domain CBECC 2025 releases for all occupancies, plus approved third party programs including EnergyPro 10, IES Virtual Environment for Title 24 2025 for nonresidential new construction, and Right-Energy Title 24 for single family new construction. A general purpose engine used for design analysis is not a compliance submission no matter how sophisticated it is.

The rule set behind the software

What the software does with your inputs is defined by the Alternative Calculation Method Reference Manual. The 2025 nonresidential and multifamily ACM manual establishes how the proposed design is described, how the standard design is generated, and which inputs are fixed rather than user selected. Reading even the schedules appendix is a fast cure for the belief that a compliance model predicts a utility bill.

The Long Term System Cost shift

For two decades California scored compliance on Time Dependent Valuation, which weighted saved energy by its value to the grid in the hour it was saved. The 2025 code replaced that metric with Long Term System Cost, the Energy Commission’s projection of the cost a design imposes on California’s energy system over a thirty year horizon.

7. Healthcare, Laboratories, Data Centres and Schools

Four building types stress the standard modelling workflow in ways worth flagging before a fee is agreed.

Healthcare

Hospitals carry ventilation and pressure relationship requirements that exist for infection control rather than energy, and those requirements are not negotiable inputs. Air change rates, filtration and pressure cascades are fixed by clinical standards, which removes many of the levers a model would ordinarily pull and concentrates the remaining opportunity in plant efficiency, heat recovery and control. Our healthcare facilities work routinely involves showing an owner that the modelled savings ceiling on a hospital is genuinely lower than on an office of the same size, and why that is not a failure of ambition.

Laboratories

The 2025 Energy Code removed the laboratory exemption that earlier cycles carried, which pulls high exhaust, once through air systems into scope. Laboratory models are dominated by ventilation rate and by process load, both of which are the least reliably known inputs at design stage. Small errors in assumed fume hood diversity produce very large errors in annual energy. For laboratory and industrial projects, parametric runs across a range of hood usage are worth more than a single precise looking answer.

Data centres

In a data centre the IT load is the building, and the IT load is set by a tenant who may not have signed yet. Models built on assumed rack density and assumed ramp rate are exercises in scenario planning rather than prediction. What the model does well here is compare cooling topologies and quantify the effect of higher supply temperatures, both of which hold up regardless of the eventual load. See our mission critical capability for context.

Schools and universities

Education buildings have the most predictable occupancy schedules of any type and the least predictable summer operation. Modelled savings on school projects tend to hold up better than average, provided the summer shutdown assumption matches district practice rather than the default. Our education sector work usually starts by confirming that one input.

8. What the Model Predicts Well and What It Misses

This is the section that matters. A model has a well defined competence, and its incompetence is equally well defined. Both are structural, which means neither can be fixed by hiring a better modeller.

What it predicts well

Everything the model does well shares one property: the answer is a comparison between two runs that share their errors. If both runs assume the same wrong occupancy schedule, the difference between them is still informative. This is why comparative results survive assumptions that would destroy an absolute prediction.

What it misses, and why

The gap between modelled and measured energy use has been studied for two decades and its causes are not mysterious. Reviews of certified buildings have found that median predicted energy use can align closely with median measured energy use while individual buildings scatter widely on both sides, and the U.S. Department of Energy has noted that even the calculation engines themselves can disagree by twenty five to thirty percent on well defined test cases before any input error is introduced. The following causes account for most of the remaining spread.

Standardised occupancy and plug load schedules

Compliance rule sets fix occupancy density, equipment power density and hourly schedules so that proposed and baseline runs stay comparable. No real building follows those curves. A 2,000 hour per year assumption meets a tenant who works twelve hour days, or a floor that sits half let for three years.

Weather that represents a typical year, not your year

Discussed above and worth repeating because it is the most consistently underrated input. The NIST work found peak HVAC demand understated by up to sixty nine percent in a bad year against a typical file. Anything sized or valued on peak demand inherits that exposure.

Idealised control assumptions

The model assumes the economizer modulates correctly, the supply air temperature reset works, the optimum start routine engages and the setback holds every night. In the field, sequences get overridden in the first cold week and never restored. This single category probably accounts for more lost savings than any other, and it is the reason commissioning services and energy modelling belong in the same conversation.

Unmodelled plug and process loads

Miscellaneous electric loads have grown from a rounding error into a major end use. Lawrence Berkeley National Laboratory puts miscellaneous and electronic loads at roughly one third of end use electricity in homes and commercial buildings, and identifies them as the fastest growing category. Models routinely carry a receptacle density set years before anyone knows what equipment will arrive.

Tenant behaviour

Lease terms, fit out standards, after hours operation, personal heaters, propped doors and thermostat wars are all outside the model and all inside the meter. A landlord and tenant matrix that permits unrestricted after hours HVAC will beat any envelope improvement the model was used to justify.

Part load performance and degradation

Plant is simulated against manufacturer performance curves generated in a laboratory on new equipment. Real chillers foul, real heat pumps lose charge, real filters load up and real dampers stick. Nothing in a design model represents the third year of a maintenance contract.

Infiltration assumptions

Infiltration is normally entered as a fixed rate per unit of envelope area, taken from a table rather than a test. Commercial envelopes leak more than the tables suggest, leakage varies with wind and stack effect, and few nonresidential buildings in California are pressure tested to confirm the figure that was modelled.

The complete absence of faults

This is the structural one. A model contains no stuck damper, no failed sensor, no simultaneous heating and cooling, no valve leaking by. Research led by Lawrence Berkeley National Laboratory for the U.S. Department of Energy estimates that faults in U.S. commercial buildings waste roughly 0.7 quadrillion British thermal units each year, worth close to fourteen billion dollars. A perfect model of an imperfect building will always predict low.

Calibration is only possible after the fact

Every other input problem could in principle be corrected by tuning against measured data. That option does not exist during design, because the measurements do not exist. Under ASHRAE Guideline 14 you need at least twelve months of metered data before a model can be called calibrated. The design model is therefore uncalibrated by definition, and describing it as validated is a category error.

9. Commissioning a Model That Is Actually Useful

None of the above is an argument against modelling. It is an argument for commissioning models deliberately. Four habits separate a model that changes decisions from a model that decorates a submission.

Define the questions before the geometry

Write down, before any modelling starts, the decisions the model is meant to inform and the date each decision is made. If a facade question is settled in schematic design, a model delivered in design development cannot inform it. ANSI/ASHRAE Standard 209 formalises this by defining a sequence of modelling cycles tied to design stages, from simple box studies at concept through load reduction, system selection, change order review and as built verification. The ASHRAE standards and guidelines library is the reference point for both Standard 209 and Guideline 14.

Insist on a documented assumptions register

Every model should be delivered with a register listing each significant assumption, its value, its source and its confidence. Schedules, internal load densities, setpoints, infiltration, plant curves and control sequences at minimum. Assumptions that come from the rule set should say so. Assumptions that come from the modeller’s judgement should say that too, because those are the ones the owner is entitled to challenge.

Ask for parametric runs, not a single answer

A model that produces one number invites false precision. A model that produces a ranked set of options with a sensitivity range around each invites a decision. Ask for the range, ask which variable drove it, and ask what would have to be true for the ranking to flip. Buildings with genuine performance targets, including anything pursuing net zero or a demanding sustainable design brief, need this more than conventional projects, not less.

Pair the model with measurement after occupancy

The model’s prediction becomes testable the moment the building is metered. Submetering by end use, a measurement and verification plan written during design rather than after handover, and a commitment to compare the first full year of data against the model are what convert a one off document into an asset. This is also where modelling and the building commissioning process reinforce each other: commissioning verifies that the sequences the model assumed are actually running, and metering verifies that the savings they were supposed to deliver actually arrived. Ongoing metric driven operation, covered in our piece on operational sustainability metrics, is the natural continuation. Benchmarking data published by the U.S. Energy Information Administration gives a useful sanity check on whether a measured result is unusual for the building type.

Key Takeaways

• A whole building energy model is an hourly simulation of loads, systems and plant against a weather file. It produces 8,760 results, and the annual total is the least informative of them.
• A compliance model uses fixed, rule set assumptions and answers whether the design beats a baseline. It is not an operating cost estimate and should never be used as one.
• Comparative results are reliable because errors are shared between runs. Absolute predictions are not, because the assumptions behind them will not hold.
• The documented causes of the performance gap are standardised schedules, typical year weather, idealised controls, unmodelled plug loads, tenant behaviour, part load degradation, infiltration assumptions and the absence of faults.
• Faults alone are estimated to waste around 0.7 quadrillion British thermal units a year across U.S. commercial buildings, and no design model represents a single one of them.
• Calibration to ASHRAE Guideline 14 criteria requires at least twelve months of metered data, so no design model is calibrated by definition.
• Under the 2025 California Energy Code the compliance metric is Long Term System Cost, which changes what the model optimises for compared with Time Dependent Valuation.
• Commission models by defining the questions first, demanding an assumptions register, requiring parametric runs, and planning measurement and verification before occupancy.

10. Where Budlong Works

Modelling practice in California is shaped by local factors as much as by state code. Climate zone assignment changes which measures pay their way, reach codes in several cities add requirements above the state baseline, and utility incentive programmes differ between service territories in both baseline and verification approach. Budlong maintains teams across the state so that the analysis reflects the jurisdiction the project sits in. You can read more about how the practice is structured on our company story page.

Sectors where modelling assumptions matter most

Healthcare has the least modelling freedom and the most clinical constraint. Laboratories are dominated by ventilation and process assumptions that nobody can confirm at design stage. Mission critical projects are modelling a tenant load that may not be defined. Commercial buildings carry the widest gap between assumed and actual plug loads.

11. Related Reading

12. Frequently Asked Questions