Building Energy Models vs Actual Bills: Why the Numbers Differ

A building energy model and a utility bill measure different things. Many design and code models assume typical weather, standard schedules and default plug loads, and they compare a design against a reference building; bills record what one building actually used under real weather, real occupants, real operation and the billing calendar of your utility. To compare them fairly, ask what weather file, schedules and loads the model used, line the bills up by meter read date, and judge any comparison against the published calibration tolerances rather than a single annual total.

Evidence label: Documentation-based guide. ClearTrade Report Editorial Team checked U.S. Department of Energy Federal Energy Management Program (FEMP) guidelines, a National Renewable Energy Laboratory (NREL) data manual and presentation, U.S. EPA ENERGY STAR Portfolio Manager technical documentation, New York State’s published adoption of ASHRAE Standard 90.1 Appendix G, and an Efficiency Valuation Organization (EVO) committee page on November 3, 2026. No hands-on testing, site visit or outside professional review was performed. Tools and programs named below are documentation examples, not recommendations. This guide contains no affiliate links.

Why a model and a bill answer different questions

Many models you will see as an owner or buyer were built to show code compliance or to rate a design, not to forecast bills. The Performance Rating Method in Appendix G of ASHRAE Standard 90.1 is one example. New York State publishes the appendix as part of its 2025 energy standard, and its Informative Note says: “Neither the proposed building performance nor the baseline building performance are predictions of actual energy consumption or costs for the proposed design after construction.” It adds: “Actual experience will differ from these calculations due to variations such as occupancy, building operation and maintenance, weather, energy use not covered by this procedure, changes in energy rates between design of the building and occupancy, and the precision of the calculation tool.” Each item on that list is a reason the two numbers can disagree even when both are correct for their own purpose.

Weather: a typical year versus the year you paid for

Design models often run on a typical-year weather file. NREL’s users manual for its TMY3 data sets says a typical meteorological year is meant to “represent a year of typical climatic conditions for a location,” that “they are not suited for designing systems to meet the worst-case conditions occurring at a location,” and that “The TMY should not be used to predict weather for a particular period of time.” Your bills, by contrast, reflect the actual heat waves and cold snaps of the billing year.

FEMP’s M&V Guidelines, Version 5.0 (2024), spell out the consequence. “Calibrating a computer simulation to measured utility data necessitates that actual weather data be used.” For savings estimates, “If savings are to be estimated for a specific year, actual weather and other data from that year must be used,” while “If savings are to be normalized to typical conditions, typical weather data (e.g., TMY data) should be used.” Either way, “both the baseline and performance period models must be run with the same weather data.”

On the bill side, ENERGY STAR Portfolio Manager offers weather normalized energy, which its Climate and Weather technical reference defines as the energy your building would have used under average conditions, also referred to as climate normals. Weather is not always the biggest driver. The same reference notes that “there is a large amount of base load” in many buildings, and “For this reason, year-to-year weather changes may not have as big of an effect on commercial buildings as you might think.” For a refresher on how normalized metrics fit with EUI and scores, see our guide to commercial energy benchmarking metrics.

Occupancy, schedules and plug loads

A model has to assume how many people are in the building, when they arrive and leave, and what they plug in. Those assumptions are inputs, and they are often defaults. FEMP’s list of data to gather for a simulation includes “Building occupants: population counts, occupancy schedules in different zones” and “Plug loads: summarize major and typical plug loads for assigning values per zone.”

Measured plug loads can differ from the assumption. In a 2012 presentation on integrating energy modeling into design, NREL reported that measured daytime plug loads were “Significantly below daytime plug load predictions,” while noting that “Continuous occupant education needed to reduce nighttime plug loads.” The point is not a number to reuse; it is that these loads should be checked against metering.

Operations, maintenance and as-built conditions

The same model can describe very different buildings depending on how the systems are actually run. FEMP’s Supplement to the M&V Guidelines (Version 4.0, 2023) says: “The key operational conditions include control strategies, set points, operating schedules.” Overrides, overnight equipment runs and failed controls show up on bills, not in design models.

The building that was built may also differ from the one that was modeled. FEMP’s guidance for simulation-based M&V says “Current as-built drawings are preferred, documenting any changes since existing as-builts were completed,” and asks for equipment data “typically collected during an audit.” The Appendix G note also lists “energy use not covered by this procedure,” so loads outside the model’s scope will appear on the bill and nowhere else.

Changes after a baseline period matter too. The IPMVP Committee of EVO, which carries forward the International Performance Measurement and Verification Protocol (IPMVP), describes non-routine adjustments as “factors that were not expected to change,” “but that will affect the building’s energy use,” “not as a result of the energy conservation measures installed as part of the retrofit.”

Rates, billing periods and meter read dates

A model reports energy, and a cost estimate depends on the rates someone assumed. The Appendix G note names “changes in energy rates between design of the building and occupancy” as a reason costs differ. When a model shows dollars, ask which tariff, which demand charges and which year’s rates it used.

Billing periods do not always line up with calendar months. Portfolio Manager handles this by allocation: “Portfolio Manager splits energy data into whole calendar months,” and “Data is apportioned to calendar months based on the average energy use per day.” It also notes that “Monthly bills in increments of 65 days or less are required for weather normalization” “because energy use is compared to monthly weather data.” FEMP asks for billing data that includes “meter read date, kilowatt-hour consumption, peak electric demand,” and stresses “documenting actual read dates of meters to capture duration of data recorded.” A bill period that does not match the model’s month creates a difference unrelated to the building.

How to compare a model with your bills

The fair comparison is a calibrated model, not a design model. FEMP explains that “Model calibration for existing buildings is accomplished by linking simulation inputs to actual operating conditions and comparing simulation results with whole building and/or end-use data.” It asks for “a minimum of 12 (and preferably 24, 36, or 48) consecutive months” of billing data, and says “All models should be calibrated to monthly data at a minimum.” It also states that “Expected calibration requirements should be specified in the project M&V plan.”

FEMP’s Table 4-2, “Acceptable Calibration Tolerances,” gives the targets, noting “Data in this table taken from ASHRAE Guideline 14-2023, Section 5.3.3.3.10.” FEMP presents them as industry standard guidelines that “may be adjusted as required to meet the needs of the project”:

• Monthly calibration: mean bias error (MBE) of plus or minus 5% and CV(RMSE) of 15%.
• Hourly calibration: MBE of plus or minus 10% and CV(RMSE) of 30%.

MBE shows whether the model runs consistently high or low; CV(RMSE) shows how much months or hours scatter. A model can match the annual total and still miss badly month to month, so ask for both indices, the period they cover, and the weather data used.

What to ask for when the numbers differ

Before accepting any conclusion about performance or savings, ask for the model input summary and the bill data side by side.

Know the limits of this guide

Evidence label: documentation-based guide. ClearTrade Report Editorial Team checked the sources below on November 3, 2026. We did not build, run or review any energy model, and we did not analyze any building’s bills. We could not read ASHRAE Standard 90.1 or ASHRAE Guideline 14 directly from ASHRAE; the Appendix G wording comes from New York State’s published 2025 energy standard, and the calibration tolerances come from FEMP’s table citing Guideline 14-2023. Other editions and other programs may differ. The NREL plug load observation comes from one presentation about one project and should not be applied to other buildings. Contracts, incentive programs and M&V plans set their own requirements; read yours.

Primary references checked November 3, 2026: New York Department of State, Normative Appendix G (NYS ASHRAE Standard 90.1-2025), Section G1.2.2 Informative Note; U.S. DOE FEMP, M&V Guidelines: Measurement and Verification for Performance-Based Contracts, Version 5.0 (September 2024), Sections 4.4 and 4.5 and Table 4-2; U.S. DOE FEMP, Supplement to M&V Guidelines, Version 4.0 (September 2023); NREL, Users Manual for TMY3 Data Sets (NREL/TP-581-43156, revised May 2008); NREL, “Integrating Energy Modeling into the Design Process” (NASA Net Zero Workshop, June 6, 2012); U.S. EPA ENERGY STAR Portfolio Manager Technical Reference, “Climate and Weather” (September 2025); Efficiency Valuation Organization, IPMVP Committee page on non-routine adjustments and events.

Frequently asked questions

Is a code compliance energy model a prediction of my utility bills?

No. The Appendix G performance rating method text, as adopted in New York’s 2025 energy standard, says neither the proposed nor the baseline building performance are predictions of actual energy consumption or costs after construction. The rating compares a proposed design with a baseline building.

Why does weather make the model and the bills differ?

Design models often run a typical-year weather file, while bills reflect the weather that actually happened. NREL says a typical meteorological year should not be used to predict weather for a particular period, and FEMP says calibration to utility data requires actual weather data.

What does it mean for a model to be calibrated?

FEMP describes calibration as linking simulation inputs to actual operating conditions and comparing results with whole building or end-use data. Its acceptable tolerances, taken from ASHRAE Guideline 14-2023, are an MBE of plus or minus 5% and a CV(RMSE) of 15% for monthly calibration, and plus or minus 10% and 30% for hourly calibration. FEMP notes these may be adjusted to meet the needs of the project.

How many months of bills should I gather for a comparison?

For simulation-based M&V, FEMP’s M&V Guidelines 5.0 call for a minimum of 12, and preferably 24, 36 or 48, consecutive months of billing data, including meter read dates and consumption. ENERGY STAR Portfolio Manager needs monthly bills of 65 days or less for weather normalization.

Disclaimer: This guide is general information based on government, national laboratory, standards-adoption and industry-protocol documentation, not professional engineering, energy, legal or financial advice for a specific building or contract. Use a qualified energy modeler or M&V professional for any analysis that affects a purchase or contract. AI tools assisted with drafting. ClearTrade Report Editorial Team checked the cited sources and edited this guide. No hands-on testing was performed.