In 2024, more than 436,000 new American homes received a HERS Index Score. A certified rater walked through each one, checked the insulation, tested the ductwork, ran the numbers through REM/Rate software, and produced a certificate. On that certificate, in bold type, was a number telling the buyer how much energy their home would save.
Compared to what?
A hypothetical house built to the 2006 International Energy Conservation Code. That baseline was set when Bush was president, gas cost $2.59 a gallon, and a code-compliant home could still have single-pane windows in some climate zones. Nobody has intentionally built to that standard in nearly two decades. Every state has tightened its energy code at least once since then. Most have done it three or four times.
Yet the HERS Index still measures everything against it.
What the Certificate Actually Promises
RESNET, the organization behind the HERS Index, publishes a one-page "Rated Home Label" for every certified home in its national registry. It shows estimated annual energy use, estimated savings, and estimated carbon reductions. Green Builder Media, in its analysis of the scoring system, identified a problem embedded in the presentation: savings are cited "relative to an average U.S. home," which is "highly ambiguous." Is that an average new home? An average 1970s ranch? An average across all 140 million American housing units, half of which were built before 1980?
It matters. RESNET claims the average HERS-rated home saves $1,100 per year compared to code-minimum construction. Multiply that across the 436,000 homes rated in 2024, and you get $479.6 million in promised annual savings. That figure assumes every home performs as modeled. Not one of those savings was measured from an actual electric meter or gas bill.
What the Utility Data Shows
Arik Levinson, an economist at Georgetown University, tried to answer a simple question: do homes built under stricter energy codes actually use less energy? He used three independent approaches: comparing homes within the Department of Energy's Residential Energy Consumption Survey, analyzing California utility billing data directly, and benchmarking California, which has had among the nation's most aggressive building codes since 1978, against all other states.
All three approaches produced the same result, and it was not what the certificates promised. After controlling for the size and location of homes, and for the income, age, number, and education of occupants, Levinson found "no evidence that homes constructed since California instituted its building energy codes use less electricity today than homes built before the codes came into effect."
Read that again. California's average HERS score in 2025 was 10. A score of 10 means the energy model says these homes are 90% more efficient than the 2006 baseline. Yet the actual electricity consumption data cannot distinguish them from pre-code homes.
Where the Models Break
A HERS rating runs on REM/Rate, a physics-based simulation. It takes the building envelope, the mechanical systems, the climate zone, and a set of standardized assumptions about occupant behavior, then calculates how much energy the home should consume. Should is doing a lot of heavy lifting in that sentence.
UK research has spent a decade quantifying the gap between should and does. A joint study by the Zero Carbon Hub and Innovate UK found that new buildings typically consume 2.6 times more energy than their design models predicted. In some cases, researchers documented gaps reaching 287%. American data is thinner, partly because nobody has built a nationwide system for matching HERS predictions to actual billing data. A DOE-sponsored study by the Rocky Mountain Institute found that algorithm-based energy assessments showed a 20 to 30 percent average absolute difference from physics-model estimates, even without introducing real occupant behavior into the equation.
Put bluntly: the models disagree with each other before they even meet reality.
And then reality shows up. Families leave windows cracked because the house feels stuffy from overly tight construction. A teenager runs a gaming PC that draws more than the living room lights. Somebody sets the thermostat to 74 instead of the assumed 72. EIA projects residential electricity rates will climb from 16.48 cents per kilowatt-hour in 2024 to 18.70 cents in 2027, a 13.5% increase in three years. Even if consumption matches the model exactly, the bill will not.
What AI Scoring Gets Right
A new generation of energy assessment tools sidesteps the physics model entirely. ClearlyEnergy and UtilityScore, both participants in the RMI/DOE study, generate estimates from actual utility billing records, tax assessor data, and satellite imagery rather than standardized simulation inputs. Machine learning models in peer-reviewed literature, using gradient-boosted decision trees like XGBoost trained on tens of thousands of real homes, have achieved R-squared values above 0.98 when predicting actual annual energy consumption. SHAP-based interpretability analysis from these models reveals that windows account for 74.3% of a home's envelope-related energy impact, floors 20.2%, walls 4.2%, roofs just 1.4%. That hierarchy does not match how most builders think about insulation upgrades.
These tools do not need a certified rater to visit the house. They do not compare against a 2006 ghost. They measure what a home like yours, in your climate, with your square footage and your vintage of equipment, actually costs to heat and cool based on what similar homes have actually cost. Not should. Did.
The Strongest Case for HERS
Defenders of the current system make a fair argument: HERS was never designed to predict utility bills. It was designed to rank relative efficiency, not to forecast a dollar amount on a monthly statement. A home with a HERS score of 40 genuinely has a better thermal envelope and tighter ductwork than a home at 70. Freddie Mac data shows HERS-rated homes sell for 2.7% more, and better-rated homes command up to a 5% premium. As a market signal, HERS works.
That argument holds right up until the buyer reads the certificate. Estimated annual energy cost. Estimated savings. Bold type. Large font. Those numbers are not hedged as relative rankings. They read as dollar amounts the buyer should expect to save, and Green Builder Media is right to call that "a huge over-promise." Builders absorb the reputational cost when the first January gas bill arrives and the homeowner starts comparing it to the piece of paper pinned on their refrigerator.
What Builders and Buyers Should Do
For builders: the HERS score is a legitimate competitive tool, but pair it with real data. If you have a subdivision with 50 occupied homes, collect anonymized utility billing from 10 of them. Show prospective buyers what similarly rated homes in the same development actually pay. That number is worth more than any model output.
For buyers: ask two questions. First, what is the HERS score? Second, what do current homeowners in this community actually pay for energy? If the builder cannot answer the second question, the first one is a physics estimate, not a financial guarantee.
For both: AI-powered energy scoring is still early. No tool is perfect, and most need more training data in underrepresented climate zones and housing types. But the trajectory is clear. Tools that learn from millions of actual utility records will eventually replace tools that simulate what a home should do under conditions that exist only inside a computer.
When that happens, the certificate on your fridge will finally match the bill in your mailbox. We are not there yet.