A homeowner in upstate New York spent $14,200 on insulation, air sealing, and a new furnace after her contractor ran an energy simulation showing she would cut heating costs by 32 percent. She did the math: $2,880 saved per year on a $9,000 annual bill, payback in under five years, and a clear path to a $4,000 IRA rebate for exceeding the 35 percent modeled-savings threshold.
Eighteen months later, her bills had dropped by 11 percent. Not 32. Eleven.
She is not an outlier, and she is not unlucky. A systematic review published in the Annual Review of Resource Economics examined 39 evaluations of 23 residential retrofit programs covering 140,977 homes that had undergone energy efficiency upgrades between 1984 and 2021. Not one program delivered deep savings of 50 percent or more. Across that entire sample, mean measured reduction in energy consumption was 7.2 percent.
Where the Models Break
Residential energy models are physics engines pretending they know how people live. They simulate heat transfer through walls, calculate furnace efficiency curves, and model air infiltration rates using blower-door test results. What they cannot model is that your teenager leaves the back door open for the dog, your spouse sets the thermostat to 74 when you leave for work, or your kitchen exhaust fan runs six hours a day because someone burns toast every morning and the smoke alarm sensitivity is set to maximum.
Practitioners call it the “realization rate”: the ratio of actual post-retrofit savings to predicted savings, and it has been studied extensively enough that anyone still quoting raw model output without calibration is either uninformed or selling something. A Performance Systems Development study for NYSERDA found the median realization rate for New York's home performance program was 61 percent. Your model says $1,000 in savings. You get $610. Your $390 evaporates into the gap between a physics simulation and human behavior inside a building envelope that was constructed by the lowest bidder thirty years ago using techniques the model assumes were executed perfectly.
That same NYSERDA study identified the primary culprit: models overestimate baseline energy use, assuming worse starting conditions than actually exist, which inflates the apparent savings from any improvement and makes every upgrade look more impressive on paper than it turns out to be in practice when the furnace cycles and the utility meter spins.
$4.3 Billion in Rebates Depend on Getting This Right
The Inflation Reduction Act's HOMES Rebate Program (Section 50121) allocated $4.3 billion for performance-based residential retrofits, distributed through state energy offices. Rebates scale with modeled or measured energy savings:
| Savings Tier | Standard Rebate | Low-Income Rebate |
|---|---|---|
| 20 to 35% | $2,000 or 50% of cost | $4,000 or 80% of cost |
| Over 35% | $4,000 or 50% of cost | $8,000 or 80% of cost |
DOE built a safeguard into the modeled pathway: all projects must calibrate their energy model to historical utility data using the ANSI/BPI-2400 standard. That calibration requirement exists precisely because the Department knows uncalibrated models overpromise. In the NYSERDA study, applying BPI-2400 calibration pushed the median realization rate from 61 percent to 91 percent. It works.
Whether contractors will do it correctly at scale is another question entirely, one that billions in federal rebate funding are about to answer.
AI-Powered Calibration: Snugg Pro and the DOE Bet
Franklin Energy's Snugg Pro became the first software platform approved by DOE for the HOMES modeled pathway in single-family homes. It automates BPI-2400 calibration by ingesting utility billing data, adjusting the baseline model to match actual consumption patterns, and generating savings projections that account for occupant behavior and as-built conditions rather than assuming a textbook home occupied by thermodynamically rational humans.
Snugg Pro also partnered with Pearl Certification, which documents and certifies home energy features in a format that follows the home through subsequent sales. One interface handles both the rebate qualification modeling and the long-term asset documentation. For a contractor, the workflow compresses from a two-day audit-and-modeling exercise to a few hours of data entry followed by automated calibration and report generation.
Simple math tells the story. If calibration increases realization rates from 61 to 91 percent, a homeowner who invests $12,000 in retrofits expecting $720 in annual savings (30 percent of a $2,400 bill) gets $655 instead of $439. Over 15 years, the calibrated model delivers $3,240 more in real savings than the uncalibrated one. That gap matters. It is the difference between a retrofit that pays for itself and one that never does.
The Strongest Case Against Worrying
DOE already mandated calibration. The worst-case 61 percent realization rate comes from programs that did not require it, and the HOMES program does. If every contractor uses approved software like Snugg Pro and enters data honestly, the accuracy problem is largely solved before the rebate check is cut.
But "if every contractor enters data honestly" is doing enormous work in that sentence. BPI-2400 calibration requires accurate utility billing history, which means the homeowner needs to provide 12 months of bills, the contractor needs to input them correctly, and the model needs to be configured to match the actual house rather than a convenient approximation of it. A contractor incentivized to show higher savings to push the homeowner into the upper rebate tier has a clear motivation to be optimistic with inputs the automated calibration cannot independently verify, like assumed occupancy patterns and thermostat schedules.
What You Should Do
If you are planning a retrofit to qualify for HOMES rebates, three questions determine whether you get real savings or a spreadsheet fantasy.
First: is your contractor using DOE-approved modeling software? Snugg Pro is currently the only approved platform for single-family homes. If your contractor is using a generic energy model without BPI-2400 calibration, the savings prediction is unreliable and your rebate application may not qualify.
Second: did the model calibrate to your actual utility bills? Ask to see the calibration report, and if modeled baseline consumption is more than 15 percent higher than your actual bills show, the savings projection is inflated. Push back before work starts.
Third: choose the measured pathway if you can wait. The HOMES program offers a separate track that calculates rebates based on actual measured savings using open-source measurement and verification software, comparing real pre-retrofit and post-retrofit energy use. You wait longer for the rebate, but you get paid for what actually happened instead of what a model predicted.
Insulation and air sealing consistently rank as the highest-return measures across decades of program evaluations. Storm windows and doors, perennially popular with homeowners, rank lowest. A programmable thermostat costs $50 and, according to the Annual Reviews analysis, delivers savings comparable to retrofit measures costing thousands.
Limitations of This Analysis
The 7.2 percent mean savings figure aggregates programs spanning 37 years, during which building codes, insulation materials, HVAC efficiency, and modeling tools all improved substantially. Current programs likely achieve higher savings than those from the 1980s and 1990s, but no recent large-scale meta-analysis separates modern programs from historical ones. The NYSERDA realization rate data comes from a single state's program and may not generalize to regions with different climates, housing stock, or contractor quality. No HOMES program has operated long enough to generate post-retrofit measured savings data, so the 91 percent calibrated realization rate remains a projection based on pre-IRA program evidence. The IRA's energy provisions face potential modification under the current administration's reconciliation legislation; rebate availability and amounts may change.