A couple in Boulder paid $62,000 more than the comparable listing next door. Triple-pane windows, a ground-source heat pump, 9.6 kW of rooftop solar, and the words “net-zero ready” on every page of the marketing brochure. Their energy model predicted annual consumption of 14,200 kWh, balanced by solar production of 14,800 kWh. A tidy surplus. The first twelve months of utility data told a different story: 28,400 kWh consumed. Twice the model. The solar panels worked fine. Everything else was off.
They are not unusual.
What the Research Measured
The New Buildings Institute’s Getting to Zero Database tracks 922 verified and emerging zero-energy buildings across 50 countries. Its January 2026 update showed a median energy use intensity (EUI) of 29 kBtu/ft²/yr for verified net-zero projects — 65% below the U.S. commercial average of 83 kBtu/ft²/yr. Impressive on paper. But dig into the verification data and roughly one in five buildings targeting net-zero certification failed to hit their designed EUI within the first two operational years.
A CIBSE/University College London meta-analysis published in 2024 put hard numbers on what building scientists now call “the performance gap” — actual energy consumption running 30% to 150% higher than design models predicted across hundreds of monitored commercial buildings. At the worst end, buildings consumed 2.5 times what their models promised. Root causes clustered around four factors: occupant behavior diverging from modeled assumptions, HVAC commissioning failures, thermal envelope defects missed during inspection, and plug loads that nobody bothered to estimate realistically.
That research focused on commercial buildings. Office towers. University lecture halls. The residential gap is almost certainly worse, because residential buildings have a variable that commercial properties can partially control: the people who live in them.
Why Identical Houses Consume Different Energy
A 2023 study published in MDPI Sustainability tracked energy consumption in physically identical homes — same builder, same orientation, same HVAC systems, same insulation package. One house consumed twice the heating energy of its neighbor. The only variable was who lived there.
Across the broader dataset, occupant behavior accounted for up to 80% of the variation in energy consumption between identical residential units. Eighty percent. Not insulation R-values, not window U-factors, not HVAC efficiency ratings. People.
Energy models assume “standard” occupant behavior — thermostat setpoints of 70°F heating and 76°F cooling, predictable occupancy schedules, reasonable plug loads. These standard occupants don’t exist. Real humans open windows while the AC runs, heat the garage, plug in space heaters because the bedroom “feels cold,” and leave the oven door open after baking because the kitchen floor is tile. No energy model accounts for the oven-door-as-space-heater occupant.
The MDPI researchers found that coupling occupant behavior models into building energy simulations could reduce prediction error by 12% and improve comfort scores by 5%. Which means even the fix gets you less than a sixth of the way there.
The Residential Blind Spot
Almost all performance gap research comes from the commercial sector. Bigger buildings, bigger energy bills, institutional owners with metering infrastructure and incentives to track. Residential net-zero homes get a model at design time, a certification label at completion, and then nobody measures anything again.
California’s 2026 Title 24 building code mandates aggressive design-stage energy performance — electric-ready systems as the default path, heat pump water heaters, EV charging infrastructure in every new build. What it does not mandate is post-occupancy verification. A builder can deliver a home that meets code on the plans, collect the net-zero premium, and never confirm whether the home performs as modeled once a family moves in.
No state requires it. No certification program enforces it over time. LEED, PHIUS, DOE Zero Energy Ready — all certify the design, not the result.
$250 Changes the Equation
AI-powered home energy monitors now cost less than a single month’s winter heating bill in most of the country. And they do something no energy model, certification, or code inspector can: measure what actually happens after move-in.
| Monitor | Cost | What It Does |
|---|---|---|
| Sense Energy Monitor | $250–$350 | Clamps onto mains, uses ML to identify individual appliances by electrical signature. Disaggregates whole-home draw into device-level consumption without per-circuit wiring. |
| Emporia Vue | $150–$220 | Per-circuit monitoring with up to 16 branch circuit sensors. Shows exactly which circuits (HVAC, kitchen, EV charger) consume what. |
| Span Smart Panel | ~$3,500 + installation | Replaces your entire electrical panel. Circuit-level monitoring plus remote control — shut off individual circuits from your phone. Integrates with solar and battery systems. |
The math on payback: at the U.S. average electricity price of $0.16/kWh and an average household consumption of roughly 10,500 kWh/yr, a 6% reduction (the conservative figure from a Hydro One utility trial across 500 Ontario homes using simple in-home energy displays) saves about $101/yr. A Sense monitor pays for itself in under three years. An Emporia Vue, under two. Active users who actually respond to the data — adjusting thermostat schedules, identifying phantom loads, catching a failing compressor early — report reductions of 8% to 12%, pushing payback under 18 months.
Span is the expensive option, but it’s the only one that both monitors and controls. For a homeowner who just paid a $50,000 net-zero premium, $3,500 to actually verify the claim isn’t a splurge. It’s the receipt.
What a Real Net-Zero Guarantee Would Require
Imagine a builder who offered this: “If metered energy consumption exceeds the model by more than 15% in the first two years, we’ll diagnose the gap and fix it at our cost.”
It would require three things that almost no builder currently provides:
Installed monitoring at close. An energy monitor in the panel, configured and transmitting, before the buyer gets the keys. Cost to the builder: $250–$500 per unit. On a $600,000 home, that’s 0.05% of the sale price.
A 12-month baseline period. Metered data compared against the original energy model, normalized for weather (heating degree days and cooling degree days from local NOAA data). Not a single month — a full seasonal cycle.
A defined remediation process. Blower door retest. Infrared thermal scan of the envelope. HVAC commissioning verification. The diagnostic tools already exist, and they’re not expensive. A residential blower door test runs $300–$500. Infrared scans cost $200–$400. Total diagnostic package: under $1,000.
No builder offers this. Not because it’s technically impossible — the tools exist and the costs are trivial relative to net-zero premiums — but because the current system lets them sell a model instead of a result.
The Strongest Argument Against Post-Occupancy Verification
It’s this: occupant behavior isn’t the builder’s fault. The MDPI study showed 80% of the variation comes from how people use their homes. A builder can deliver a perfectly constructed, perfectly insulated, perfectly commissioned net-zero house, and the occupant can still blow the energy budget by heating the garage to 72°F and running a server rack in the basement.
That’s fair. Builders shouldn’t be liable for behavior they can’t control.
But the counterargument to the counterargument is this: if your energy model assumes occupant behavior that you know doesn’t exist — if you’re using “standard” thermostat setpoints and “typical” plug loads that decades of research have shown don’t reflect real households — then the model itself is the misrepresentation. The performance gap isn’t just about bad occupants. It’s about unrealistic models that builders have no incentive to make realistic, because nobody checks.
A weather-normalized, behavior-adjusted verification protocol could separate the two. AI disaggregation tools like Sense can identify whether excess consumption comes from the HVAC system (a building problem) or from the gaming PC and the hot tub (a lifestyle problem). The technology to assign accountability already exists. The will doesn’t.
What Buyers Should Demand
If you’re paying a net-zero premium — and $30,000 to $80,000 is the range we’re seeing on certified net-zero new construction in markets from Austin to Portland — ask for three things before signing:
The energy model inputs, not just the output. What occupant behavior assumptions were used? What plug load assumptions? What thermostat setpoints? If the builder won’t share the model file, the number it produces is marketing.
An installed energy monitor at closing. Sense or Emporia at minimum. If the builder declines a $250 addition to a $600,000 sale, ask yourself what they’re afraid you’ll measure.
A first-year performance review clause. Written into the purchase agreement: 12 months post-occupancy, the builder reviews metered data against the model. No penalty required. Just a conversation, documented, with the data on the table. Builders confident in their work won’t object.
Limitations
This article relies primarily on commercial building data for performance gap statistics — the NBI and CIBSE/UCL datasets cover office, institutional, and mixed-use buildings far more thoroughly than single-family homes. We extrapolate to residential, but hard residential performance gap data at scale doesn’t exist yet, precisely because post-occupancy monitoring is so rare in homes. The AI monitor energy savings figures (6–12%) come from utility trials and manufacturer-reported data; long-term, peer-reviewed residential studies with control groups remain scarce. Cost figures for Sense, Emporia, and Span reflect retail pricing as of March 2026 and do not include electrician labor for installation where required. The Boulder couple referenced in the opening is a composite based on reported cases, not a single identified household.
Priya Greenwood covers sustainability and green building for AI Home Building. She believes in insulation, metered data, and never trusting a number nobody verified.