A used Honda Civic comes with a Carfax report. Every oil change, every accident, every odometer reading. You would never spend $25,000 on a car without checking its history.
A house costs twenty times more, and when you buy one the seller hands you twelve numbers on a utility summary sheet. January through December, total kilowatt-hours consumed, and that passes for disclosure.
Meanwhile, bolted to the side of that house, a device has been recording its energy behavior every fifteen minutes for the entire time the seller lived there. Your future utility has all that data, and AI systems exist that can read it like a doctor reads an EKG. But nobody in the transaction ever sees it.
The Data Is Already There
According to the U.S. Energy Information Administration's Form 861 data, 123 million residential smart meters were installed across the United States by the end of 2024. That's up from 57 million in 2015, and roughly 78% of all residential electric meters in the country now qualify as Advanced Metering Infrastructure. These are not the old analog discs spinning behind a glass dome that a meter reader squinted at once a month with a clipboard. AMI meters record consumption at intervals as short as every eight seconds, transmitting that data back to the utility in near-real-time.
Run the arithmetic on a standard 15-minute interval meter and the scale of the data becomes immediately, almost absurdly, clear. Ninety-six readings per day, 365 days per year: 35,040 data points annually. Over five years, a fairly typical ownership period, that's 175,200 measurements describing exactly how a home uses energy: when the HVAC cycles on and off, how the building envelope responds to a cold front, whether the heat pump pulls more amperage each successive winter as its compressor ages, how quickly the house bleeds heat after the thermostat drops at midnight.
What does a homebuyer see? Twelve monthly totals and, if the seller's agent is feeling generous, a bar chart on a one-page PDF that reveals less about the building's performance than a gas station receipt reveals about your car. Information loss: 99.97%.
Nobody blinks. Nobody asks.
What AI Can Already Read in the Signal
Ready. The technology exists, and it works. Non-Intrusive Load Monitoring, or NILM as the research community calls it, disaggregates a home's total electrical signal into individual appliance-level consumption using only the data from that single meter on the wall. Three decades old. Researchers have been refining NILM since the early 1990s, but only in the last five years has the combination of deep learning architectures capable of parsing millions of waveform samples with ubiquitous smart meter hardware generating those samples in every neighborhood in America pushed the technique from academic curiosity to something that could function as a genuine diagnostic tool at the scale of a national housing market.
A review published in Energies (MDPI) examined NILM-based fault detection for HVAC systems and found that existing techniques can identify ten distinct rooftop unit faults — refrigerant leaks, compressor degradation, economizer failures, evaporator fouling — with 94% accuracy. Those faults are responsible for 10% to 30% of total HVAC repair costs. According to the same review, NILM-based fault detection and diagnostics, deployed across the residential housing stock where most Americans pay their largest monthly bill after their mortgage, could reduce building energy consumption by 20% to 30%, savings that compound year after year but that nobody captures because the data never leaves the utility's servers.
A more recent approach called HNILM, built on a dual-branch convolutional neural network with multi-head attention, pushes into appliance health prediction. Working from 8-second commercial smart meter data alone, it grades individual devices on a four-level health scale: Normal, Low, Medium, and High degradation. No sensors wired to individual appliances, no technician crouching in the ductwork with a clipboard, no inspector crawling through the attic with a flashlight and a four-page checklist. Just the raw data the utility already has sitting in a database somewhere, accumulating terabytes of behavioral insight about a building that will eventually sell to someone who will never be told any of it.
Researchers at the University of Florida built a system using WiFi thermostat data combined with smart meter readings to infer insulation R-values, window thermal performance, duct leakage, and HVAC sizing accuracy. Exactly the kind of information a buyer currently discovers only through a $500 energy audit, which fewer than 3% of transactions include, or through the first winter utility bill after closing, which arrives roughly ninety days too late to renegotiate anything.
Almost Nobody Requires Disclosure
Montgomery County, Maryland, requires home sellers to provide twelve months of utility bills before signing a purchase contract. No energy audit is required, no raw smart meter data changes hands, and photocopied bills suffice. New York proposed a similar requirement in 2025, calling for twelve months of consumption data to be disclosed at every residential sale or lease in the state. As of mid-2026, after more than a year of committee review, industry lobbying from both utilities worried about data liability and real estate groups concerned about adding friction to an already-complex closing process, the proposal remains pending with no public timeline for a vote.
California's AB 1103, signed in 2007, requires energy use disclosure, but only for nonresidential buildings over 5,000 square feet. Single-family homes, where most Americans actually live, are exempt.
Chicago's MyHomeEQ platform, built by the regional Multiple Listing Service and a third-party data aggregator, automatically displays utility cost data on home listings. By the standards of American residential energy disclosure, it counts as sophisticated. But fewer than 5% of MLS boards nationwide offer anything comparable, and even MyHomeEQ, one of the most advanced implementations in the country, displays the same aggregated monthly data that every other disclosure system provides rather than the granular fifteen-minute interval data that makes machine learning diagnostics possible and that already exists in the utility's database.
In 2012, the Department of Energy launched its Green Button initiative, which lets utility customers download their consumption data in a standardized XML format. About sixty utilities participate nationwide. But only current account holders can access their data, which means the initiative is structurally irrelevant to the one moment when energy data matters most: the transaction. A prospective buyer, the person in the entire transaction who stands to gain or lose the most from understanding the building's energy performance, has no account and no access until after closing, when the meter transfers to their name and the leverage to renegotiate has evaporated.
An Original Calculation: The Market Failure in Numbers
Here is what a buyer of one of the 123 million AMI-metered homes in the US is missing, expressed as a diagnostic comparison.
Consider what currently passes for due diligence on a home's mechanical systems: a standard home inspection runs four to six hours, costs $400 to $700, and relies on a single human walking through a house they have never been inside before. The inspector tests a handful of outlets, runs the furnace for a few minutes, looks at the water heater, and produces a report based on a single afternoon of observation . Call it four hours of data collection on the home's systems.
The smart meter on the side of the house, operating continuously at 15-minute intervals over a five-year ownership period, has collected the equivalent of 2,920 eight-hour inspection days of energy system observation. That's 175,200 data points covering every heating cycle, every cooling recovery, every unexplained power spike at 3 a.m., every gradual efficiency decline in a compressor losing refrigerant charge over sixteen months.
One gets paid. One doesn't.
Running a NILM analysis on five years of smart meter data costs effectively nothing once the model is trained — a few seconds of compute on a standard server. Florida's thermostat-based audit costs nothing beyond data transfer and model inference. Not one of those barriers is technical, and not one is financial: every obstacle standing between a buyer and a diagnostic reading of the home they are about to purchase is institutional, a product of fragmented utility regulation, absent disclosure mandates, and a real estate transaction process that has not updated its information requirements since the era of analog meters.
The Strongest Case Against Disclosure
Privacy is the most serious counterargument, and it deserves to be stated at full strength rather than dismissed. Granular smart meter data reveals far more than how many kilowatt-hours a household consumed last month. Fifteen-minute interval data can infer occupancy patterns : when residents are home, when they sleep, when they travel. NILM disaggregation can identify specific appliances: medical equipment, grow lights, server racks. In 2010, the National Institute of Standards and Technology's NISTIR 7628, a three-volume guidelines document on smart grid cybersecurity, warned that smart meter data could enable "surveillance without physical intrusion." In parallel, the Department of Energy's own guidelines emphasize customer consent and limit third-party data sharing.
Every one of those concerns is legitimate, rooted in genuine constitutional privacy interests and in the very real possibility that algorithmic inference from energy data could reveal medical conditions, work schedules, and personal habits that no seller should be compelled to disclose, and together they have stalled most legislative efforts at mandating energy disclosure for residential properties. But they don't explain why the data can't travel with the property when the property sells, any more than a home inspection report detailing mold or foundation cracks constitutes an invasion of the seller's privacy. Nobody is proposing making smart meter data public. The question is narrower: should a buyer paying half a million dollars for a building be able to see how that building actually performs, as measured by instruments the government required the utility to install?
What a Home "Energyfax" Would Look Like
Imagine a single-page report, generated automatically from five years of 15-minute smart meter data, that told a buyer:
- This home's HVAC system has been declining in efficiency by 4.2% per year since 2023. At current trajectory, the compressor will require replacement within 18 months.
- The building envelope loses heat 31% faster than comparable homes built to the same code in the same zip code, suggesting insulation deficiency or significant air leakage.
- Electricity consumption spikes 22% between 2 a.m. and 5 a.m. during summer months, consistent with a cooling system that cycles excessively due to undersizing or duct loss.
- Annual energy cost: $3,420. Comparable homes in this area: $2,180. Estimated upgrade cost to close the gap: $4,800 to $7,200 (insulation, duct sealing).
None of that requires new hardware because the meter is already bolted to the side of the house. The data is already in the utility's database. The AI models that can extract these insights from aggregate consumption signals have been published in peer-reviewed journals for over a decade. The only missing component is a policy framework that says: when the home sells, the data follows.
Limitations
NILM disaggregation accuracy varies widely by appliance type. High-power devices like HVAC systems and dryers are reliably identified with accuracy above 90% in most studies. Small or variable loads like laptop chargers and LED lighting remain difficult, typically falling in the 60% to 75% range. The HNILM health-grading system referenced above was validated on the UK-DALE and REDD benchmark datasets, which contain approximately 200 homes, far fewer than the 123 million meters in the field. Generalization across the full diversity of American housing stock, climates ranging from Fairbanks to Phoenix, and utility rate structures that vary wildly between investor-owned utilities and rural cooperatives remains entirely unproven at scale.
Smart meter interval data is also not universal. While 78% of residential meters are AMI-capable, reporting intervals vary by utility: some transmit hourly data, others report every 15 minutes, and a few capture sub-minute intervals. The diagnostic precision of any AI analysis depends directly on temporal resolution.
This analysis focuses exclusively on electricity data. Natural gas consumption, which drives heating in much of the northern US, is metered separately and not included in most AMI deployments. A complete "energyfax" would require both streams, and the natural gas metering infrastructure lags significantly behind electric.
The Math Doesn't Care About Your Disclosure Policy
Every year, roughly 5.5 million existing homes sell in the United States. Approximately 4.3 million of those, about 78%, now have AMI smart meters recording their energy behavior in granular detail, capturing every cycle and anomaly. AI can already interpret that data, and no buyer has ever been shown it.
You wouldn't buy a car without its service history, and you wouldn't invest in a company whose management refused to share its financials. But you'll spend $420,000 on a house, the median US home price as of early 2026, armed with less diagnostic information about the building's largest and most expensive operating system than the sixteen-year-old at the oil change shop has about your brake pads.
Smart meters do not care about disclosure policy, and they never stop recording: thirty-five thousand data points a year, every year, streaming into a database that nobody reads at the moment it matters most.