Medina County, Ohio. Population 182,000. A new leadership team at the county assessor's office had a problem they could smell but not measure: the property sketches in their CAMA database, the official records that determine what gets taxed and what doesn't, were riddled with gaps. Somebody would add a deck, convert a garage, sink a pool, and never pull a permit. It happens constantly. Each addition would exist on the ground but not in the records, invisible to the tax rolls for years, sometimes decades, accumulating into a gap so large that nobody in the assessor's office could estimate its true size.
So they measured it.
They fed every sketch into EagleView's Sketch Inspect, an ML system that overlays assessor records onto high-resolution aerial imagery and flags mismatches, and the results came back inside 30 days.
More than 9,000 properties had major mismatches or missing sketches: pools that don't appear in any record, garages that grew by 400 square feet without explanation, additions that doubled a home's footprint without a single line drawn in the county database, all adding up to $40 million in newly identified taxable value. Forty million. Recovered property tax revenue: $14 million.
How It Works, Mechanically
Every county assessor's office maintains sketches: simple outlines of each structure on a parcel, with dimensions and labels for garages, patios, living space. EagleView's system takes those sketches and superimposes them onto building outlines extracted from aerial imagery captured two to four times per year in metro areas. A proprietary algorithm classifies every parcel: green for alignment, orange for minor discrepancy, red for major mismatch, purple for missing sketch entirely. A reviewer can filter by severity, zoom in, and confirm or dismiss the flag without leaving their desk.
Nearmap runs a parallel operation at different scale. Its single global ML model generates 130-plus semantic layers from aerial captures, identifying not just building footprints but roof material, pool condition, solar panels, and construction sites. In 2025 alone, the system produced more than 7 billion AI insights across its 100-petabyte imagery archive, a volume of analysis that would have required every assessor in the country working full-time for a century to replicate manually. SEMCOG, the Southeast Michigan Council of Governments, used Nearmap AI to identify 3,425 buildings that were entirely missing from its regional inventory.
Not miscategorized, not slightly off in dimensions, but entirely missing from the database, as if those 3,425 structures had never been built.
What Inspectors Cannot Reach, Satellites Can
Prince George's County, Maryland illustrates why this technology is spreading. Before 2021, building inspectors there were forbidden to enter properties posted with "no trespassing" signs. Getting legal access to a property posted with "no trespassing" signs could take 18 months of court filings and bureaucratic process. Behdad Kashanian, the county's inspections division associate director, described the consequence bluntly, and the picture he painted should keep every homeowner with unpermitted work awake at night: "If it is a case involving unpermitted construction, the illegal construction can be completed, the house sold to an unsuspecting home buyer and the person responsible for the illegal construction will have moved away."
In August 2021, Prince George's County launched an aerial inspection program using manned police flights and drones. They identified more than 230 problem sites in the first few months. Drones reached 400 feet, traveled 50 miles per hour, and surveyed a quarter-mile radius per flight. No warrant needed, no trespass dispute, no 18-month wait for a judge to grant access to a backyard the assessor already had the right to view from the sky.
In Europe, ESA's InCubed program funded a satellite-based system specifically designed to detect unpermitted pools and building extensions by cross-referencing ML detections against municipal building registries. Spanish municipalities are the primary market, but the technical pattern is identical to what EagleView and Nearmap sell domestically: compare what exists to what the records say exists, and flag the gap.
Guernsey County's Math
For anyone who thinks Medina County is an outlier, Guernsey County, Ohio ran the same system and got results on the first day that silenced the skeptics. Three reviewers examined 31 buildings and found $3,154,200 in increased appraised value and $50,614 in additional tax revenue. That works out to roughly $3,000 per hour per reviewer in found money, which makes this one of the highest-ROI line items any county government can add to its budget.
Fairfax County, Virginia, population 1.15 million, was the first in the nation to implement EagleView's ChangeFinder, a companion product that compares sequential aerial captures to detect new construction between survey periods. Residential appraiser Teresa Rhinehart told EagleView she uses it every day. Her colleague Forrest Mathews noted that it solved their persistent access problem: gated communities and locked backyards that used to eat hours of field time now get reviewed remotely.
What This Means for the Homeowner Who Skipped the Permit
In California, any substantial addition to real property triggers reassessment under a 1983 state law: new rooms, pools, garages, all of it, followed by a supplemental tax bill that reflects the increased value. If you built without a permit, you face both the reassessment and the penalty for unpermitted work. Fines for unpermitted construction range from $1,000 to $10,000 or more depending on the jurisdiction. A pool permit, by comparison, runs $500 to $1,500. Do the math. It isn't close.
Before aerial AI, enforcement was almost entirely reactive: you got caught during a property sale when a buyer's inspector noticed the addition didn't match public records, or a neighbor complained, or a field inspector happened to drive by on a Tuesday. Now enforcement is systematic, and a county that runs Sketch Inspect isn't sampling properties at random. It is scoring every parcel against satellite imagery in a single pass, prioritizing the biggest mismatches, and sending reviewers to those addresses first.
If you have unpermitted work, the window for voluntary self-reporting is narrowing. Retroactive "as-built" permits remain available in most jurisdictions and cost substantially less than fines, forced remediation, or the discovery-during-sale scenario where the buyer renegotiates or walks entirely. Talk to your local building department before the satellite does. Seriously.
Privacy and Pushback
County assessor aerial inspections have existed for decades in analog form, and no court has ruled that viewing private property from public airspace violates the Fourth Amendment for assessment purposes, but what has changed is the automation and the completeness of coverage. A human reviewer looking at printed aerial photos might cover 20 parcels in a day. An ML system compares the entire county in a month and classifies every mismatch by severity. Scale changes the nature of surveillance, even when each individual observation was always legal, because the difference between a county assessor glancing at twenty aerial photos over coffee and an algorithm scoring every parcel in the jurisdiction overnight is not a difference of degree but of kind.
Civil liberties advocates have not yet mounted a sustained legal challenge to ML-based property assessment from aerial imagery, which is itself revealing. But the political dynamics are unstable. A county recovering $14 million in tax revenue has a powerful incentive to deploy; a homeowner who built a pergola and now faces a $5,000 fine plus reassessment has a powerful incentive to push back. Expect this tension to produce state-level legislation within two years, likely framed around notice requirements and appeal rights rather than outright prohibitions on the technology, because no state legislature is going to vote against $14 million in recovered revenue when the alternative is asking taxpayers to cover the difference.
Limitations of This Analysis
Medina County and Guernsey County data comes from EagleView's own marketing materials, not independent audits, and we could not verify the $14 million recovery figure against county financial records. Prince George's County's 230-plus sites were reported in an October 2021 newsletter and may not reflect current program scope. False positive rates for sketch verification systems are not publicly reported by any vendor. California reassessment rules cited here are state-level; individual county implementation varies.