Industries / Construction and property / Real estate / PropTech
Property data
The shared records that property software runs on: registers of buildings and addresses, property numbers, building outlines and 3D models, and commercial databases.
6
Milestones achieved, from 1977 to 2020.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1977
Denmark registers every building
Denmark set up its Building and Housing Register (BBR) in 1977. Every property owner filled in a form with that year's general property valuation, the 16th, and the answers became a national register of buildings and dwellings kept by the municipalities. The register was complete and in use by 1980. Its floor areas feed property valuation and tax, and estate agents, mortgage lenders and banks also use its data.
Credit
Danish Ministry of Housing and the municipalities
1987
A researched commercial property database
Andrew Florance founded CoStar in 1987 to sell information on commercial property. Its researchers gather data by phone and email, from public records and data feeds, and by visiting and photographing buildings to record space for lease, sales and tenants. In its 2025 annual report CoStar Group says it has spent more than 35 years building such databases and now also runs home and apartment listing sites.
Credit
Andrew Florance, CoStar Group
2004
Audit finds register unchecked
Denmark's national audit office, Rigsrevisionen, reported in April 2004 that municipalities had checked the BBR's floor-area figures for only 51.1% of buildings by 2003, that sample checks in some municipalities had found many errors, and that no ministry had ever analysed the register's errors as a whole. A new BBR law followed in 2005, and a state agency began supervising how municipalities keep the register up to date.
Lessona building register stays accurate only when someone is paid and checked to keep it current.
Credit
Rigsrevisionen
2018
Machines draw building outlines
In June 2018 Microsoft began publishing building footprints, the outline of each building seen from above, for the whole United States, traced by a neural network from Bing aerial imagery and released as open data. The set now holds 129,591,852 footprints; by Microsoft's own test, 98.5% of the outlines it draws are real buildings (precision) and it finds 92.4% of buildings (recall). Google published Open Buildings in 2021 from satellite images of Africa and has extended it to 1.8 billion detections across Africa, South and Southeast Asia, Latin America and the Caribbean. A 2024 study by the European Commission's Joint Research Centre compared Microsoft's footprints with three other open building sets in five EU countries and found that all are more complete in towns than in the countryside.
github.comsites.research.googleisprs-archives.copernicus.org
Credit
Microsoft Maps; Google Research
2020
Property numbers made open
On 1 July 2020 the UK's Geospatial Commission, with Ordnance Survey, GeoPlace and others, released Unique Property Reference Numbers (UPRNs), a number for each addressable property, together with street and map-feature numbers and their locations, under the Open Government Licence. Other data can now carry the number freely. In 2021 the housing ministry added UPRNs to its published energy performance certificate data, so a certificate can be matched to a property without matching addresses by hand.
Credit
Geospatial Commission, Ordnance Survey and GeoPlace
2020
Open 3D city models
Japan's Ministry of Land, Infrastructure, Transport and Tourism (MLIT) started Project PLATEAU in 2020 to publish 3D city models as open data in CityGML, an international format for city models. Each building carries attributes such as its use and year of construction. The models covered 56 cities in 2021 and about 250 cities now. MLIT plans to link them with building information models and with a 17-digit real estate ID built from the land registry's property number.
Credit
MLIT
Watch for
Stable IDs for every building
Overture Maps Foundation, set up in December 2022 by Amazon Web Services, Meta, Microsoft and TomTom to publish open map data, replaced all its feature IDs with random UUIDs in its June 2025 release. For buildings and six other feature types it says "the same identifier refers to the same real-world entity in every release", so other datasets can join on the ID. It does not promise a link from a removed ID to its replacement when a building is split or merged.
Achieved whenan outside measurement over several monthly releases reports what share of buildings kept their ID, and property records held outside Overture are joined to buildings by it.
Claimed by
Overture Maps Foundation, June 2025
Heights for the world's buildings
Xiao Xiang Zhu and colleagues say their GlobalBuildingAtlas gives outlines for more than 2.75 billion buildings and LoD1 models, simple blocks with a height, for 2.68 billion of them, made from PlanetScope satellite images. They give the height error as a root-mean-square error of 1.5 m to 8.9 m, depending on the continent.
Achieved whenan independent group checks the heights or floor counts against official building registers or field surveys and publishes the errors.
Claimed by
Xiao Xiang Zhu, Sining Chen, Fahong Zhang, Yilei Shi and Yuanyuan Wang, June 2025
Open
01
Check building IDs over time
An independent tracker of whether Overture's building IDs stay attached to the same building from one monthly release to the next, scored against the Netherlands' register of addresses and buildings (BAG), which the municipalities keep and the Kadaster publishes as open data. For each release, match Overture buildings to BAG buildings by outline, then count how many BAG buildings keep the same Overture ID, get a new one, are split, or lose their match.
The briefCredit · unclaimed02
Check estimated floor areas
A public test of how well floor area estimated from open data, outline area times a number of floors worked out from height, matches the floor area recorded in Denmark's BBR. Use at least 10,000 Danish buildings of mixed types and sizes, and have a person review the 100 largest disagreements, marking each as a model error, a register error or unclear.
The briefCredit · unclaimed03
Map who shares parcel data
A public table of every US county saying whether its parcel map, the official outline of each property, can be downloaded free, under what licence, in what format, when it was last updated, and whether a statewide file includes it, with a map of the results.
The briefCredit · unclaimed
Open 01 · the brief
Check building IDs over time
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent tracker of whether Overture's building IDs stay attached to the same building from one monthly release to the next, scored against the Netherlands' register of addresses and buildings (BAG), which the municipalities keep and the Kadaster publishes as open data. For each release, match Overture buildings to BAG buildings by outline, then count how many BAG buildings keep the same Overture ID, get a new one, are split, or lose their match.
Why it is still open
Overture says its building IDs refer to the same building in every release and tells users to measure ID changes on the data they use. A search found only Overture's own release notes, and no outside measurement for buildings.
Built with what exists
Overture's monthly GeoParquet releases, BAG open data from the Kadaster, DuckDB with its spatial extension, and Python.
First step
Download two consecutive Overture building releases and the BAG buildings for the province of Utrecht, match them by footprint overlap, and count unchanged IDs.
Done looks like
A public table for at least six consecutive releases covering one Dutch province or more, giving for each release the share of BAG buildings whose Overture ID stayed the same, changed, split or vanished, with the code to rerun it.
Open 02 · the brief
Check estimated floor areas
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of how well floor area estimated from open data, outline area times a number of floors worked out from height, matches the floor area recorded in Denmark's BBR. Use at least 10,000 Danish buildings of mixed types and sizes, and have a person review the 100 largest disagreements, marking each as a model error, a register error or unclear.
Why it is still open
Roeland Boeters and colleagues at TU Delft (2015) computed usable floor areas from detailed 3D models of Hoogvliet-Zuid, a district of Rotterdam, and found only 38% of homes within the tolerance the Dutch standard allows when compared with the national register, or 77% under what they took to be the register's older definition. That test covered one district and used detailed models. GlobalBuildingAtlas publishes heights and block models for 2.68 billion buildings but gives no floor areas, and a search found no study that checks floor areas estimated from global open data against a register. Valuers, lenders and energy analysts would use such estimates where registers are missing.
Built with what exists
GlobalBuildingAtlas or Overture heights and outlines, BBR building records, Python with GeoPandas.
First step
Pick one Danish municipality, take its BBR building records and the matching outlines and heights, and compute floor area errors for detached houses first.
Done looks like
Median and 90th-percentile error, in square meters and in percent, by building type and height band, for at least 10,000 buildings, with the hand-reviewed list of 100 large disagreements published.
Open 03 · the brief
Map who shares parcel data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of every US county saying whether its parcel map, the official outline of each property, can be downloaded free, under what licence, in what format, when it was last updated, and whether a statewide file includes it, with a map of the results.
Why it is still open
The National States Geographic Information Council (NSGIC), the body of US state GIS coordinators, said in 2024 that the parcel data now collected and made available "is not complete nor reliable", and it describes availability state by state, not county by county. Parcel maps are kept county by county. New York's state portal offers 38 of the state's 62 counties in one file because, it says, "we respect each county's right as the data owner to control the distribution" of its data. Regrid sells nationwide parcel data but its product page gives no county-by-county account of which data is free. One search found no national table.
Built with what exists
State GIS clearinghouses and county GIS portals, a spreadsheet, and optionally a web agent such as Browser Use (see the Note "Deleting 80% to start over") to visit county sites, with people checking a sample.
First step
Fill the table for New York's 62 counties and for one state without a statewide parcel file, timing how long each county takes.
Done looks like
Every county listed with its status and a source link, 10% of entries rechecked by a second person with the agreement rate published, and the date of each check.