Denmark's 2017 Ejendomsvurderingslov (Property Assessment Law) required the Tax Ministry to replace manual, caseworker-driven home valuations — the last full round dated to 2011 — with an automated valuation model (AVM) covering every property in the country.
The system, built and operated by Vurderingsstyrelsen (the Property Assessment Agency, part of the Danish Tax Agency), uses a statistical/machine-learning model trained on roughly 19 parameters per property: recent sale prices in the neighbourhood, distance to forests, coastline, schools and railway lines, plot buildability and ground pollution, among others. It generates a taxable value for each of Denmark's approximately 1.7 million homes without an individual case review.
Originally due to launch in mid-2019, the rollout slipped repeatedly. Denmark's National Audit Office (Rigsrevisionen) reported in December 2025 that the project finally closed in June 2025 — eight years behind the original schedule and at a total cost of DKK 4.9 billion against an initial budget of DKK 96 million, an overrun of roughly 51 times. The Tax Ministry told the Audit Office it still expects error rates to fall only once remaining functions are delivered in the first half of 2026, at a further cost of DKK 44 million.
Independent observers flagged problems well before the 2025 closeout: AlgorithmWatch's Automating Society report documented an official +/-20% accuracy band around each valuation and widespread homeowner confusion, and researchers at the University of Southern Denmark wrote separately about citizens 'facing an opaque system' they could not interrogate.
This practice is included as a cautionary case: a technically ambitious, fully-scaled national AI system whose public record is now dominated by its own auditor's findings of years of schedule slippage, cost overrun and unresolved error, rather than by measured public benefit.
Read the full analysis: https://www.lovguiden.dk/det-offentlige/rigsrevisionen/2025-11-27-ejendomsvurderingssystem-afsluttet-8-aar-forsinket-og-til-49-mia-kr
Where this practice's information was retrieved from, and when.