Checks of firms flagged by a Danish tax model find misuse in about half

Rigsrevisionen reports that the Danish Tax Agency checked 74 companies that its machine-learning model flagged in 2019. By preliminary results, the hit rate was 47% and the revenue DKK 5.3 million.

  • 2 for an organization

    By the agency's preliminary results, checks of 74 companies that the model flagged had a hit rate of 47% in the control's first year.

  • 1 for society

    The checks brought in DKK 5.3 million for the state by the end of 2019, a preliminary figure.

Depth runs from 1, a convenience or time saved, to 5, a life or a livelihood preserved.

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

An independent descriptive report supports the benefit.

The system’s part
Not isolated. Rigsrevisionen relays the agency's preliminary results for checks of companies that the model scored 0.5 or above. It gives no hit rate for reversals checked without the model, since the agency had no separate check of reversals before 2019. The agency then checked 74 of the 222 flagged companies by the end of 2019. The report's glossary defines machine learning as a form of artificial intelligence that finds patterns in large amounts of data and predicts the probability of outcomes.
Later
Nothing later is known. Rigsrevisionen reports that the agency uses two data-mining models to pick declarations for control, one of which looks for possibly unjustified reversals. It gives no results for the models and does not say whether this is the model of 2019.
Not yet checked
LimitsCosts and harmsLearns from data

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
Scoring model based on machine learning (scoringsmodel baseret på machine learning) for reversals of withheld income tax (A-skat) in eIndkomst
Made by
Skattestyrelsen (Danish Tax Agency)
Used by
Skattestyrelsen (Danish Tax Agency)
Where
Denmark (national)
Stage
In routine use
Date
Kind
Aggregate result
Broad benefit
Evaluating Information to Determine Compliance with Standards, Organizing, Planning, and Prioritizing Work
Last revised
31 October 2026

Sources

  1. Descriptive report, by authors with no tie to the system.

    Rigsrevisionen (National Audit Office of Denmark). Beretning om Skatteministeriets kontrol med A-skat (beretning nr. 4/2020), signed 12 November 2020 and submitted to the Folketing with the Statsrevisorerne's remarks of 20 November 2020. Chapter 4.3, paragraphs 67-68, printed p. 40 (PDF p. 47). Rigsrevisionen relays the Danish Tax Agency's preliminary results, drawn from the agency's memo on the scoring model and its results by company.

    Ud af de 222 virksomheder med en score på 0,5 eller derover havde Skattestyrelsen ultimo 2019 kontrolleret 74 virksomheder, der samlet havde tilbageført A-skat for 8,5 mio. kr. De foreløbige resultater viste en træfprocent på 47 % og et provenu på 5,3 mio. kr.

  2. Descriptive report, by authors with no tie to the system. Context or contrary evidence.

    Rigsrevisionen. Opfølgning i sagen om Skatteministeriets kontrol med A-skat (beretning nr. 4/2020). Fortsat notat til Statsrevisorerne, RN 1108/24, 4 July 2024, paragraph 20. Rigsrevisionen relays how the agency describes its data-mining models.

    Den ene model leder efter risikofyldte, positive angivelser af A-skat (fx fiktive indberetninger), mens den anden leder efter negative angivelser, hvor der kan være tale om uberettigede tilbageførsler af A-skat.

Details

AI system

Workflow

The model scored companies that reversed withheld income tax in eIndkomst on a scale from 0 to 1. Companies with a score of 0.5 or above had to be picked for control. Those below 0.5 could be picked. The agency checked the picked companies for fictitious and erroneous reversals.

Money

Kind of value

Revenue (provenu) from the agency's checks of 74 companies that the model scored 0.5 or above, by preliminary results at the end of 2019

Amount

5,3 mio. Kr.

Recipients and payers

The Danish state

Follow-up

Conditions for replication

The agency must check the companies that the model flags. By the end of 2019 it had checked 74 of the 222. Misuse or fraud must be present among the flagged reversals. The agency judged that most reversals are ordinary corrections of errors.

Entry details

Sector

Tax administration.

Purpose

The Danish Tax Agency built the scoring model to find fictitious and erroneous reversals of withheld income tax and to score companies from 0 to 1 for control.

38 more questions not yet checked

model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, currency and prices, period and unit, realization, gross or net, included costs, comparison, method, return and payback, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.

Cite this record

Peter H. Vartanian. (2026). Checks of firms flagged by a Danish tax model find misuse in about half (AIBD-0924). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-0924.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)

BibTeX Data Field definitions