Swedish risk model finds more errors in parental-benefit claims

Sweden's Social Insurance Inspectorate found that 55% of parental-benefit claims picked by a risk model in 2017 had errors, against 34% of claims checked at random.

  • 2 an organization

    ISF measured errors in 55% of the claims that the model picked, against 34% of claims checked at random.

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

Status
Reported benefit.
Evidence

foundstudiedreportedclaimed

An independent audit report supports the benefit.

Compared with

Claims picked at random for checks in 2017. Measured.

Without the system
34 procent
With it
55 procent
The system’s part
ISF compared claims that the model picked with claims picked at random in the same year, which isolates the model's effect on the hit rate. The model was built from register data and the errors found in random checks, and ISF writes that the agency builds such profiles with machine-learning methods.
Limits
Not yet checked.
Costs and harms
Not yet checked.
Later
Withdrawn. Försäkringskassan told the Authority for Privacy Protection that it had taken its machine-learning risk profile for these checks out of use. It did so about a month before its reply of 5 September 2025 and had no plans to bring the profile back.

Download data Cite Suggest a correction

Benefit type
Improved Task Accuracy
System
Försäkringskassan's risk-based selection profile (urvalsprofil) for checks of temporary parental benefit, run through the RaKuR infrastructure
Made by
Försäkringskassan (Swedish Social Insurance Agency), Urvalsanalys unit
Used by
Försäkringskassan (Swedish Social Insurance Agency)
Where
Sweden
Stage
In routine use
Date
Kind
Aggregate result
Broad benefit
Evaluating Information to Determine Compliance with Standards, Organizing, Planning, and Prioritizing Work
Collections
Scrutiny and findings
Last revised
31 October 2026

Sources

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

    Inspektionen för socialförsäkringen (ISF). Profilering som urvalsmetod för riktade kontroller: En granskning av träffsäkerheten, effektiviteten och rättssäkerheten i Försäkringskassans modeller för riskbaserade kontroller. ISF Rapport 2018:5, March 2018. Chapter 6, printed pp. 109-110, Tabell 7.

    Genom att välja ut vilka som ska kontrolleras mer systematiskt ökar alltså träffsäkerheten väsentligt.

  2. Regulatory document, by authors with no tie to the system. Context or contrary evidence.

    Integritetsskyddsmyndigheten (IMY). Beslut efter tillsyn enligt dataskyddsförordningen – Försäkringskassan. IMY-2025-11828, 18 November 2025.

    I Försäkringskassans svar till IMY, som inkom den 5 september 2025, uppgavs att Försäkringskassan har använt en maskininlärningsbaserad riskprofil inom ramen för det aktuella arbetet men att denna togs ur bruk en dryg månad före tidpunkten för Försäkringskassans svar.

Details

AI system

Workflow

The model computed each claim's probability of error from register data, reported data and the errors uncovered in random checks. The agency picked claims for checks by risk value, amount and expected future use of the benefit. Random checks continued beside the model and supplied errors to train it.

Follow-up

Conditions for replication

Random checks must continue to show the true distribution of errors. Selection must pass a test of equal treatment across groups.

Entry details

Sector

Social insurance control.

Purpose

Försäkringskassan built the risk-based selection profile to pick the temporary parental benefit claims most likely to contain errors for checks.

Not yet checked: model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, 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). Swedish risk model finds more errors in parental-benefit claims (AIBD-0713). Artificial Intelligence Benefits Database, edition of 31 October 2026. https://petervartanian.xyz/aibd/records/AIBD-0713.html (archived at https://doi.org/10.7910/DVN/AADSII, version 4.0)

BibTeX Data Field definitions