EST-003

OTT (Otsustustugi) — AI Decision-Support Tool for Unemployment Service Tailoring

Download PDF
Estonia Europe & Central Asia High income Full Production Deployment Confirmed

Estonian Unemployment Insurance Fund (EUIF)

At a Glance

What it does Prediction (including forecasting) — Decision support for eligibility and benefits
Who runs it Estonian Unemployment Insurance Fund (EUIF)
Programme OTT (Otsustustugi) Decision-Support Tool
Confidence Confirmed
Deployment Status Full Production Deployment
Key Risks Operational and system integration risks
Key Outcomes Awarded 'Best Data-Based Digital Service' in Estonian public sector (2021).
Source Quality 6 sources — Report (multilateral / development partner), News article / media, Other, +2 more

OTT (Otsustustugi, meaning 'decision support' in Estonian) is a machine-learning-based decision-support tool deployed by the Estonian Unemployment Insurance Fund (EUIF) to assist employment counsellors in configuring and tailoring service packages for unemployed clients. The system was developed through a collaboration between the EUIF, the Centre for IT Impact Studies (CITIS) and the Estonian Centre for Applied Research (ECePS) at the University of Tartu, software implementation partner Nortal, and data warehouse provider Resta. Development began in 2018, with the system integrated into EUIF's IT infrastructure in 2019, pilot testing conducted across five branch offices from mid-2020, and full operational deployment across all EUIF offices achieved by October 2020.

The core function of OTT is to calculate two key predictions for each registered unemployed person: the probability of finding employment within 180 days, and the estimated risk of becoming unemployed again after re-employment. Based on these predictions, the system segments clients into risk categories corresponding to recommended service intensity levels — high, medium, or low support — and suggests appropriate service types such as retraining, reskilling, language courses, digital skills training, or work capacity re-evaluation. The tool also advises on the appropriate service channel, recommending either in-person counselling or digital self-service depending on the client's risk profile and circumstances. This segmentation enables counsellors to prioritise clients who are furthest from the labour market and require the most intensive intervention packages.

Technically, OTT employs supervised machine learning, initially using a Random Forest model which was subsequently switched to gradient-boosted trees in 2022. The model analyses approximately 45 indicators for each unemployed person, drawing on five years of historical data aggregated from multiple national registries via Estonia's X-Road secure data exchange layer. These indicators include employment and unemployment histories, socio-demographic characteristics (age, gender, family status, residence location), education level and qualifications, ICT and digital literacy indicators, benefit receipt history (unemployment insurance, social assistance), risk-group flags (disability, long-term unemployment, youth, persons aged 50+), regional labour market indicators (vacancy rates, sector composition), and prior EUIF service participation and outcomes. Data sources include the Tax Office (salary records), the Board of Education (qualifications), the Social Welfare Board (subsidy information), and the Health Insurance Fund, among others. The model is retrained quarterly to reflect changing labour market conditions, and individual risk scores are refreshed nightly to incorporate new information. The system processes records from over 100,000 clients.

OTT is integrated into EUIF's primary case management system, EMPIS-2, through which counsellors access the tool's recommendations. The system provides counsellors with visibility into which factors most strongly influence each individual's risk score, supporting a degree of interpretability. Crucially, OTT operates in an advisory capacity only: employment counsellors (referred to as 'specialists' within EUIF) retain full decision-making authority over service allocation. Counsellors must review OTT advisory scores, can override recommendations based on professional judgement and qualitative information not captured in administrative data, and are required to record accuracy feedback on whether OTT recommendations were appropriate. This feedback loop enables ongoing model improvement. Management dashboards aggregate risk distributions across the client population for workload planning purposes but do not automate individual service decisions.

Research by Vihalemm et al. (2025) documented the implementation experience of specialists and middle managers at EUIF, finding a mixed reception. While EUIF leadership and some counsellors reported that OTT improved workload planning and helped identify appropriate interventions more quickly, other specialists reported that they seldom used the tool in daily practice, citing difficulties understanding the cause-effect relationships in the model and requesting more comprehensive training. Some counsellors expressed scepticism about the tool's recommendations and found certain features unclear. The study highlighted tensions between the formal obligation to provide feedback on OTT scores and specialists' actual engagement with the system.

The Nortal case study reports a 95% forecast accuracy rate and notes that reducing unemployment by a single day could save 3.8% of Estonia's yearly unemployment expenditures. In 2021, OTT was awarded 'The Best Data-Based Digital Service' in the Estonian public sector. Feedback statistics indicate that 93% of welfare specialists consider OTT scoring to be accurate. The system has been recognised internationally as a public-sector AI exemplar through platforms including the ITU AI for Good initiative and e-Estonia briefings. However, no published causal impact study has quantified employment outcomes directly attributable to OTT.

OTT operates under the EU General Data Protection Regulation (GDPR) framework and Estonian national data protection law, including the Estonian Administrative Procedure Act requirements for reasoned decisions. The Unemployment Insurance Act authorises EUIF's use of automated decision-making processes. The system does not make automated eligibility, payment, or sanctioning decisions — all such decisions require human counsellor approval. A 2021 maintenance contract with CITIS at the University of Tartu provides for ongoing model development and retraining. The EUIF has also explored extending predictive capabilities to identify employed persons at risk of future unemployment, with pilot plans for preventive services targeting the currently employed population.

The system runs on Estonian government infrastructure, with data retrieved from national registries via X-Road. There is no evidence of external or non-sovereign hosting arrangements. The development involved a mix of in-house EUIF capacity and third-party partners (University of Tartu for model development, Nortal for software integration, and Resta for data warehousing).

Classifications follow the DCI AI Hub Taxonomy. Hover over field labels for definitions.

Social Protection Functions

Implementation/delivery chain
Profiling, job matching and support services primaryAssessment of needs/conditions + enrolment Case management
SP Pillar (Primary) The social protection branch: social assistance, social insurance, or labour market programmes. Labour market programmes
SP Pillar (Secondary) The social protection branch: social assistance, social insurance, or labour market programmes. Social insurance
Programme Name OTT (Otsustustugi) Decision-Support Tool
Programme Type The type of social protection programme, classified under social assistance, social insurance, or labour market programmes. View in glossary Public employment services
System Level Where in the social protection system the AI is applied: policy level, programme design, or implementation/delivery chain. View in glossary Implementation/delivery chain
Programme Description AI-enabled decision-support tool within the Estonian Unemployment Insurance Fund that uses supervised machine learning to predict employment probabilities and unemployment recurrence risk, segment clients by risk level, and recommend tailored service packages to employment counsellors.
Implementation Type How the AI output is produced: Classical ML, Deep learning, Foundation model, or Hybrid. Affects validation, compute requirements, and governance profile. View in glossary Classical ML
Lifecycle Stage Current stage in the AI lifecycle, from problem identification through to monitoring, maintenance and decommissioning. View in glossary Monitoring, Maintenance and Decommissioning
Model Provenance Origin of the AI model: developed in-house, adapted from open-source, commercial/proprietary, or accessed via third-party API. View in glossary Developed in-house
Compute Environment Where the AI system runs: on-premise, government cloud, commercial cloud, or edge/device. View in glossary Not documented
Sovereignty Quadrant Classification of data and compute sovereignty: I (Sovereign), II (Federated/Hybrid), III (Cloud with safeguards), or IV (Shared Innovation Zone). View in glossary I — Sovereign AI Zone
Data Residency Where the data used by the AI system is stored: domestic, regional, or international. View in glossary Domestic
Cross-Border Transfer Whether data crosses national borders, and if so, whether documented safeguards are in place. View in glossary None
Decision Criticality The rights impact of the decision the AI supports. High criticality requires HITL oversight; moderate requires HOTL; low may operate HOOTL. View in glossary Moderate
Human Oversight Type Level of human involvement: Human-in-the-Loop (active review), Human-on-the-Loop (monitoring), or Human-out-of-the-Loop (periodic audit). View in glossary HITL
Development Process Whether the AI system was developed fully in-house, through a mix of in-house and third-party, or fully by an external provider. View in glossary Mix of in-house and third-party
Highest Risk Category The most significant structural risk source identified: data, model, operational, governance, or market/sovereignty risks. View in glossary Operational and system integration risks
Risk Assessment Status Whether a formal risk assessment, informal assessment, or independent audit has been conducted for this system. Informal assessment

Impact Dimensions

  • Grievance mechanism
  • Human oversight protocol
  • Independent evaluation
CategorySensitivityCross-System LinkageAvailabilityKey Constraints
Administrative data from other sectorsSensitiveLinks data across multiple systemsCurrently available and usedData aggregated from Tax Office (salary/employment histories), Board of Education (qualifications), Social Welfare Board (subsidy records), and Health Insurance Fund via X-Road. Approximately 45 indicators over 5-year rolling window. Data quality depends on accuracy and completeness across multiple registries.
Beneficiary registries and MISPersonalLinks data across multiple systemsCurrently available and usedEUIF case management system (EMPIS-2) containing unemployment registration, service participation history, and benefit receipt records. Primary operational data source for the model.
Social registriesPersonalLinks data across multiple systemsCurrently available and usedSocio-demographic data including age, gender, family status, residence location, risk-group flags (disability, long-term unemployment, youth, 50+), and regional labour market indicators.

AI-FORA (n.d.). 'Country Case Study Estonia.' Mainz: Johannes Gutenberg University Mainz. Available at: https://www.ai-fora.de/estonia/

View source Report (multilateral / development partner)

AlgorithmWatch (2020). 'Estonia — Automating Society Report 2020.' Berlin: AlgorithmWatch. Available at: https://automatingsociety.algorithmwatch.org/report2020/estonia/

View source Report (multilateral / development partner)

Bushell, J. (2021). 'How Estonia is using AI to tackle unemployment.' Tech Monitor. Available at: https://techmonitor.ai/technology/ai-and-automation/how-estonia-using-ai-tackle-unemployment

View source News article / media

Nortal (2022). 'Estonian Unemployment Insurance Fund Prevents Unemployment with Artificial Intelligence.' Tallinn: Nortal AS. Available at: https://nortal.com/insights/estonian-unemployment-insurance-fund-prevents-unemployment-with-artificial-intelligence

View source Other

Vihalemm, T., Manniste, M., Trumm, A. and Solvak, M. (2025). 'Specialists and Algorithms: Implementation of AI in the Delivery of Unemployment Services in Estonia.' In: Participatory AI in Public Social Services. Cham: Springer. Available at: https://link.springer.com/chapter/10.1007/978-3-031-71678-2_5

View source Academic journal article

e-Estonia (2021). 'AI to Help Serve the Estonian Unemployed.' Tallinn: e-Estonia Briefing Centre. Available at: https://e-estonia.com/ai-to-help-serve-the-estonian-unemployed/

View source Government website / press release
Deployment Status How far the system has progressed into real-world operational use, from concept/exploration through to scaled and institutionalised. View in glossary Full Production Deployment
Year Initiated The year the AI system was first initiated or development began. 2018
Scale / Coverage The scale and geographic or population coverage of the deployment. Fully deployed across all EUIF branch offices nationwide since October 2020; processes records from over 100,000 clients
Technical Partners External technology vendors, academic partners, or development partners involved. University of Tartu (CITIS/ECePS) — model development and research partnership; Nortal — software implementation and system integration; Resta — data warehouse infrastructure. 2021 maintenance contract with CITIS for ongoing model development and retraining.
Outcomes / Results Awarded 'Best Data-Based Digital Service' in Estonian public sector (2021). 95% forecast accuracy reported by Nortal. 93% of welfare specialists consider OTT scoring accurate. Qualitative reports indicate improved workload planning and ability to prioritise high-risk clients. Recognised internationally via ITU AI for Good and e-Estonia briefings. Nortal estimates reducing unemployment by one day could save 3.8% of yearly expenditures. No published causal impact study quantifying employment outcomes attributable to OTT.
Challenges Vihalemm et al. (2025) found mixed specialist reception: some counsellors reported seldom using OTT in daily practice, citing difficulty understanding cause-effect relationships in the model. Specialists requested more comprehensive training beyond the 22-page manual. Some expressed scepticism about recommendations. Tension between mandatory feedback obligation and actual engagement with the tool. Initial model design pivot required — first iteration predicted long-term unemployment probability but proved 'difficult and a bit useless'; redesigned to predict job-finding probability.

How to Cite

DCI AI Hub (2026). 'OTT (Otsustustugi) — AI Decision-Support Tool for Unemployment Service Tailoring', AI Hub AI Tracker, case EST-003. Digital Convergence Initiative. Available at: https://socialprotectionai.org/use-case/EST-003 [Accessed: 1 April 2026].

Change History

Updated 1 Apr 2026, 08:11
by system (system)
Created 30 Mar 2026, 08:39
by v2-import (import)