evidoria

← Back to browse

Good practice Imported

Next-Step — VDAB's Machine-Learning Profiling of Jobseekers' Re-Employment Risk (Flanders, Belgium)

Belgium · Brussels · See the Belgium profile · See the Brussels profile

Evidence: Descriptive / self-reported Top 50% 60/100 · Ask Evidence Copilot about this practice

Since ~2018, Flanders' employment service VDAB has used a Random Forest ML model, "Next-Step", to predict jobseekers' re-employment odds and route those at highest long-term unemployment risk to intensive caseworker support, screening new registrants within six weeks.

Details

Maturity
Established
Promoter
VDAB (Vlaamse Dienst voor Arbeidsbemiddeling en Beroepsopleiding) — Flemish Public Employment Service
Period
2018–present
Region (NUTS)
BE2
Keywords
public employment services, algorithmic profiling, labour-market forecasting, explainable AI, jobseeker support

Context

Since approximately 2018, VDAB (Vlaamse Dienst voor Arbeidsbemiddeling en Beroepsopleiding), the public employment service for Belgium's Flanders region, has used a machine-learning profiling tool internally called 'Next-Step' to estimate each new jobseeker's probability of finding work within a given time horizon.

Objectives

The tool aims to classify all new jobseekers within six weeks of registration, using predicted re-employment probability to prioritise caseworker time toward those assessed as least likely to find work unaided, rather than spreading support evenly across all registrants.

Activities

The underlying model is a Random Forest trained on jobseekers' registration and labour-market history data. VDAB has presented the model's design at EU-level forums such as Cedefop, and peer-reviewed research has applied explainable-AI techniques such as TreeSHAP to VDAB-linked profiling data to assess interpretability and fairness.

Conclusions

Unlike comparable profiling systems in Poland (ruled unconstitutional and discontinued in 2019) and Austria (blocked by the data-protection authority before nationwide rollout), Next-Step has remained in continuous operational use, but no independently audited accuracy or misclassification figures specific to it have been published.

Implementation
Indicative cost
Medium (€50k–€500k) — In-house Random Forest profiling model integrated into existing VDAB registration workflow; no public cost figures found in source.
Time to results
Long (> 3 years) — Operational continuously since approximately 2018 to present.
Staffing & skills
VDAB caseworkers applying model outputs to prioritise support, VDAB data/technical team maintaining the Random Forest model

Conditions for success

  • Screening embedded into standard registration workflow (classification within six weeks)
  • Continuous operation since 2018, avoiding the legal/constitutional challenges faced by comparable Polish and Austrian systems
  • Openness to external academic scrutiny (TreeSHAP-based fairness/interpretability research) and public presentation of methodology at EU forums (Cedefop)

Common failure modes

  • No independently audited accuracy or misclassification-rate figures specific to Next-Step have been published
  • Comparable systems elsewhere (Poland, Austria) were discontinued or blocked over legal and fairness concerns

Where it fits

Governance type
regional public employment service
Scale
regional (Flanders, Belgium — all new jobseeker registrants)
Income level
high-income (Belgium)

Commonly funded by

National / regional programmes ESF+ — European Social Fund Plus

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful