Your Job Risk · Belgium
Methodology →

Methodology

How these numbers are built — and what they are not.

This page walks through every figure on the site: the province exposure scores, the occupation ranking, the layoff classifications, and the deliberate choices that shaped each one.

1. What "AI exposure" means here

Exposure is not displacement. A high score means a job is made up of tasks that today's AI systems (large language models, vision systems, tabular ML, process automation) can do part of the work for. Whether that leads to layoffs, augmentation, new jobs or nothing visible depends on firm strategy, institutions and policy.

The underlying construct is the AI Occupational Exposure (AIOE) index from Felten, Raj & Rock (2021), which links applications of AI to the O*NET abilities that an occupation relies on. Treat the 0–100 number as a relative ranking of cognitive task exposure, not a probability of job loss.

2. Occupation scores (ISCO-08 major groups)

The AIOE is defined at the US SOC occupation level. For Belgium the site works at the coarser ISCO-08 major-group level (1 through 9) because:

  • Statbel publishes Belgian employment at the ISCO major-group level consistently and openly.
  • Crosswalking each 4-digit SOC to 4-digit ISCO would introduce false precision for a narrative site like this.

For each ISCO major group, the score shown is an employment-weighted AIOE average of the SOCs that map into that group, rescaled to 0–100 so clerical support (the most exposed group) sits around 82 and agriculture around 22. The examples in each row are representative, not exhaustive.

The AIOE numbers are cross-checked against more recent task-exposure work that captures generative AI specifically — Eloundou et al. (OpenAI, 2023) on GPT exposure, the ILO's Generative AI and Jobs refreshes (2023 and the 2025 update), the OECD Employment Outlook 2023–2025 chapters on AI, the Anthropic Economic Index (2025), and newer firm-level evidence on junior employment at GenAI-adopting firms. They agree on the direction of the ranking, even where absolute scores differ.

3. Province exposure scores

Each province gets a single composite score by combining:

  1. Occupation mix: the share of that province's employed residents in each ISCO major group, from the Belgian Labour Force Survey (Statbel LFS, 2024 annual release).
  2. Occupation AIOE: the group-level scores from step 2.
  3. Sector adjustment: a small correction for province-level sector intensity (e.g. EU institutions in Brussels, chemicals in Antwerp, auto/logistics in Limburg, retail/services in Luxembourg province) drawn from Eurostat regional accounts.

The resulting score is rescaled to 0–100 so Brussels-Capital sits near the top (72) and Luxembourg province at the bottom (43). Two things to remember:

  • It reflects where workers live, not where they work. Cross-border commuting into Brussels is significant but not explicitly modelled.
  • The sector adjustment is indicative. I did not run this through Statbel's detailed regional sector tables or reweight by firm size — both would move scores by a few points.

4. Province totals and headline numbers

The ~5.0 M workers figure is the sum of modelled employed residents per province (rounded Statbel 2024 numbers), and sits within a few percent of the official Belgian employed population.

The "at elevated AI exposure" aggregate counts workers in ISCO groups scoring ≥ 60 on the occupation scale — professionals, managers, technicians and clerical support. That's a defensible threshold but it is a threshold: move it to 55 or 65 and the headline changes by roughly ±0.5 M.

5. Layoff classification (AI likely / plausible / non-AI)

Every restructuring announcement in the last 24 months is tagged with an editorial AI factor. The rubric is:

  • AI likely — the company explicitly cites AI, automation, digital transformation or back-office modernisation as a driver, and the roles being cut sit in occupation groups the AIOE flags as highly exposed. Example: Proximus' transformation plan, BNP Paribas Fortis branch + IT consolidation.
  • AI plausible — stated cause is primarily market, demand or regulation, but the profile of roles cut (clerical, HQ, R&D support) sits in the high-exposure zone. Example: bpost back-office, AGC Glass HQ / R&D, Pfizer commercial efficiency.
  • Non-AI — the restructuring is primarily a geopolitical, energy, demand or sector-specific shock, targeting roles (production, retail, plant) the AIOE does not flag. Example: Audi Brussels, Cora, Van Hool, BASF Antwerp, ExxonMobil, Villeroy & Boch, BAT.

These tags are judgements, not causal attribution. The counterfactual — "would this layoff have happened in a world without generative AI?" — is genuinely unknowable. Hover a tag on the homepage to see the one-line rationale.

6. What this site is not

  • Not a forecast. There is no Monte Carlo simulation, no 2030 / 2035 projection, no "X% of jobs will be lost" claim.
  • Not a policy tool. The figures are indicative, intended to frame a conversation, not to size a training budget or a regional transition plan.
  • Not commercial. There are no ads, trackers, email captures or affiliate links.

7. Sources

Core exposure construct:

Cross-checked against more recent task-exposure work:

Belgian labour and geography:

Layoff announcements (April 2024 – April 2026):

Last data refresh: April 2026. The AIOE itself has not been rewritten since 2021 — newer work (ILO, OECD, Anthropic) extends it rather than replacing it, so the ranking is stable; absolute levels should be read with a wider confidence band than the two-digit scores suggest.

8. Corrections

If a number is wrong, a layoff is missing, or a classification is unfair to a company, the code and data are in a single file (lib/data.ts) in the repository — open an issue or a pull request and it can be fixed in minutes.

← Back to the map