86 occupations. AI exposure. 2030 forecast.

The French forecast uses the same engine as the Dutch, UK, US and German maps (JPE-prognose v3). It starts from the national occupational projection and adds AI and robotics as separate layers. The design was fixed on 27 September 2026, before the data was recalibrated, before the backtest and before the blind check. The JPE score per occupation did not change.

Data

• Classification: each occupation is mapped to one or more FAP 2021 families (DARES). The old codes on this map were not valid PCS 2020 codes and have been replaced.
• Employment: DARES, Portraits statistiques des métiers, 2024 (Enquête Emploi, all workers including self-employed and civil servants), summed over the families of an occupation and rounded to 1,000. The series 2015-2024 is carried across the FAP 2009 to FAP 2021 break by DARES.
• Growth: 2021 to 2024 as shown on the pages; 2015 to 2024 as context in the dataset.
• Salary: DARES median net monthly salary 2023-2025 × 12, for employees only. It is not converted to gross, so it is not comparable with the gross salaries on the other maps. Where DARES publishes no median for a family, the median of the parent level is used and labelled.
• Self-employed: where 40% or more of the workers are self-employed, the page notes that the salary covers employees only.
• Shared family: Consultants en management and Contrôleurs de gestion have no own DARES family (both fall inside L5X90, administrative and financial managers). They keep a page with an estimated job count, but are not counted in totals, averages or the backtest.
• Jump warning: where the DARES figure changed by 25% or more between 2022 and 2024, the page flags the growth figure and the forecast as extra uncertain. This affects display only.

Model

• Base: France Stratégie / DARES, Les métiers en 2030 (reference scenario) (March 2022), net job creation 2019-2030 per FAP 87 family. The annual rate is applied to 2025-2030. Where an occupation spans several families, the rate is weighted by jobs in 2019.
• Long-term decliners: where DARES employment fell both from 2013 to 2019 and from 2021 to 2024 (17 occupations), the base is blended half and half with the occupation's own trend (the average annual rate over both periods), as in the Netherlands and Germany.
• AI layer: −0.25 percentage points per year for every JPE point above 4, over five years. Les métiers en 2030 already contains digitalisation losses (administration, banking, accounting), but it was made on a 2019 base, before generative AI; the AI layer measures the extra substitution since.
• Robotics layer: the midpoint of an expert range per occupation, only when physical automation starts scaling up in France before 2030 (for example self-checkouts, warehouse automation, cleaning and agricultural robots). Robotaxis and autonomous trucks are not expected to scale up in France before 2030.
• Forecast: (1 + base) × (1 + AI) × (1 + robotics) − 1, the midpoint.
• Band: base ±3 percentage points (an assumption), combined with the favourable and unfavourable AI slope (−0.1 to −0.5) and the robotics range, rounded outward to 5 percentage points. The band matters more than the midpoint.

Trend distribution: 22 growth, 44 stable, 18 decline and 2 strong decline. No occupation is expected to change by more than ±40%.

Backtest

Les métiers en 2030 (annual rate 2019-2030) against actual DARES growth from 2019 to 2024 for the families of each occupation (84 occupations; error in percentage points per year, weighted by jobs in 2019; paired bootstrap, 2,000 draws).

The new model is significantly more accurate than the old one (difference −0.49 pp/yr, 95% interval −0.84 to −0.13) and not significantly worse than the macro benchmark (−0.21, interval −0.52 to +0.08). Caveat: the projection appeared in March 2022, so 2019-2021 was partly known at publication; the base itself is 2019.

Sensitivity

• Occupations expected to shrink by more than 3%: 17 with an AI slope of −0.10, 20 with −0.25 (used) and 31 with −0.50.
• Without the rule for long-term decliners the base shifts for 17 occupations, most for pharmacists (−1% to +6%), laboratory technicians, electricians and carpenters; for bank clerks and secretaries the base becomes milder.
• Double-counting risk: for 8 occupations where Les métiers en 2030 already expects a decline of 10% or more, the forecast without the AI layer is 1 to 5 points higher (for example accounting clerks −14% to −9%, secretaries −18% to −14%, bank clerks −19% to −15%).

Blind check

15 randomly chosen occupations (seed 20260927), judged against external French sources: 11 plausible, 4 doubtful, 0 unlikely. Same pattern as in the US and Germany: strong, persistent trends are pulled towards zero (farmers and secretaries probably lower, childcare assistants higher). The model was deliberately not changed afterwards (preregistration).

Limitations

• Base 2019: Les métiers en 2030 is seven years old and predates generative AI and the post-Covid labour market. Its successor (horizon 2035) had not been published on 27 September 2026.
• FAP 87: the projection differentiates per FAP 87 family (FAP 2009). Twenty FAP 2021 families have no one-to-one match (construction, metalworking, process industry, banking and insurance, administration); occupations within one family get the same base.
• Net salary: salaries are net medians for employees; they exclude the self-employed and are not comparable with the gross salaries of the other countries.
• Broad proxies: several occupations cover more than the name suggests, for example lawyers (lawyers and in-house legal staff), refuse collectors (all urban maintenance), estate agents (including letting and property management), government administrative officers (PCS 2020 no longer separates civil servants) and construction tradespeople and retail shop owners (FAP has no employment status). The page of each such occupation carries a note.
• Small cells: for small occupations such as judges the survey count is noisy; for judges no growth, series or salary is shown and the occupation is not ranked.
• No salary figure: for 7 occupations DARES publishes no median for any of their cells; no amount is shown and they are left out of salary averages.
• Childcare assistants: the base uses the Métiers 2030 row for childminders (about −1%); in the older classification childcare assistants were counted with care assistants (about +15%), so their forecast is probably too low.
• Laboratory technicians: the long-term decline rule is triggered by a break in the DARES series before 2019, while employment grew about 40% from 2019 to 2024 (partly Covid testing). Their expected decline is uncertain.
• Reception and customer service: receptionists, customer advisers and secretaries get a Métiers 2030 rate that also weighs in the public-sector administrative family (P0Z), which is projected to shrink more than the private-sector rows.

The general JPE methodology (scoring rubric, Felten AIOE validation) is on the methodology page. Design document (Dutch): docs/jpe-prognose-v3-fr.md.