Fifty-seven billion euros a year to train French workers. On paper, it is one of Europe’s most generous budgets. In practice, this system concentrates training on those who already have the most access to it, and leaves without retraining those for whom it would be decisive. The choice that 2027 will have to make is this: change who decides how funds are allocated, or continue financing the demand of those who know how to formulate it.
A manager trains twice as often as a factory worker
A manager is twice as likely as a factory worker to receive training in any given year [2]. Access rates to non-formal vocational training are 68% for managers, compared to 33% for factory workers [2]. These figures return with each survey cycle, unchanged in their structure.
Older workers pay the same price. Among those aged 55-64, only 24% completed non-formal vocational training in the year; among those in employment in this age bracket, the rate rises to 39%, but remains far below the 54% reached among those aged 35-44 [2]. Yet these are the workers whose skills risk becoming most outdated, and whose professional horizon would precisely require such reinvestment.
The EFE-e survey conducted by Céreq, Dares and France compétences over 2005-2020 shows that the share of training companies and the proportion of trained employees each year did not change over fifteen years [13]. Several major reforms (2009, 2014, 2018) passed through this space without shifting it.
Skills stagnate, gaps widen
The OECD’s PIAAC survey, conducted in 2022-2023 across 31 countries and published in December 2024, provides the most rigorous benchmark available [3] [4]. In France, more than one adult in four shows weak proficiency in literacy, numeracy and problem-solving using digital tools [3]. France sits slightly below the OECD average in these two domains.
What strikes more than the level itself is the immobility. The performance observed in 2023 changes little compared to the first PIAAC survey conducted in 2012 [3]. In eleven years of reforms, average performance stagnated. The gaps, meanwhile, widened: in numeracy, those aged 55-65 are twice as likely in 2023 to show weak proficiency as those aged 16-24; in literacy, three times as likely [3].
The length of this blockade says something about its depth. In 1998, manager access to vocational training was already around 68%, compared to 37% for factory workers [5]. The inequality has been in place for a generation. All the financing reforms have not changed its structure by a millimeter.
Billions spent where the market signal was already favorable
France compétences, the body created in 2018 to steer the system, accumulated deficits from its earliest years. Its deficit reached 4.6 billion euros in 2020 [7]. In 2022, without exceptional state subsidies (4 billion in total), it would have reached 7.4 billion [7]. Spending was exploding: not because less-qualified people were being trained better, but because apprenticeship at bachelor’s degree level and above, and use of the personal training account (CPF), had taken off to benefit already well-established workers [15].
Apprenticeship rose from 360,000 entrants in 2019 to 880,000 in 2024, exceeding one million apprentices in training [1]. This success is real for the job placement of young people who access it. But it was financed on credit, and its profile shifted toward higher levels, where the market signal was already favorable. France compétences’ cumulative deficit reached 9 billion euros by end-2023, despite state subsidies of 4 billion in 2022 and then 1.6 billion in 2023 [7].
Facing this budgetary wall, the response was to compress spending uniformly. The 2026 budget provides for approximately 12 billion euros in spending, roughly 1.5 billion less than in 2025, a decline of more than 10% in one year [6]. What recedes first are the tools that less-qualified people could theoretically activate.
CPF spending came close to 3 billion euros in 2021; it fell below 2 billion in 2025, and the 2026 budget provides for 1.32 billion [6] [8]. The amount will have been cut by more than half in four years. These are skills assessments and short training courses that are disappearing: the only tools accessible to someone just starting out.
AI is attacking exactly where the system concentrates its resources
Until now, the system rested on an implicit assumption: graduates need continuing training because their professions evolve quickly; less-qualified people need initial training to enter the job market. Agentic AI breaks this assumption from above.
It affects disproportionately cognitive and non-routine occupations that call on judgment and interpretation, as of 2026 [9]. These are the office jobs of graduates (law, finance, accounting, human resources, information technology) that previous waves of automation had spared. According to the Coface study co-authored by Axelle Arquié and Aurélien Duthoit, published in April 2026, one job in six, or roughly 5 million jobs in France, will be automatable within three to four years [9].
This reversal has a budgetary consequence that Daron Acemoglu has documented [10] [11]. AI targets the best-paid occupations, those that pay the most taxes and social contributions. It therefore exposes public finances to a scissor effect: tax revenues decline just as training and unemployment spending increases. A training system concentrating its resources on the already-qualified loses its reason for being at the precise moment when these workers are most exposed to substitution. And it has built nothing for those who needed retraining urgently.
The tax structure aggravates the imbalance. Labor is taxed far more heavily than the intangible assets destined to automate it. Acemoglu and Johnson argue for fiscal symmetry: equalizing the marginal tax rates on hiring and training with those applied to software and algorithms, in order to encourage companies to choose complementarity with humans rather than substitution [10].
Transfer fund allocation to territories and industry branches
Today, it is the demand of individuals and businesses that allocates funds. Those who know how to activate the CPF, who have an employer willing to finance, who have access to a career development advisor, get the training. There is no guarantee that training benefits those who need it most. The logic is that of a market without correction.
The architecture to be built is different. It transfers a significant portion of fund allocation power to branches and territories, linked to diagnoses of employment vulnerability by sector, by region and by qualification level. It makes a right to qualifying training opposable for any worker whose occupation is documented as exposed to automation. Bertrand Martinot formulated it: making vocational training more modular and accessible throughout working life requires adapting the system to a more diverse and changing professional world [12]. Modularity is not enough if it is always the same public that activates the modules.
This transfer requires an institution or a mandate given to branches and regions, with the authority and data to target training toward workers whose skills are becoming obsolete, before they become unemployed. It closes the freedom of undirected individual choice, which today benefits the best-informed. It opens a real right for those whom the market does not reach. Its cost is political as much as budgetary: it requires removing from competency operators (OPCO) a portion of their decision-making power, and imposing on branches a responsibility for diagnosis that they have never assumed alone.
Sources
[1] DARES, “Appendix to the 2026 Budget Bill: Yellow Budget on Vocational Training,” Ministry of Labor, 2025, https://www.assemblee-nationale.fr/dyn/dyn/contenu/visualisation/1089989/file/7-Jaune2026_Formation_Professionnelle.pdf (consulted 28/09/2026).
[2] DARES, INSEE, “More frequent recourse to training in 2022 for employed persons and those with higher qualifications,” INSEE Première no. 1994, April 2024, https://www.insee.fr/fr/statistiques/8177288 (consulted 28/09/2026).
[3] DARES, DEPP, INSEE, “Adult skills in 2023: what proficiency in finding and using information in the digital age?” DARES Analyses no. 73, December 2024, https://dares.travail-emploi.gouv.fr/sites/default/files/12a714bec043c83622cbeec936b360b3/Dares_DA_Piaac_.pdf (consulted 28/09/2026).
[4] OECD, “Survey of Adult Skills 2023: Country Note France,” December 2024, https://www.oecd.org/content/dam/oecd/fr/publications/reports/2024/12/survey-of-adults-skills-2023-country-notes_df7b4a60/france_ba42b6a4/21bacb82-fr.pdf (consulted 28/09/2026).
[5] INSEE, “Adult Training,” INSEE Première no. 1468, 2012, https://www.insee.fr/fr/statistiques/1281375 (consulted 28/09/2026).
[6] Senate, Special Reporters, “2026 Budget Bill: Labor, Employment and Administration of Social Ministries,” 2025, https://www.senat.fr/rap/l25-139-331/l25-139-33110.html (consulted 28/09/2026).
[7] France compétences / Certiforma, “France compétences Financial Situation: Vocational Training Financing 2026,” September 2026, https://certiforma.fr/france-competences-situation-financiere-financement-formation/ (consulted 28/09/2026).
[8] DARES, “Appendix to the 2025 Budget Bill: Yellow Budget on Vocational Training,” Ministry of Labor, 2024, https://dares.travail-emploi.gouv.fr/sites/default/files/Dares_Jaune2024_formation_professionnelle.pdf (consulted 28/09/2026).
[9] Axelle Arquié, Aurélien Duthoit (Coface), study on job automation by agentic AI, April 1, 2026; results cited by Le Devoir, “The massive effects of artificial intelligence on employment are yet to come,” March 24, 2026, https://www.ledevoir.com/economie/techno/966348/effets-massifs-ia-emploi-restent-venir (consulted 28/09/2026).
[10] Daron Acemoglu, Simon Johnson, “Rebalancing AI,” Finance & Development (IMF), December 2023, https://www.imf.org/fr/publications/fandd/issues/2023/12/rebalancing-ai-acemoglu-johnson (consulted 28/09/2026).
[11] Daron Acemoglu, Simon Johnson, “The Simple Macroeconomics of AI,” Economic Policy, vol. 40, no. 121, 2025, https://doi.org/10.1093/epolic/eiae042 (consulted 28/09/2026).
[12] Bertrand Martinot, Work is the Solution, CFE-CGC Orange seminar, July 2025, https://www.cfecgc-orange.org/2025072415407/nos-seminaires/le-travail-est-la-solution-bertrand-martinot-seminaire-ete-2025.html (consulted 28/09/2026).
[13] Céreq, DARES, France compétences, EFE-e survey (Employer Training Survey: European), 2005-2020 data, results cited by UNSA, October 2024, https://www.unsa.org/Developpement-des-competences-les-inegalites-d-acces-a-la-formation-persistent.html (consulted 28/09/2026).
[14] National Assembly, Amendment no. II-CF2504, PLF 2025, Finance Committee, 2024, https://www.assemblee-nationale.fr/dyn/17/amendements/0324C/CION_FIN/CF2504.pdf (consulted 28/09/2026).
[15] France compétences, “Report on Vocational Training Fund Use: 2025 Edition,” February 2026, https://www.francecompetences.fr/app/uploads/2026/02/RUF25_VF.pdf (consulted 28/09/2026).
