What happened
Pôle emploi, now France Travail, has run La Bonne Boîte since 2015. The platform predicts which firms are likely to hire in each occupation, and France's Cour des comptes says that its predictions rest on machine learning. Five economists, two of them at the Paris School of Economics, designed a model that chose which firms to recommend to each job seeker. Between 19 November and 4 December 2019, about 500,000 registered job seekers received emails linking to recommended firms. About 300,000 others formed the control group. In an IZA discussion paper of February 2024, the authors report that the emails raised job finding in short-term contracts by 1%, with a p-value of 0.09. They found no statistically significant effect on job finding across all contracts. Applications to recommended firms were 2.7 times as efficient as applications to the average firm. The authors designed the recommendations that they evaluated. They followed hiring for four months, until the Covid-19 lockdown.
Sources
-
Audit report, by authors with no tie to the system. Context or contrary evidence.
Cour des comptes. France Travail et l'intelligence artificielle. Observations définitives S2025-1558, adopted 15 October 2025 and published 8 January 2026, section 1.1.2, p. 20. Describes La Bonne Boîte's predictions without assessing the experiment.
permet d'identifier les entreprises à fort potentiel d'embauche, selon une analyse prédictive reposant sur un module d'apprentissage automatique (machine learning)
-
Report, by the makers.
Behaghel L, Dromundo S, Gurgand M, Hazard Y, Zuber T. The Potential of Recommender Systems for Directing Job Search: A Large-Scale Experiment. IZA Discussion Paper No. 16781, February 2024. Abstract, p. 1. The authors designed the recommendations with the support of Pôle emploi and La Bonne Boîte, with funding from ANR grant ANR-17-EURE-0001.
Our intervention induces a 1% increase in job finding rates for short term contracts.
Details
AI system
- Workflow
-
La Bonne Boîte predicted hiring for each firm and occupation from administrative data on all French firms. The authors' model chose recommendations to maximize total expected employment, allowing for congestion at each firm. Job seekers received two or four emails, each with links to at most two recommended firms.
Follow-up
- Conditions for replication
-
Job seekers must apply to the recommended firms, and the authors estimate that the emails raised application rates by only one percentage point. Recommendations must favor firms with high predicted hiring, since applications to them succeeded far more often.
Entry details
- Sector
-
Public employment services (job matching).
- Purpose
-
Pôle emploi built La Bonne Boîte to direct job seekers' unsolicited applications to firms that it predicts will hire. The study's authors added a model that chose which firms to recommend to each job seeker.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). La Bonne Boîte tips lift short-term hiring a little in France (AIBD-1152). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1152.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)