Out of 2,000 cases across the cantons between 2020 and 2023, the recommendation optimised using machine learning increased the three-year employment rate by 5.2 percentage points.
Between January 2020 and June 2023, the Swiss State Secretariat for Migration entrusted around 2,000 refugee cases to a controlled experiment: each case received a recommendation for the host canton, either calculated by an employment algorithm or drawn in such a way as to replicate usual practices. Neither the placement officers, who had the last word, nor the refugees knew which group they were in. The quotas by canton and origin were identical in both arms: the measured gain comes from a better match, not from a mass referral to the most dynamic labour markets.
Effects that grow over time
On the main indicator, the share of months spent in employment during the first three years, the average effect is +2.2 points (about 10% more than the control group, at 22.3%), and +3.9 points for the 2022-2023 cohorts, after the Covid shock. At 36 months, the employment rate increases by 5.2 points, an effect that the authors consider comparable to that of hundreds of hours of intensive language courses.
The study provides one of the few pieces of evidence on the ground that an AI-based decision support tool can improve high-stakes public allocation at a modest cost and at scale.
Read the full article on the original source
Go to the source →