Piloting a Machine Learning-Based Job-Matching Algorithm

The objective of this note is to present and discuss the findings of piloting a task-based job matching tool developed by the World Bank and implemented in partnership with the Regional Labor Office of Pomerania, Poland. The aim of the pilot was to assess whether simple ML-based tools could contribute to improve the efficiency of PES delivery and job-seeking behaviors compared to rule-based, knowledge-driven approaches. By combining labor demand data from local occupational barometers and the descriptions of tasks in the national taxonomy of occupations, the tool provides jobseekers a menu of potential jobs available in the local labor markets that match the tasks performed in previous work experiences. Results show that jobseekers were satisfied with the proposed occupations resulting from the tool (as beyond their thinking) and had the intention to expand job search efforts, though job-seeking behaviors could not be monitored. Career advisers recognized that the lack of information on jobseekers' education, skills, and preferences limited the efficiency of the proposed job matches