Abstract:  In Welfare-to-Work programs, profiling the job-finding probability of unemployed workers is essential for directing them to the most appropriate policy instruments. An optimal profiling strategy is achieved by integrating dynamic learning about workers' skills within a principal-agent framework. The paper outlines the optimal Welfare-to-Work program for each worker, detailing aspects such as policy referrals, unemployment benefits, wage taxes post-reemployment, and the accuracy and timing of profiling. These factors are tailored based on the worker's expected skills and the generosity of the program. Implementing this optimal profiling strategy in the United States is estimated to generate average per-capita net present savings of $19,677.