Efficient unemployment assistance is tailored to workers' human capital. Since human capital is difficult to infer, assistance is provided on the basis of its expected level. Alternatively, workers can be profiled and their actual level of human capital be detected. A profiling program establishes (i) whom to profile, (ii) at what stage of the program and (iii) what benefits to transfer to them, depending on the new information obtained. The paper identifies the determinants of optimal profiling along these three dimensions in a dynamic principal-agent framework with non-contractible effort and two-sided uncertainty about workers' human capital. There are two main findings. First, workers with higher expectations on human capital are incentivized to search for a job, thanks to larger returns on search effort. They are profiled only at a successive stage of the unemployment spell, once the gains from optimal matching between policies and workers outweigh the cost of profiling. Second, since the incentive cost is increasing in the generosity of benefits promised to workers, profiling is used also to lower promised benefits for those workers who are incentivized to search after it.