We introduce a novel methodology for adaptive targeted experiments. Our Tempered Thompson Algorithm balances the goals of maximizing the precision of treatment effect estimates and maximizing the welfare of experimental participants. A hierarchical Bayesian model allows us to adaptively target treatments at different groups. We implement our methodology in a field experiment. We examine the impact of three interventions designed to improve formal employment outcomes of Syrian refugees and local jobseekers in Jordan: one treatment to address liquidity constraints, one to address information frictions, and one to address challenges of self-control. Six weeks after being offered treatment, none of the interventions has a significant or meaningful impact on the probability that individuals are in wage employment; we estimate that our targeting algorithm had a positive but small effect on aggregate employment (approximately 1 percentage point). However, we find large employment effects of all treatments for refugees at the two-month follow-up, and suggestive evidence of four-month impacts for the cash grant; liquidity appears to be a key barrier to employment for refugees.