Paper presented at the 2016 Jobs & Development Conference, 2- 3 November. Non-cognitive (“soft”) skills – allocating time and money effectively, teamwork, leadership, relationship management, acquiring and assimilating information – account as much for long-term economic success as cognitive ability and educational attainment. But these skills may be very difficult to teach in adulthood, especially to those with low baseline skill sets. Moreover, firms may be reluctant to invest in workers’ skills if attrition rates are high, which is particularly the case for frontline workers. We carried out a randomized experiment with female garment workers in Bengaluru, India to test whether it is possible to impart soft skills to frontline workers, and evaluate the labor productivity, retention, and profitability consequences for firms. Treated workers are less likely to leave during the program, and exhibit substantially higher productivity up to nine months after program completion. This leads to being assigned to more complex tasks and a greater likelihood of promotion. Treated workers are also more likely to enroll in workplace skill development and production incentive programs. Survey evidence supports the hypothesis that the stocks of soft skills improved in key dimensions. Two-stage randomization allows us to estimate spillovers within production teams; spillovers in productivity are substantial and persistent. Using actual costing data we find that the program pays for itself several times over by the end of the evaluation period, implying that teaching soft skills in the workplace can be profitable for firms even in high-turnover environments.