Abstract
Generative AI can substantially raise individual productivity, but whether these gains translate into higher firm performance depends on how AI reshapes work within organizations. This project studies generative AI as an organizational technology through a randomized rollout of a retrieval-grounded legal-AI tool specialized to the country-specific case law, across ~300 small-to-medium-sized law firms. Firms are randomized to delayed access, partial access for approximately half of their lawyers stratified by seniority, or firm-wide access. The design identifies direct effects on users, spillovers onto coworkers who do not receive the tool, and firm-level effects of adoption within the same organizations. It also tests whether effects differ between junior and senior lawyers. To our knowledge, we provide the first causal design that separately identifies direct effects, coworker spillovers, and firm-level output within the same organizations.
The study addresses a central tension in AI adoption within firms: AI may free senior workers from routine tasks and generate organizational gains, but imperfect AI output may instead create verification burdens that attenuate productivity gains. These forces have important implications for how AI affects the distribution of work and demand for junior professionals. If specialized AI allows small firms to overcome constraints in scarce professional expertise, it could lower the cost and expand the availability of these services. If effective adoption instead depends on existing organizational capacity and senior expertise, technology could reinforce productivity differences across firms. The results will provide evidence on whether AI can help small professional-services firms in developing countries overcome constraints to productivity and scale, and on how the gains from adoption are distributed across firms and workers.