The ability of the state to collect tax revenue is a crucial indicator in the process of economic development. Nonetheless, we know relatively little about what drives tax compliance and how it can be improved in low-capacity countries. I test a text-based tax encouragement scheme in conjunction with the Uganda Revenue Authority, and find a 6x rate of return to this low-cost, easily implementable intervention. I use machine learning tools to estimate state tax capacity using an extensive, granular, nationwide dataset on public good provision, and assess impact heterogeneity across differences in pre-intervention capacity. I find that the treatment was most effective where state tax capacity was low, suggesting that the intervention may have functionally increased the presence of the state in such areas. Additional evidence implies that individuals are also more responsive in the wake of recent, high-visibility investments in state presence.