Understanding why forest conservation initiatives succeed or fail is essential to designing cost-effective programs at scale. In this study, we investigate direct and indirect impact mechanisms of a REDD+ project that was shown to be effective in reducing deforestation during the early years of its implementation in the Transamazon region, an area with historically high deforestation rates. Using counterfactual impact evaluation methods applied to survey and remote-sensing data, we assess the impact of the project over 2013-2019, i.e., from its first year until two years after its end. Based on the Theory of Change, we focus on land use and socioeconomic outcomes likely to have been affected by changes in deforestation brought about by the initiative. Our findings highlight that forest conservation came at the expense of pastures rather than cropland and that the project induced statistically greater agrobiodiversity on participating farms. Moreover, we find that the project encouraged the development of alternative livelihood activities that required less area for production and generated increased income. These results suggest that conservation programs, that combine payments conditional on forest conservation with technical assistance and support to farmers for the adoption of low-impact activities, can manage to slow down deforestation in the short term are likely to induce profound changes in production systems, which can be expected to have lasting effects.