Cocoa production remains a cornerstone of Nigeria’s rural economy and export revenue. However, the excessive use of non-approved pesticides threatens compliance with international Maximum Residue Limits (MRLs), risking market rejection and income losses. To address this, the Nigerian authorities introduced a list of approved pesticides (APs), yet their adoption remains limited. This study investigated the socio-economic and institutional drivers of AP adoption and quantified its causal impact on cocoa productivity and profitability using advanced machine learning methods. Survey data from cocoa farmers in Osun State were analyzed using logistic regression and random forest models to identify adoption and profitability drivers. Key predictors of AP use included education, the farm size, off-farm income, and awareness. Profitability was influenced by age, experience, the farm size, and the input use (fungicides and insecticides). To estimate the causal effects, we employed Causal Forests, revealing that AP adoption increases the log output by 41.6% and the log profit by 44.9%. These findings highlight the transformative impacts of approved pesticide adoption on farm welfare. Promoting awareness and targeted support policies can scale adoption and enhance sustainability. Machine learning methods enriched the analysis by revealing both average and heterogeneous treatment effects, offering evidence-based insights for agricultural policy and development planning.