Decision Support Systems (DSS) with embedded best management practices help reduce pesticide use. However, costly experimentation is necessary to learn about the quality of their data processing and recommendation models. We study farmers’ choices to try such a DSS. Due to order effects, learning can be inefficient and stop after inadequate recommendations, leading to crop losses. Subsidizing crop insurance conditional on DSS use is a cost-effective policy to foster experimentation. A 4-year living lab on-farm experiment involving French wine cooperatives, an insurer, and researchers supports our theoretical results on learning dynamics and provides additional insight into information requirements for subsidized green insurance.