The global decline of insect pollinators, especially bees, is cause for concern, and there is an urgent need for cost-effective conservation measures in agricultural landscapes. While landscape context and habitat quality are known to influence species richness and abundance of bees, there is a lack of evidence from manipulative field experiments on bees' responses to adaptive management across differently structured landscapes. We present the results of a large-scale study that investigated the effects of a targeted agri-environment scheme (AES) on bumble bees (Bombus spp.) over three years in the United Kingdom. Forage patches of different sizes were sown with a conservation flower mixture across eight sites covering a broad range of agricultural land use types. Species richness and worker densities (especially of the longer-tongued Bombus species for which the mixture was targeted) were significantly higher on sown forage patches than on existing non-crop control habitats throughout the three-year study, but the strength of this response depended on both the proportions of arable land and abundance of herbaceous forb species in the surrounding landscape. The size of sown patches also affected worker density, with smaller patches (0.25 ha) attracting higher densities of some species than larger patches (1.0 ha). Our models show that a targeted AES can deliver greater net benefits in more intensively farmed areas, in terms of the number and species richness of bumble bees supported, than in heterogeneous landscapes where other foraging habitats exist. These findings serve to strengthen the evidence base for extending agri-environment schemes to boost declining pollinator populations to a larger number of agricultural landscapes across the globe.