Context: Agri-environment schemes support land management interventions that benefit biodiversity, environmental objectives, and other public goods. Process-based model simulations suggest the English scheme, as implemented in 2016, increased wild bee pollination services to pollinator-dependent crops and non-crop areas in a geographically heterogeneous manner. Objectives: We investigated which interventions drove the scheme-wide predicted pollination service increase to oilseed rape, field beans and non-cropped areas. We determined whether the relative contribution of each intervention was related to floral and/or nesting resource quality of the intervention, area of uptake, or placement in the landscape. Methods: We categorised interventions into functional groups and used linear regression to determine the relationship between predicted visitation rate increase and each category’s area within a 10 km grid tile. We compared the magnitude of the regression coefficients to measures of resource quality, area of uptake nationally, and placement to infer the factors underpinning this relationship. Results: Hedgerow/woodland edge management had the largest positive effect on pollination service change, due to high resource quality. Fallow areas were also strong drivers, despite lower resource quality, implying effective placement. Floral margins had limited benefit due to later resource phenology. Interventions had stronger effects where there was less pre-existing semi-natural habitat. Conclusions: Future schemes could support greater and more resilient pollination service in arable landscapes by promoting hedgerow/woodland edge management and fallow interventions. Including early-flowering species and increasing uptake would improve the effect of floral margins. Spatial targeting of interventions should consider landscape context and pairing complimentary interventions to maximise whole-scheme effectiveness.