Sustainable intensification and climate-smart agriculture initiatives promote complex systems-based innovations to simultaneously improve yields and conserve natural resources. These innovations are usually tested under near-perfect experimental conditions with purposively selected farmers. Using a quasi-experimental approach and geographic information system, we evaluate a systems-based sustainable intensification project in Malawi aiming at improving whole-farm productivity and nutrition through integrated agricultural innovations. We find adopters of these innovations to systematically differ from non-adopters and suggestive evidence of potential systematic targeting of project locations and households. Econometric results using efficient influence function and propensity score matching methods show consistently higher maize yield and value of harvest, on average and across quantiles, for project beneficiaries, compared to that of randomly selected non-beneficiary households in non-target villages. Our findings highlight the need to rethink selection criteria for systems-based innovations, something that could potentially bear severe implications upon scaling up.