Globally, obstetric haemorrhage (OH) remains the leading cause of maternal mortality. Much of the associated mortality is ascribed to challenges surrounding deployment of innovations rather than lack of availability. In low- and middle-income countries (LMICs), where the burden is highest, there is a growing interest in implementation research as a means to bridge the ‘know–do’ gap between proven interventions and their reliable implementation at scale. In this systematic review, we identified and synthesized qualitative and quantitative data across the implementation outcomes of OH prevention innovations in LMICs using a taxonomy developed by Proctor et al. We also identified service outcomes for the included innovations, as well as implementation strategies and implementation facilitators and barriers. Eligible studies were empirical, focused on the implementation of OH prevention programmes or policies and occurred in an LMIC. Eight databases were searched. Two authors independently assessed studies for selection and extracted data; the first author resolved discrepancies. Narrative synthesis was used to analyse and interpret the findings. Studies were predominantly focused in Africa and on primary prevention. Interventions included prophylactic use of uterotonics (n = 7), clinical provider skills training (n = 4) and provision of clinical guidelines (n = 1); some (n = 3) were also part of a multi-component quality improvement bundle. Various barriers were reported, including challenges among intervention beneficiaries, providers and within the health system; however, studies reported the development and testing of practical implementation solutions. These included training and monitoring of implementers, community and stakeholder engagement and guidance by external mentors. Some studies linked successful delivery to implementation outcomes, most commonly adoption and acceptability, but also feasibility, penetration and sustainability. Findings suggest that innovations to prevent OH can be acceptable, appropriate and feasible in LMIC settings; however, more research is needed to better evaluate these and other under-reported implementation outcomes.