The carbon-dense peatlands of Indonesia are a landscape of global importance undergoing rapid land-use change. Here, peat drained for agricultural expansion increases the risk of large-scale uncontrolled fires. Several solutions to this complex environmental, humanitarian and economic crisis have been proposed, such as forest protection measures and agricultural support. However, numerous programmes have largely failed. Bundles of interventions are proposed as promising strategies in integrated approaches, but what policy interventions to
combine and how to align such bundles to local conditions remains unclear. We evaluate the impact of two types of interventions and of their combinations, in reducing fire occurrence through driving behavioural change: incentives (i.e. rewards that are conditional on environmental performance), and deterrents (e.g. sanction, soliciting concerns for health). We look at the impact of these interventions in 10 villages with varying landscape and fire-risk contexts in Sumatra, Indonesia. A private-led implementation of a standardised programme allows us to study outcome variability through a natural experiment design. We conduct a systematic cross-case comparison to identify the most effective combinations of interventions, using two-step qualitative comparative analysis (QCA) and geospatial and socio-economic survey data (n = 303). We analysed the combined influence of proximate conditions (interventions, e.g. fear of sanction) and remote ones (context; e.g. extent of peat soil) on fire outcomes. We show how, depending on the level of risk in the pre-existing context, certain bundles of interventions are needed to succeed. We found that, despite the programme being framed as rewards-based, people were not responding to the reward alone. Rather sanctions and soliciting concern appeared central to fire prevention, raising important equity implications. Our results contribute to the emerging global interest in peat fire mitigation, and the rapidly developing literature on PES performance.