Up to one third of adolescents in the UK are bullied. Bullying and aggression are associated with poorer physical and mental health, and worse social and economic wellbeing across the life-course, making the need for effective interventions imperative. Evaluations of a wide range of interventions have found promising results, but outcomes are often inconsistent across contexts. INCLUSIVE is a school-based cluster-randomised controlled trial (RCT) evaluating Learning Together, a whole-school intervention that integrated restorative practices, social and emotional skills curricula, and student/staff action groups. INCLUSIVE is also the first RCT to be underpinned explicitly by a realist approach aiming to understand how intervention mechanisms interact with context to generate outcomes. This approach has been criticised by some as philosophically incongruent and practically unfeasible. Specific criticisms include the beliefs that 1) trials are inherently positivist; 2) randomisation and control make it impossible to assess context-mechanism-outcome configurations; and 3) trials are insufficiently theorised and concerned only with estimating effect sizes. This mixed-methods study uses data from INCLUSIVE’s process and outcome evaluations. Process evaluation data were analysed to understand fidelity, feasibility and acceptability of intervention resources, and how these varied by context. Further analysis of three diverse case-study schools sought to understand participant descriptions of their contexts and how changes were described as occurring as a result of using intervention resources. These were then used to inform theorisation of context-mechanism-outcome configurations which were testing using qualitative comparative analysis. Depending on both their needs and capacities, schools used resources in novel ways to decrease bullying with varying degree of success. Three key social mechanisms for reducing bullying were identified: building commitment to the school community; building healthy relationships by modelling and teaching pro-social skills; and de-escalating bullying and enabling re-integration of perpetrators back into the school community. Analysis suggests that these mechanisms were also activating in some control schools, indicating that they are plausible and potentially transferable. Based on these findings and other analyses of INCLUSIVE data, I reflect on the process of conducting the first realist RCT and conclude that they are philosophically cogent, and can produce nuanced findings about how an intervention works, for whom and under what conditions.