EmpaTeach was the first intervention to address teacher violence to be tested in a humanitarian setting and the first to focus on reducing impulsive use of violence, but a cluster randomised trial found no evidence that the intervention was effective in reducing physical and emotional violence from teachers. We aimed to understand why. We conducted a quantitative process evaluation to describe the intervention implementation process (what was implemented and how); examine teachers' adoption of positive teaching practices (was the content of the intervention taken up by participants), and test mechanisms of impact underlying the program theory (how the intervention was supposed to produce change). Despite participation in the intervention activities and adoption of intervention-recommended strategies (classroom management and positive disciplinary methods), we show that teachers who used more positive discipline did not appear to use less violence; and teachers in intervention schools did not experience gains in intermediate outcomes such as empathy, growth mindset, self-efficacy or social support. Our findings suggest that the intervention did not work due to the failure of some key hypothesised mechanisms, rather than because of implementation challenges.