AbstractThe heterogeneity in recovery outcomes for individuals with First Episode Psychosis (FEP) calls for a strong evidence base to inform practice at an individual level. Between 19–89% of young people with FEP have an incomplete recovery despite gold-standard evidence-based treatments, suggesting current service models, which adopt a ‘one-size fits all’ approach, may not be addressing the needs of many young people with psychosis. The lack of consistent terminology to define key concepts such as recovery and treatment resistance, the multidimensional nature of these concepts, and common comorbid symptoms are some of the challenges faced by the field in delineating heterogeneity in recovery outcomes. The lack of robust markers for incomplete recovery also results in potential delay in delivering prompt, and effective treatments to individuals at greatest risk. There is a clear need to adopt a stratified approach to care where interventions are targeted at subgroups of patients, and ultimately at the individual level. Novel machine learning, using large, representative data from a range of modalities, may aid in the parsing of heterogeneity, and provide greater precision and sophistication in identifying those on a pathway to incomplete recovery.