The main goal of psychiatric high-risk research--the personalised early recognition and intervention of schizophrenic and affective psychoses--is one of the biggest challenges of current clinical psychiatry due to the immense socioeconomic burden of these disorders. In this regard, this review discusses the prospects and caveats of new clinical, neuropsychological, neurophysiological and imaging-based concepts aimed at optimising the current state-of-the-art of early recognition. Finally, multivariate modelling and machine learning methods are presented as a novel methodological framework facilitating the decoding of early psychosis into different intermediate phenotypes. In the future, these phenotypes could be employed for a more objective risk stratification that operates at the single-subject level. This could allow us to generate clinically applicable prognostic biomarkers for these disorders that would propel the individualised prevention of disease transition, chronification and psychopharmacological treatment resistance of psychotic disorders.