<p>The evolution of metastasis in humans is considerably less well understood than the biology of early carcinogenesis. For over a century, clinicians and scientists have been debating whether metastatic potential is the intrinsic property of a cancer, pre-determined by the molecular characteristics of the tumour founder cell, or whether metastatic capacity evolves in a stepwise fashion as the tumour grows, akin to the multistage accumulation of oncogenic alterations that give rise to the first cancer cell. In this Perspective, I examine how genetic analyses of primary tumours and matched metastases can distinguish between these two competing metastasis evolution models, with particular emphasis on the utility of metastatic randomness — a quantitative measure that reflects whether metastases arise from a random selection of primary tumour subclones or whether they are enriched for descendants of privileged lineages that have acquired pro-metastatic traits. Probable metastasis evolution trajectories in tumours with high and low baseline metastatic capacity are discussed, along with the role of seeding rates and selection at different metastatic host sites. Finally, I argue that trailblazing insights into human metastasis biology are immediately possible if we make a concerted effort to apply existing experimental and theoretical tools to the right patient cohorts.</p>