We study the distribution of line-of-sight velocities of galaxies in the vicinity of SDSS redMaPPer galaxy clusters. Based on their velocities, galaxies can be split into two categories: galaxies that are dynamically associated with the cluster, and random line-of-sight projections. Both the fraction of galaxies associated with the galaxy clusters, and the velocity dispersion of the same, exhibit a sharp feature as a function of radius. The feature occurs at a radial scale $R_{\rm edge} \approx 2.2R_{\rm{\lambda}}$, where $R_{\rm{\lambda}}$ is the cluster radius assigned by redMaPPer. We refer to $R_{\rm edge}$ as the "edge radius." These results are naturally explained by a model that further splits the galaxies dynamically associated with a galaxy cluster into a component of galaxies orbiting the halo and an infalling galaxy component. The edge radius $R_{\rm edge}$ constitutes a true "cluster edge", in the sense that no orbiting structures exist past this radius. A companion paper (Aung et al. 2020) tests whether the "halo edge" hypothesis holds when investigating the full three-dimensional phase space distribution of dark matter substructures in numerical simulations, and demonstrates that this radius coincides with a suitably defined splashback radius.