We consider the downlink of a wireless system with an M-antenna base station and K single-antenna users. A limited feedback-based scheduling and precoding scenario is considered that builds on the multiuser random beamforming (RBF). Such a scheme was shown to yield the same capacity scaling, in terms of multiplexing and multiuser diversity gain, as the optimal full CSIT-based (channel state information at transmitter) precoding scheme, in the large number of users K regime. Unfortunately, for more practically relevant (low to moderate) K values, RBF yields degraded performance. In this work, we investigate solutions to this problem. We introduce a two-stage framework that decouples the scheduling and beamforming problems. In our scenario, RBF is exploited to identify good, spatially separable, users in a first stage. In the second stage, the initial random beams are refined based on the available feedback to offer improved performance toward the selected users. Specifically, we propose...