All mammals exhibit flexible decision policies that depend, at least in part, on the cortico-basal ganglia-thalamic (CBGT) pathways. Yet understanding how the complex connectivity, dynamics, and plasticity of CBGT circuits translate into experience-dependent shifts of decision policies represents a longstanding challenge in neuroscience. Here we present the results of a computational approach to address this problem. Specifically, we simulated decisions during the early learning process driven by CBGT circuits under baseline, unrewarded conditions using a spiking neural network, and fit an evidence accumulation model to the resulting behavior. Using canonical correlation analysis, we then replicated the identification of three control ensembles (responsiveness, pliancy and choice) within CBGT circuits, with each of these subnetworks mapping to a specific configuration of the evidence accumulation process. We subsequently simulated learning in a simple two-choice task with one optimal (i.e., rewarded) target and found that, during early stages of learning, feedback-driven dopaminergic plasticity on cortico-striatal synapses effectively increases reward rate over time. The learning-related changes in the decision policy can be decomposed in terms of the contributions of each control ensemble, whose influence is driven by sequential reward prediction errors on individual trials. Our results provide a clear and simple mechanism for how dopaminergic plasticity shifts subnetworks within CBGT circuits so as to increase reward rate by strategically modulating how evidence is used to drive decisions.