Shared mobility has an important role in supporting existing transportation options in cities. However, when not deployed carefully, shared services may have operational inefficiencies such as low occupancies and increased deadheading. One reason is the spatio-temporal variance in the distribution of urban trip demand, which may lead to an unbalanced fleet displaced in cities thus unable to serve requested trips. Strategically siting holding areas (depots for dispatching and relocating fleets) could help improve fleet performance. Therefore, this paper considers shared autonomous vehicle (SAV) fleet operations by modeling the impacts of different holding area policies on service performance. Modeling and comparing multiple holding area policies for tactically deploying SAVs is novel, and the insights from this paper can inform service providers on how to site holding areas for improved performance. We develop a model of SAV fleet with pooling in the City of Toronto, with 27,951 total SAV trip requests across a 16-h period. We then integrate four holding area policies estimated using different spatial clustering methods, centralized positioning, and existing taxi stands. Findings indicate that using agglomerative clustering results in superior SAV fleet performances (average passenger waiting times reduced by about 20% compared with the worst performing policy), with increased served demand and reduced deadheading. A single holding area at a high trip density location yields efficient service performance at lower fleets but struggles to serve sparse demand (producing worst results); this method may suffice for operating SAV services within a small geofence with high trip densities.