Transportation networks are one of the fundamental tools for human society to work, more so in our globalized world. The importance of a correct, efficient design of a transportation network for a given region or country cannot be overstated. We here study how network design is affected by the geography of the towns or nuclei to be connected, and also by the decision process necessary to choose which connections should be improved (in a generic sense) first. We begin by establishing that Delaunay networks provide an efficient starting point for the network design and at the same time allow us to introduce a computationally amenable model. Subsequent improvements lead to decentralized designs in geographies where towns are more or less homogeneously distributed, whereas radial designs arise when there is a core-periphery distribution of nodes. We also show that optimization of Delaunay networks outperforms that of complete networks at a lower cost, by allowing for a proper selection of the links to improve. In closing, we draw conclusions relevant to policy making applied to designing transportation networks and point our how our study can be useful to identify mechanisms relevant to the historical development of a region.