Before getting start to modeling, I have a good habit to create a block out of the scene [or the model]. Blocking out the scene allows me to set the sizes, proportions, the presence of details, and it's really useful to identify critical parts or parts not required [out of views].
First things first, I synchronized my camera view with Unity standard first person angle of view [Modo = 80°, Unity = 60°] just to view exactly what I will see into game view.
Then I build the scene using primitives with grid snapping. The main reason for the grid snapping usage is to cut the ground in "blocks" to easily discard from memory geometries not present in the frustum culling. If the ground were a single block, I mean a single mesh, it will entirely remain in memory every single frame. Same things for the buildings: every single floor will be a separated mesh [game object].
And this is the block out result:
