A key challenge in the theoretical study of quantum many body systems is to overcome the exponential growth of the Hilbert space with the system size. Many successful approaches are variational, i.e., they are based on choosing suitable families of states that capture key properties of the system. Prominent examples range from Gaussian states to matrix product states and tensor networks. In this talk, I will review the geometric structures of variational manifolds and how we can use them to systematically (a) estimate ground state energies, (b) compute approximate excitation spectra, (c) predict the linear response of these systems and (d) study quenches in the presence of conserved quantities. Taking the Bose-Hubbard model as an example, I show how our methods give rise to a systematic extension of the traditional Bogoliubov theory.