In this talk I will show how the general framework of effective quantum gravity proposed by Brunetti, Fredenhagen and myself can be used to study problems in quantum cosmology. This formulation allows to derive some known results in (first order) cosmological perturbation theory from first principles and it provides a robust tool for computing higher order corrections. Conceptually, it works as a test ground for new ideas concerning construction of gauge invariant observables in effective quantum gravity. The non-local character of these observables leads to new interesting combinatorial structures arising in renormalization.