I have recently proposed a generalization of the cluster variational method which can be used to study the out of equilibrium dynamics and the steady states of kinetic Ising-like models. In this talk I will show that this approach provides a general framework in which several previous results of a mean-field nature are reproduced as particular cases and can be systematically improved. I will consider applications to Ising-like models on various graphs, epidemic models on networks and asymmetric exclusion processes. In the last case I will discuss in particular a recently discovered dynamical transition.