We present a fluctuation-response relation for arbitrary Markov processes, relating the time correlation of a variable with its linear response to a external perturbation [Prost, Joanny, Parrondo, Phys. Rev. Lett. 103, 090601 (2009)]. This relation allows us to apply to general systems the strategy that non equilibrium statistical mechanics developed to study the response of a physical system close to equilibrium: Langevin equations, fluctuating linear irreversible thermodynamic, etc. More precisely, we will give some hints on how to build a reduced linear model that reproduces such relation, providing a general tool to reduce the number of variables and complexity of a system under weak external perturbations.