We consider the non-equilibrium dynamics of disordered systems as defined by a master equation involving transition rates between configurations (detailed balance is not assumed). To compute the important dynamical time scales in finite-size systems without simulating the actual time evolution which can be extremely slow, we propose to focus on first-passage times that satisfy 'backward master equations'. Upon the iterative elimination of configurations, we obtain the exact renormalization rules that can be followed numerically. To test this approach, we study the statistics of some first-passage times for two disordered models : (i) for the random walk in a two-dimensional self-affine random potential of Hurst exponent \(H\), we focus on the first exit time from a square of size \(L \times L\) if one starts at the square center. (ii) for the dynamics of the ferromagnetic Sherrington-Kirkpatrick model of \(N\) spins, we consider the first passage time \(t_f\) to zero-magnetization when starting from a fully magnetized configuration. Besides the expected linear growth of the averaged barrier \(\bar{\ln t_{f}} \sim N\), we find that the rescaled distribution of the barrier \((\ln t_{f})\) decays as \(e^{- u^{\eta}}\) for large \(u\) with a tail exponent of order \(\eta \simeq 1.72\). This value can be simply interpreted in terms of rare events if the sample-to-sample fluctuation exponent for the barrier is \(\psi_{width}=1/3\).