This paper presents an analytical model for the performance prediction of queueing networks with batch services and batch arrivals, related to the fluid limit of a suitable single-parameter sequence of continuous-time Markov chains and interpreted as the deterministic approximation of the average behaviour of the stochastic process. Notably, the underlying system of ordinary differential equations exhibits discontinuities in the right-hand sides, which however are proven to yield a meaningful solution. A substantial numerical assessment is used to study the quality of the approximation and shows very good accuracy in networks with large job populations.