As I've rapidly discovered, trying to calculate aggregate stats over the basis of Poisson arrivals... is a pain. To work out anything interesting, you end up studying piecewise-deterministic markov processes in more detail than you'd wish.