We review the mathematical tools used to describe binary-state dynamics on complex networks. We then show their limitations and failures and we propose a generalization to improve its accuracy and scope. Specifically, the improved method allows one to calculate correlations, fluctuations and finite-size effects of the description variables.