The low energy excited states of the conformational isomers of solvated azobenzene are calculated with several DFT methods accounting for the solute-solvent interaction implicitly with the polarizable continuum model or explicitly with subsystem DFT. For the latter, embedding potentials are calculated for 21 sampled snapshots of the solvent molecules. First, we find that accounting for the solvent implicitly or explicitly has little effect on the predicted cis-trans S1 excitation energy gap. Second, we find that azobenzene's S1 cis and trans energies are accurate as long as a screened range-separated hybrid exchange-correlation functional is employed. Finally, we also tested a simplified workflow whereby a single, averaged, embedding potential is used. Unfortunately, we find larger deviations against the experiment for the simplified workflow. This highlights a basic flaw in the approach, where the time scale of solvent averaging is much longer than that of the solute's electronic polarization.