A recent approach to simulating localized feedback from active galactic nuclei by Power et al. (2011) uses an accretion disc particle to represent both the black hole and its accretion disc. We have extrapolated and adapted this approach to simulations of Milky Way-sized galaxy mergers containing black holes and explored the impact of the various parameters in this model as well as its resolution dependence. The two key parameters in the model are an effective accretion radius, which determines the radius within which gas particles are added to the accretion disc, and a viscous time-scale which determines how long it takes for material in the accretion disc to accrete on to the black hole itself. We find that there is a limited range of permitted accretion radii and viscous time-scales, with unphysical results produced outside this range. For permitted model parameters, the nuclear regions of simulations with the same resolution follow similar evolutionary paths, producing final black hole masses that are consistent within a factor of two. When comparing the resolution dependence of the model, there is a trend towards higher resolution producing slightly lower mass black holes, but values for the two resolutions studied again agree within a factor of two. We also compare these results to two other AGN feedback algorithms found in the literature. While the evolution of the systems vary, most notably the intermediate total black hole mass, the final black hole masses differ by less than a factor of five amongst all of our models, and the remnants exhibit similar structural parameters. The implication of this accretion model is that, unlike most accretion algorithms, a decoupling of the accretion rate on to the black hole and the local gas properties is permitted and obtained; this allows for black hole growth even after feedback has prevented additional accretion events on to the disc.