The reason for invoking theory is that it's crucial to understand within which theoretical framework a software package operates. Sure, we solve the numerical equations as exactly as possible, but the equations themselves are set up under certain assumptions, since it's impossible to solve the full quantum-mechanical problem for a system of more than 1 or 2 atoms. This is no different from a macroscopic device simulator which perhaps assumes a constant resistance, or parabolic bands, etc, which may not hold for hot electrons or when dimensions are reduced, etc.
In the case of ATK, one of these assumptions is the absence of inelastic scattering, e.g. in the shape of phonons, or any phase decoherence. That is, we solve the problem in the coherent, ballistic regime. It may certainly be the case that phonons etc will limit the mobility of a realistic device, and you may also not have the exact same geometrical configuration in each fabricated device (small fluctuations, random dopants, etc), but what ATK is designed to do is to provide a theoretical limit to the performance of the device, and to give some picture of the variability (you can insert randomness explicitly and see how it affects the conductance or not). ATK is also intended as a tool to make choices between possible design choices, to judge which one will likely be the better candidate, before making an expensive experiment, and finally to provide insight into the mechanisms that control the device operation. If these are understood on a fundamental level, you have a better chance to optimize the design or come up with ideas for new and improved device configurations.
