I think ArcGIS can be used to develop a grid and then sample a road network. However, I am not sure how such a sample can be stratified on the basis of some observable attribute such as business and office location etc.?
My general opinion: simulation of real objects with random samples will lead us to false science. Scientists should look for hidden regularities in the cloud of randomness.