This talk will describe recent work with A. Carlotto on a method for approximating general asymptotically flat initial data sets by ones which are trivial outside cones of arbitrarily small angle. We will explain why this type of approximation is optimal in a certain sense, and we will give some applications of the construction.