Insufficient or inappropriate soil testing can lead to a range of undesirable consequences, and yet there is no guideline for optimal investigation. This study analyses the influence of a two layer, virtual soil profile with an undulating boundary on site investigation performance. Factors investigated include the method of representing the boundary within the soil model, the stiffness ratio of the two layers, choice of test type, the pile length relative to the boundary length, and the number of boreholes and piles. The relative error contribution from the uncertainty sources of layer geology and soil variability is also examined. Investigation performance is assessed through Monte Carlo analysis in terms of total expected project cost, while implicitly incorporating the risk of damage from poor investigation. It has been shown that the optimal investigation can save in the order of AUD$1.5 million and that 2D soil models can represent 3D soils.