Once a design concept has been chosen and parameterised, the embodiment design stage consists of choosing materials and dimensions to ensure a 'good matching' with the expected performances. In this context of preliminary design stages, several approaches exist, which correspond to slightly different complexities and issues and must consequently be used at different moments. We consider in this paper three families of approaches: 1) exploring design (parametric) dimensioning under uncertainty (through constraint programming techniques, representations of feasible design points or Pareto frontiers) 2) robust design and multidisciplinary optimisation 3) design for reliability. We advocate and state in this paper that these approaches must be used in that order of increasing complexity. Indeed, applying an approach allows one to quickly figure out inadequacies with performance specifications or initial allowable bounds of design parameters and then to backtrack or to refine the design issue before proceeding to the next stage or approach. We illustrate that phenomenon by successively applying the three approaches on a dimensioning issue of a two-member truss structure. We clearly show that the successive optimal designs obtained are notably different, but that the optimal point obtained in a given approach is used to explore its surroundings within the next approach.