In recent years, the shipbuilding industry has been facing transformative changes to reduce the environmental impact of maritime travel. Recent regulations for the reduction of emissions fuel the demand for lighter ships and the adoption of alternative energy storage solutions, requiring the designers to face novel structural configurations. In this work, in collaboration with Fincantieri S.p.A., we propose a transfer learning approach to speed up the optimization of passenger cruise ship designs. At the beginning of the design process, after the geometry and functional areas of the ship have been finalized, the designers are tasked with ensuring the structural reliability under extreme wave conditions. The behavior of the ship is controlled via the thickness of primary structural components (e.g., decks, bulkheads, shell) and the simulation of the entire ship via finite element analysis under multiple load conditions. The use of data driven surrogates instead of the high-fidelity solver enables the use of black-box optimization to select the cheapest, but structurally sound, configurations from the vast parametric domains. However, the best results are achieved through iterative refinement of the optimization problem, adding new decision variables according to the sensitivity of different regions of the ship in an increasingly detailed manner. The resulting optimization process is effective and appropriately tailored to the early design phase, but it is still too expensive to be used in the preliminary design, when the users are focused on studying the structural responses and identifying defects unsolvable with simple steel allocation. To address this requirement, we propose a one-shot parameterization procedure that, by acting on the initial model and a single instance of finite element analysis, is able to generate an effective optimization problem without the need for surrogates to study the sensitivity of structural responses. The procedure is tested on a set of simplified midship sections, and on a set of recent commercial designs by Fincantieri. The results achieve good trade-offs between the total number of high-fidelity simulations and the quality of the final optimal configuration, streamlining the preliminary design phase.