It is investigated if material based structural safety can be replaced with safety obtained from SHM. SHM on its own does not add any value to a structure unless there is a decision policy attached. The further attachment of a loss function enables upfront calculation of the expected utility. In the presented case, damage sensitive features from sensitivity based damage detection are used in a supervised learning and optimization scheme. A Bayes detector, i.e. the static decision rule that minimizes the expected utility, is utilized for the system design. The technique is demonstrated in a simulation case of the NREL 5MW wind turbine tower subjected to bending fatigue and horizontal circumferential cracking at weld locations. Decision driven SHM is shown to change the initial design safety of the structure in the fatigue limit state. A common optimum of material safety and classifier threshold is found, enabling the calculation of the expected value of SHM. A sensitivity analysis of the feature extraction and of the loss function is included.