The "factor loadings" are really the weights attributed to different variables that predict default. If you increase the value of these factor loadings, you increase the prediction of default, thereby making the model more conservative.
Whether factor loadings are high enough ex ante is often defined by ex post events. If you had a sample of firms a certain percentage, x, of which defaulted, you might begin by adding up the predicted default rate of the sample and comparing it to x. If the predicted rate were too low/high, you would increase/decrease the factor loadings in order to get the predicted default rate to approximate the actual default rate, x.
This is an iterative process that requires a bunch of trial and error, and basically assumes that the future samples of companies will be much like the ones on which you did your adjustments of factor loadings.