We introduce an approximate variant of the NORTA method which aims at generating structured data from a given prior sample. The technique accommodates for any combinations of marginals (especially continuous/discrete mixtures) and a wide range of correlation structures. We focus on the interesting case where the sample includes categorical data, both ordered and unordered. We provide an application in the financial industry through a test of our iterative Newton-like algorithm on a dataset comprising the results of a questionnaire. We show that the sampled data, similarly to the NORTA technique, matches both the marginal and correlation structures of the original dataset closely. Consequently, analyses such as decision tree modeling or Support Vector Machine classification and regression, can be carried out on the new, much larger, sample without altering the core properties of the original sample.