It would be unusual not to check the performance of the model at predicting the target variable, which would validate whether or not the derived feature is useful.