Recently we have developed and validated model-based forest methodology (mobForest) to predict patient outcome under hypothetical treatment assignments (i.e. what if a patient is assigned to an alternative treatment). We used mobForest as well as regularized regressions, trees, and a hybrid model to predict patient response to treatment. We present the results of two analyses: (1) the comparison of treatments across the patients and testing for the uniform superiority of one treatment vs. another; (2) identification of patients that benefit from personalized treatment assignment. This is achieved by testing the outcome differences between patient assignment to the most effective treatment, to a random treatment, and to the least effective treatment. We applied this methodology to clinical studies of alcohol- and heroin-abuse treatment. We tested the difference between the best, the second best, random, and the worst treatments. We show that for some patients there is no significant difference in the outcomes, while others could substantially benefit from personalized best choice of treatment. We illustrate the results through innovative "hurricane" graphics.