In the talk we consider the use of multiple comparisons procedures in the following applied statistical problem. Given the results of customers survey for a large company with a network of branches, we want to identify a) the three "best" branches, which will get bonuses; b) all the branches, where the results of the survey were substantially below the median level, which have to be additionally investigated and possibly re-organized. To distribute bonuses among the three "best" branches in a fair way, we propose the procedure for testing differences among them, which takes into account that they are the three best points among a given number of observations.The number of the 'worst' branches is not fixed from the beginning and must be determined via some decision making procedure, which also uses the distribution of extreme observations.