A workload to be handled by a database system can be identified. The workload can include at least one query that the database system is to handle. A set of at least one candidate statistical views (statviews) to be utilized when optimizing the workload can be enumerated. A benefit value and a cost value of the each of the enumerated candidate statistical views relative to the entire workload can be computed. The cost value can reflect a cost of constructing and collecting statistics on the associated statistical view. A set of the candidate views most beneficial for handling the workload can be determined based upon the computed benefit values and computed cost values. A generalization phase that augments the candidate view set with higher value candidate views for consideration during the recommendation phase. The optimum subset of views from the determined set of candidate views can be recommended, which can cause them to be constructed and utilized by a database optimizer.