Traditional physical retail stores often attract customers by sending email-flyers (or promotional electronic Direct Mail, DM). However, most customers will treat electronic DM as junk mail if it includes too many types of commodities or is sent too frequently. To avoid this problem, we propose a recommendation algorithm to select a small and fixed number of commodities for preparing customized electronic DM suitable for a well-known Taiwanese supermarket. Our method first selects a small and fixed number of commodity types that customers will most likely purchase; then, at product level, the method selects one commodity of each commodity type to promotion. Considering the habits of customers have already purchased commodities will buy again, when the recommendation success rate for recommending commodity types (or commodities) reached a certain value, we could replace some commodity types (or commodities) those the customers had already purchased in the last month. Compared to two item-based collaborative filtering approaches, Cosine and Bigram, the experimental results show that our approach has higher recommendation success rate and then increases the possibility of customers back to physical retail stores to purchase.