有效促銷品預測模型之研究－以台灣超市資料為例

Abstract

降價促銷是最常見且能最快達到行銷目的的手法，然而研究顯示有超過一半的商家在降價促銷時無法獲得利潤，因此，除了在促銷期間吸引新客群來增加銷售額外，在促銷期間後是否能讓消費者願意以原價回購是商家更關切的議題，促銷後銷量增加才能達成企業以促銷活動增加該商品的客群的目的。本研究將行銷經驗系統化，以人工智慧彌補人工判斷的不足或缺漏，幫助專業人員在行銷時選擇適合的促銷品，利用數據分析發現潛在且有效的促銷產品，以技術模擬人的智慧尋找更好的銷售策略。研究中使用台灣大型連鎖超市所提供之四個月交易資料進行分析，為了使回購率高、促銷銷量大的產品脫穎而出，本研究將定義「促銷有效度」算法，並使用K-Prototypes分群法根據產品性質先作粗略的分群，最後將重要的產品性質加權，計算產品與「高有效促銷品」的相似度後，找出潛在的高有效促銷品，提供商家在進行降價促銷時的參考依據，配合行銷人員豐富的經驗找到適合的促銷產品。

Price promotion is the most common marketing tactics that can provide the visible short-term sales results. However, research indicates that nearly half of the companies would not be able to make a profit from price promotion. Besides attracting new product users, companies are more concerned with consumers’ repurchase intention of the product after the promotion period, as their promotion goal is to increase product sales without price reduction. This thesis intends to systematize experts’ marketing experience by using artificial intelligence to enhance the human decision making and help experts select the best products for promotion. Based on the patterns discovered in the large data sets, potential effective sales products will be identified and stronger promotion strategies could be found. Four-month transaction data used in this thesis was provided by a large chain supermarket in Taiwan. “Promotion effectiveness” is defined in this thesis to set potential products which are not only highly purchased during promotions but also highly repurchased after promotions apart from other products. K-Prototypes was used as a clustering algorithm to preliminarily categorize products based on their attributes. Ten important attributes were used to calculate the weighted similarity between all products and high promotion effectiveness products. Products that have high similarity to those high promotion effectiveness products will be identified and provided to companies as potential sales products. With experienced sales experts, products could be promoted more appropriately.