고객세분화를 위한 기대가치산출모형에 관한 연구

Abstract

Currently, companies recognize the relationship between company and customers as critical to succeeding in marketing competition. Companies accumulate data by actively acquiring information related to customers using advanced information technologies, and with this accumulated database, studies are made actively to segment customers. These studies help identify the customers who could bring profits to companies in the future, calculate the future expected values of customers by analyzing their past purchase performance, and then segment them based on the calculated expected values. The notable models for calculating customer values are the recency, frequency, monetary(RFM) model and the customer lifetime value (CLV) model. However, these models have several limitations in measuring the future expected values of customers with difficulty in applying real marketing practice and effectively. Therefore, this study aims to develop an expected valuation model from the probable increase in purchasing frequency based on the recurring purchases of customers with a certain frequency and the subsequent expected sales. First, after presenting the customer movement model based on purchasing frequency, this study derived the detailed components of the expected valuation model, and analyzed detailed models on each component. Research on the detailed models used the data of 130,000 transactions from 60,094 customers in five fashion retail stores of company A for about nine years since December 2006. In detail, the probability of purchasing activity during the forecast period after the last purchasing date was estimated by deriving the discrete probability distributions for each group according to the purchasing frequencies. A model was developed to calculate the retention rate of customers for each group of purchasing frequency and the returning rate of customers who have left. The expected customer value model was completed by applying the average purchasing amount of customers in the past purchasing frequency based on the probability of additional purchasing. The model was evaluated using the same data of the same store and same customers for two years from August 2015 as a reserve sample. On the basis of the customer values calculated using this model, the total customers were divided into 10 groups, and each group showed a high correlation of more than 0.99 between the estimated and actual values. The accuracy of the prediction model, measured by the mean absolute percentage error(MAPE), was found to be "excellent." From these results, the expected value calculation model using the discrete probability distribution of each purchasing frequency group's repurchase cycle can be considered a future prediction model that reflects the customers' past purchase behavior well. This study provides the following theoretical implications: First, a "repurchasing cycle" variable was added to three main variables of the RFM model for each purchasing frequency group to obtain a model that can calculate the expected values of an individual customer based on the discrete probability distribution of the repurchasing cycle. Second, this study presented a model that can estimate whether a customer will leave or remain based on quantitative values. Third, for customers who left but returned to re-establish relationships, this study adopted the concept of "customer's return" to manage the purchasing performance of each customer continuously without disconnecting with the earlier records, unlike past research that commonly classify them as new customers. Furthermore, this study presents the following practical implications. First, a company managing the transaction data of individual customers as a database can apply the data to any case regardless of business or industry type even in an industry with long or irregular repurchasing cycles by generalizing it. Second, whenever a purchasing activity occurs, the expected customer values reflecting the latest customer behavior parameters will be calculated. Third, customers can be segmented by only the expected customer values calculated using the model proposed in this study, and this model can be applied to customer segmentation in a variety of ways such as segmentation based on the purchasing performance and the expected values calculated of each customer. For marketing communication with all the customers of the same level in the data who had a transaction in the past, the issues of effectiveness in terms of cost and performance compared to goals come to the fore. Therefore, if the person in charge of marketing calculates the expected customer values using the model proposed in this study and segments customers using the calculated values, it would lead to better marketing strategies for each subgroup.

Key words: CRM, Customer Segmentation, Expected Value Model, Customer Value, RFM, CLV, Repurchase Interval, Probability of Purchase Activity, Probability of Customer Retention, Probability of Customer Come Back