STOCHASTIC MODEL OF MARKOV CHAIN TO CALCULATE CUSTOMER LIFETIME VALUE

• Main Author: Permana, Dony
• Format: Dissertations
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/62860
• Institution: Institut Teknologi Bandung

A customer of a company's products has a measure of value (in the currency) that is some of potential benefits of the customer company. This value will measure a customer on how profitable or not for a company. One of measure is customer life value or CLV. This value is calculated based on probability of a customer to buy a company's products continously at any given time period from the present to a long period in the future, or as long as a lifetime being the company’s customer. CLV is calculated by summing the potential benefits from a consumer that became customer for a long time in the future. This profit is calculated based on the number of net contributions or a price of one unit of product (good or service) that has been cleaned from its cost of production and then reduced by cost of retention and acquisition of the customer. High value of CLV means that the customer will bring greater benefits for company than marketing costs to acquire him as a customer. Marketing costs contain the costs of customer acquisition and cost of customer retention. Acquisition is an attempt to attract the consumer to buy and try a product. While retention is an attempt to retain customers who have ever purchased and countinously buy at every period of time. There are two characteristics of CLV calculations. First the element of probability of customer acquisition and probability of customer retention from one time period to the next period. Both of these probability make CLV estimation is stochasticly. The second feature, namely the factor of discount rate because calculation of benefits in future are taken into account in the present. This concept is called present value. By the time horizon, CLV is calculated for a long time in the future or time toward the infinite. But for specific purposes, CLV can also be calculated until the time T is limited in the future. In this dissertation, a consumer is assumed through some status when viewed by a company. For example, status of prospect customers, customers and former customers. Consumer behavior in the status are very diverse and may not be the same for every customer. Consumer character, is highly dependent on several factors such as price, quality, service, product competition, and others. Calculation of CLV on consumer who has a lot of status is done by a stochastic model of Markov chain. This model has matrix of transition probability that describe behavior of transition between consumer status. This matrix contains probability for customer acquisition and retention. Acquisition probability from prospect customer and former customers is the probability of transition from the status of prospect customer and former customer to customer status. While probability of customer retention is the probability of transition from customer status to itself. Through the Markov chain, CLV calculations performed more efficiently because Markov calculate CLV at any initial status in everywhere at once. The concept also has convergence properties if the process runs for a long time in the future. Besides that, probability of first transition from one status to another status can also used by Markov chain models. For example, it used to calculate a transition probability from a customer that have come out to buy product again in the future. The use of stochastic concept is more advantageous because it can obtain better results than simply calculation value of the CLV. In this dissertation, CLV explored by behavior of variables influence it. The variables are the probability of retention, probability of acquisition, interest rates, net contribution, cost of retention, and cost of acquisition. The purpose of the exploration is to see the direction change of CLV based on increasing of its variables. From the behavior, company can make policy settings especially in terms of the amount of the interest rate, the net contribution, and marketing in order to obtain financing CLV expected behavior. In another part of this dissertation, the model of CLV calculation is including inverse problems. On this topic, input of vector CLV is considered as the target company to a customer for a time T in the future. While the output is estimated probability of acquisition and probability of retention. This Feedback can be used as the target company to a customer CLV. While the output is a reference to the marketing team to conduct customer acquisition and retention in order to generate the target CLV that already set. Inverse problems on the model involving numerical approach for dealing systems of nonlinear equations with many variables. Numerical approach that chosen is differential evolution algorithm.