英语论文网

ding high fl exibility andability to handle categorical variables.Particularly interesting for practicalimplementation is the identifi cation ofworkable rules that allow managers to easilyclassify customers according to theirattributes and thus implement profi tmaximisingmarketing strategies.The structure of the paper is as follows.Following a short conceptual overview ofthe measurement and prediction ofcustomer lifetime value in earlier studies, wepresent an introduction to decision trees.Next, we illustrate our approach with a casestudy using customer data from a majorEuropean airline. This section describesattributes of the database and presents thedistribution and segmentation of customervalues in our database. We then use decisiontrees to predict the value segment of agiven customer based on observabledemographic and behavioural features.Depending on the duration of the previouscustomer – fi rm relationship, differentinformation is available to us; therefore, weobtain very different results. Mostnoteworthy — even for prospectivecustomers for whom no prior behaviouraldata are available — we obtain very goodpredictions of future customer value. To thebest of our knowledge, such evaluation ofprospects has not yet been addressed in thecustomer lifetime value literature. Weconclude by discussing the results andmanagerial implications.Concept and prediction of customerlifetime valueTo choose the appropriate investment intoacquiring or retaining a customer, a fi rstquestion that must be answered is how tosegment customers according to their utilityfor the fi rm. As Woo and Fock 1 note, ‘ themain differences in characterising right andwrong customers are satisfaction level andprofi tability ’ . One measure used by severalresearchers in recent years to express longruncustomer profi tability is lifetime value , forwhose computation several different modelshave been proposed (for an overview, seeJain and Dingh 6 and Gupta et al . 7 ).Customer lifetime value is most commonlydefi ned as the sum of the discounted netcash fl ows generated by a customer duringhis / her relationship with the company. Thisdefi nition can be expressed analytically bythe following formula, based on standardmethodology for the fi nancial evaluation ofassets: 8CLVj = ΣTt = 0rj,t  cj,t(1+i)twhere j refers to the customer beingevaluated, r j , t and c j , t are the revenue andcost generated by customer j in time t , i isthe constant discount rate taking intoaccount the time value of money, and T isthe duration of the relationship with thecompany. According to this formula for thecomputation of the lifetime value of aparticular customer, we must forecast therevenues and costs at each future time stepand estimate the remaining time interval(ie the lifetime) during which the customergenerates revenues for the company.Too often, however, because of dataavailability and complexity constraints,models only predict the average customerlifetime value at an aggregate level for thewhole customer base 9 without taking intoconsideration characteristics of the singlecustomer. This limitation is a seriousdrawback, since profi tability is usually notTirenni, Kaiser and Herrmann132 Database Marketing & Customer Strategy Management Vol. 14, 2, 130–142 © 2007 Palgrave Macmillan Ltd 1741-2439 $30.00distributed uniformly among customers 10and a primary objective of the lifetime valueapproach is to identify highly profi tablecustomers in order to keep existing onesand attract