Research Article A CBR-Based and MAHP-Based Customer Value ...downloads.hindawi.com/journals/tswj/2014/459765.pdf · Chan s model, this study proposes a CBR-based and MAHP-based customer

Research ArticleA CBR-Based and MAHP-Based Customer Value PredictionModel for New Product Development

Yu-Jie Zhao Xin-xing Luo and Li Deng

Business School Central South University Changsha 410083 China

Correspondence should be addressed to Yu-Jie Zhao yuanyuan860831163com

Received 12 February 2014 Accepted 18 June 2014 Published 4 August 2014

Academic Editor Juan M Corchado

Copyright copy 2014 Yu-Jie Zhao et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In the fierce market environment the enterprise which wants to meet customer needs and boost its market profit and share mustfocus on the new product development To overcome the limitations of previous research Chan et al proposed a dynamic decisionsupport system to predict the customer lifetime value (CLV) for new product development However to better meet the customerneeds there are still some deficiencies in their model so this study proposes a CBR-based and MAHP-based customer valuepredictionmodel for a new product (CampM-CVPM) CBR (case based reasoning) can reduce expertsrsquo workload and evaluation timewhile MAHP (multiplicative analytic hierarchy process) can use actual but average influencing factorrsquos effectiveness in stimulationand at same time CampM-CVPM uses dynamic customersrsquo transition probability which is more close to reality This study not onlyintroduces the realization of CBR and MAHP but also elaborates CampM-CVPMrsquos three main modules The application of theproposed model is illustrated and confirmed to be sensible and convincing through a stimulation experiment

1 Introduction

Under the furiously competitive market the competitionamong similar enterprises has been focused on new productdevelopment that is a powerful guarantee of the enterprisewho wants to enhance the core competitiveness and gainsuperiority However new product development is not onlyof high cost but also of higher cost of failure developmentSuccessful new product development is not just a purelytechnical problem moreover it needs to consider variousinfluencing factors To improve success rate developers mustcomprehensively integrate various factors and then correctlyevaluate and choose the better development New productdevelopment is a dynamic process that includes raisinganalyzing and solving problemsThedecisionmaking whichis affected bywide range factors such as the productrsquos complexstructure is semistructured or even nonstructured Withthe maturation of information processing technology andsoftware auxiliary tools the decision making contributes tonot only reducing costs and increasing efficiency but alsoimproving business strategy [1] Developers are beginning tomake full use of the superiority of software auxiliary toolsfor better and easier obtaining of new product development

In recent years many models for new product developmenthave been proposedHowever these highlight the influencingfactors including product and market and seldom regardcustomer needs Only Chan et al overcome this shortcomingand propose a new model that integrates three recognizedinfluencing factors of new product development-productattributes specified by designers customer requirementsand satisfaction and marketing competence Using systemdynamics the progress for a new product launching the mar-ket is stimulated and the customer lifetime value (CLV) canbe obtained Relative to previous research Chanrsquos model hasincomparable superiority whether on evaluation perspectiveand method or the influencing factor But judging from ldquothebasic goal of software engineeringrdquo [2] effectively developingsoftwarewith systematical construction and engineeredman-agement it is obvious that developing systematic software canbetter meet user needs

From above perspective this study finds out three majordeficiencies by further analyzing Chanrsquos model (1) To eval-uate the effectiveness of each influencing factor the expertsrsquoworkload is too heavy and work time is too long Meanwhilethe evaluation results have obvious subjectivity deviations (2)The effectiveness of each influencing factor for different levels

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 459765 18 pageshttpdxdoiorg1011552014459765

2 The Scientific World Journal

of customers is not yet distinguished (3) Different levels ofcustomers have the same transition probability which is staticbut dynastic To overcome the above three deficiencies inChanrsquos model this study proposes a CBR-based and MAHP-based customer value prediction model for new productdevelopment (CampM-CVPM) and its advantages are as fol-lows (1) Case based reasoning (CBR) can reduce expertsrsquoevaluationworkload andwork time which couldmake up fordeficiency one (2) Multiplicative analytic hierarchy process(MAHP) uses the actual effectiveness of the influencing factorrather than the overall average in stimulation which couldmake up for deficiency two (3) The dynamic customersrsquotransition probability is more close to reality which couldmake up for deficiency three

This section has given the general background to thestudy Section 2 discusses the literature on new productdevelopment The methodology for the proposed CampM-CVPM is presented in Section 3 Section 4 introduces themodelrsquos framework and its three modules A stimulationexperiment is conducted in Section 5 Some concludingremarks are offered in Section 6

2 Literature Review

21 The Present State of the Research In recent years manyconventional and mature models for new product devel-opment have been proposed Xu et al use fuzzy set andutility and theory to evaluate the typical phases duringnew productrsquos whole lifecycle [3] Besharati et al mea-sure new product development according to the size ofcustomer expected utility that is market demand basedon customersrsquo preference designersrsquo experience preferenceand uncertainty of product designing characteristics [4]Alexouda applies algorithm and enumeration method innew product development and then combines the optimalproduct from the perspective of customer utility maximiza-tion and at last returns market sharersquos maximization [5]Matsatsinis and Siskos discuss market share maximizationby historical data (economic or customer survey data) [6]Hu and Bidanda utilize Markov to build sustainable greenproduct lifecyclersquos evolution system and then make decisionsfor new product development from the maximization of thepresent discount value during the whole product lifecycle[7]

The following three areas which cover various influenc-ing factors have been identified as significant and requisite innew product development (a) product attributes specified bydesigners [3 8ndash10] (b) customer requirements and satisfac-tion [11ndash14] and (c) marketing competence [15ndash18]

However none of the currently available models consid-ers all of these areas concurrently Only Chan and Ip [19]integrates three recognized factors and takes the time valueof money into consideration so a new comprehensive modelis proposed on new product development Within Chanrsquosmodel the involved influencing factors are classified intofive parts overall product attractiveness (OPA) overall cus-tomer satisfaction (OCS) word ofmouth (WOM)marketingapproach (MA) and remarketing approach (RA) Based onpurchasing frequency new product customers are divided

into five levels potential customers first-time customersregular customers frequent customers and loyal customers

While the system is running the overall average effec-tiveness of the above five influencing factor is evaluated byexperts or historical data Then the customer purchasingmodel is established by system dynamics and the customerlifetime value (CLV) can be predicted byMarkovmodel Andat last this model could evaluate new product development

22 Analysis of the Research Status Relative to previousresearch Chanrsquosmodel has incomparable superioritywhetheron evaluation perspective and method or the influencingfactor But in order to better meet user needs after furtheranalyzing with the evaluationmethods used in Chanrsquos modelit can easily find out three specific deficiencies which need tobe solved urgently

Deficiency (1) The expertsrsquo workload must be reduced byauxiliary software To be specific whenChanrsquosmodel predictscustomer lifetime value of newproduct the parametersrsquo effec-tiveness such as production cost is basically obtained with theconcerted efforts of experts However there is no availabledata for the product that does not yet enter the marketTherefore it is hardly feasible to investigate and evaluate thoseparameters one by one This is because firstly it will imposelarge workloads to experts and last a long period of time toevaluate each parameter Secondly system developers maynot have necessarily adequate ability to accurately evaluateeach parameter In addition the subjectivity from expertscould easily lead to certain deviations

Deficiency (2) For improving the accuracy of evaluation it isnecessary to distinguish the effectiveness of the influencingfactors for different levels of customers Chan thinks that dif-ferent levels of customers have the same effectiveness of oneinfluencing factor that equals to overall average effectivenessof all levels of customers However in real life the high-frequency purchasing customers often have a relatively highrecognition to the product during their purchasing Whenthe productrsquos overall average effectiveness has reached acertain value itmay not attract the low-frequency purchasingcustomers but it can attract the high-frequency ones

Deficiency (3) Dynamic customersrsquo transition probability ismuchmore realisticWhen stimulating the purchasing Chanuses original nonzero probability as all levels of customersrsquotransition probability Namely the level of (119894) customersrsquotransition probability equals the retained customersrsquo numberat time (119894 minus 1) divided the customersrsquo number at time (119894 minus 2)and it is obvious that the probability is irrelevant to time SoChan believes that different levels of customers have the sametransition probability that is static but dynastic However thereal system is filled with dynamics complexity and productperiodicity As time passed each level of customersrsquo transitionprobability will change within a certain range

This study takes the advantages of CBR and MAHP andintroduces the dynamic transition probability Based on whatis discussed above CampM-CVPM is proposed In responseto the three deficiencies from Chanrsquos model this study puts

The Scientific World Journal 3

forward three solutions which are the main contributionsand innovations simultaneously

Solution (1) CBR in CampM-CVPM helps decision makersto match out the new product which is most similar tothe preexisting products from enterprisersquos historical dataThe preexisting productsrsquo parameters are used for the newproduct If and only if it lacks historical data the expertsrsquoevaluation is required

Solution (2) MAHP in CampM-CVPM distinguishes the effec-tiveness of each influencing factor for different levels ofcustomers though calculating the influencing weight fromexpertsrsquo preference matrix

Solution (3) The dynamic customersrsquo transition probabilityin CampM-CVPM means that the probability is changing ina certain range so that the model has a more realisticsignificance

3 Study Approach

31 CBR (Case Based Reasoning)

311 Overview of CBR CBR is firstly proposed by RogerSchank in 1982 [20] and has become an important machine-learning method in the field of artificial intelligence [21]CBR can obtain the solutions of new problems by adjustingsolutions of old problems similar to new ones [22] Cognitivescience is the logical origin of CBRrsquos theoretical system[23] Cognitive science suggests that human can pass on theperceived information to their brains [24] and then brainsstore and remember the information So in the future ifbrains encounter with one problem which shares the sameor similar stored information brains can quickly providethe information to solve this problem As is known to usthis form of human cognition is a common way to solvethe problem in reality And its development inspires andenlightens the artificial intelligence experts and scholarsgreatly

Based on the above way of human cognition CBRuses the knowledge and experience that happened in theaccumulated cases to provide reference for those similarcases Using CBR to deal with the present case is not entirelyrestarting over again but it matches the past cases with thepresent case and modifies the existing methods in order tomake the method more suitable to the present case

Therefore CBR is more appropriate for those areas suchas imperfect knowledge but relatively rich experience CBRis different from the traditional reasoning method and it cancollect the methods resolved the past problems which is thesame as or similar to the present one whereas the traditionalreasoning method requires complete domain knowledgeCBR is generally divided into four steps which are usuallycalled 4R [25] retrieve reuse revise and retain (see Figure 1)For a new case CBR will retrieve the same and most similarcase from the case vase reuse the case and correspondinglyrevise the suggested solution to get a confirmed solutionfor the new case Simultaneously the amended case will beretained as a new case in the case-base for subsequent uses

New case

Case base

Retrieve

Reus

e

Revise

Retain

Similar case

Confirmed solution

Suggestedsolution

Figure 1 The flowchart of CBR

312 Scope of CBR Nowadays CBR has been playing animportant role in practical applications because of its advan-tages such as the online services applications [26] schedulingand process planning [27 28] hydraulic mechanical design[29] architectural design [30] customer relationship man-agement [31] troubleshooting [32 33] design and implemen-tation of knowledge management [34 35] the prediction ofinformation systems outsourcing success [36] and customerand market plans [37 38] CBR has solved many problemssuccessfully Analyzing and summarizing the findings CBRis mainly applied in the following cases

(1) The large amount of tacit knowledge stored in expertsrsquobrains is not easily extracted Therefore when it isnot convenient to extract knowledge from vast areasquickly and easily CBR can straightly extract thevaluable tacit knowledge out of the expertsrsquo brains tosolve problems

(2) When using rule-based reasoning the amount of tacitrules and professional knowledge stored in expertsrsquobrains is huge So it is easy to cause combinatorialexplosion between each rule and field Eventually itwill lead to reasoning fault and low efficiency How-ever CBR can avoid such combinatorial explosionproblem

(3) While demanding to quickly update the domainknowledge and rules CBR can avoid people beinginvolved in the problems of exponential knowledgeand information by incremental learning methodAnd to a certain extent CBR can also reduce theproblems caused by outdated or incomplete previousexperience and knowledge

(4) When facing incomplete past experience knowledgeand rules or difficult model and structure such aseffective plans of suddenly occurred events CBRcan effectively avoid some problems that are causedby these complex social management or economicsystem

(5) CBR can be applied in the field of knowledge man-agement which attaches great importance to tacitknowledge It is far reaching significant for effectivelymining the tacit knowledge andmaking it into explicitknowledge The knowledge or information carried in


the case from CBR is relatively complete fragmentsand it includes tacit knowledge to be effectivelyminedin knowledge management

313 Application of CBR in This Study With aggravatedglobal competition in order to gain greater competitiveadvantages enterprises have to develop new products amongthe same present products [24] Belecheanu et al point outthat CBR is more suitable than KBS (knowledge systems) tosolve the problem of complex and uncertain new productdevelopment [39] In fact CBR has been widely used in theproduct development at present Therefore it is feasible toapply CBR in this study

32 MAHP MAHP (multiplicative analytic hierarchy pro-cess) is an improvement and extension of AHP (analytichierarchy process) so it is necessary to do an overview ofAHP

321 Overview of AHP In 1970s Saaty a famous Americanoperational research expert proposes analytic hierarchy pro-cess (AHP) which divides multicriteria and multiobjectivedecision making into object criteria and project AHP doesquantitative and qualitative analysis on hierarchy AHP candeeply analyze the nature of complex decision problems fac-tors and intrinsic relationships then use minor quantitativeinformation to make thinking progress mathematical andat last provide an easy way for complex decision makingmixed of qualitative and quantitative problems AHP dividesrelevant elements into goals guidelines and programs andbased on which it makes qualitative and quantitative analysisdecisions [40]

AHP belongs to subjective weighting methods That isthe decision makers compare pairwise the importance formultiple criteria according to their previous knowledge andexperience and then determine the final weight for each crite-rion from comparisonmatrix through relevant mathematicalmodel

322 Overview of MAHP Although AHP has been widelyused in multiobjective decision it still has many shortcom-ings For example when using this method an importantstep is to check whether the comparison matrix meets thecondition of consistency or not namely to revise the compar-isonmatrix to ensure the consistency of matrix However theconsistency always exists in the real world and this becomesthe most fundamental and fatal flaw of AHP In additionthe presence of incompatibility incomplete information andreverse caused by the inconsistency of judgment matrix arealso the flaws of the AHP

To solve these problems in 1990s Lootsma first proposeda method to improve the original AHP known as multi-plicative analytic hierarchy process (MAHP) [41] MAHPmakes up three decencies in AHP-weight synthesizing orderpreserving and weight scaling MAHP can improve deter-mination of the weights under various criteria of decisionmaking [42]

323 Application of MAHP in This Study Therefore thisstudy uses MAHP in order to better determine the weightsof influencing factors for different levels of customers [1 43]

4 CampM-CVPM

41 The System Framework of CampM-CVPM The main func-tions of CampM-CVPM are providing the customer value fornew product development through changing-trend diagramof the net customer lifetime value (NCLV) and the customersrsquonumber The implementation of the functions is divided intothree modules CBR MAHP and NCLV prediction moduleEach module is running by data or related technical supportthroughmodel server (MS) knowledge server (KS) databaseserver (DBS) data warehouse server (DWS) and onlineanalytical processing and data mining server (ODS) (seeFigure 2)

While CampM-CVPM is running CBRmodule could workout the decision-set attributes for the preexisting productwhich ismost similar to the new product and then input theseparameters as the new productrsquos corresponding parameters(1198641 1198642 1198644 and 119864

5) to MAHP module The decision-set

attributes include each parameter and developing experienceThe development can adjust reevaluate and also modifythe inappropriate attributes to gain more satisfying resultsThe development experience is the lessons learned fromthe preexisting product marketing or remarketing experi-ence In addition if the matched similarity is lower thanthe threshold set by experts during running CBR moduleCampM-CVPM will ask for expertsrsquo assistance to provide thevalue of those parameters to be evaluated through expertsrsquoinvestigation or information provided from the relevantservers

CampM-CVPM uses the integrated construction on net-work [44] each server uses both CS (clientserver) modeon which the resources can be shared and services can beprovided to other severs or different clients simultaneously

CampM-CVPMrsquos users include the product developer whois the actual operator of the system the domain expertswho are responsible for stabling and maintaining the sharingresources on all kinds of servers such as the knowledge in KSand the models in MS and the system administrators whoshould be duty bound to maintain the whole system from theinformation technology perspective

The clients correspond to the integrated components oftraditional decision support system the system for problemsynthesis and interaction Firstly the new produce devel-opment is inputted into the clients and then the relevantsystemcontrol program is generated at last all types of serversthrough the network are called to evaluate the new productdevelopment Meanwhile the developer can also query theshared resources on all types of serves to meet their ownneeds MS KS DBS DWS and ODS are responsible forrelated data and technical support

Database server (DBS) updates single userrsquos databaseand database management systems and adds network proto-col communication protocols concurrency control securitymechanisms and other server functions mainly for storing


Developer

Expe

rt CBR

MAHP

NCLV prediction

New

pro

duct

desig

n

Dec

ision

-set

attr

ibut

e

E1E2

E4E5

D(Eic)

New

pro

duct

NCL

V an

d cu

stom

ernu

mbe

r-te

nden

cy g

raph

CampM-CVPM

Client i

Client j

Model server

Network

(MS)

Online analyticalprocessing and datamining to the server

(ODS)

Databaseserver(DBS)

Knowledgeserver(KS)

Datawarehouse

server (DWS)

CSMCRCPC

Figure 2 The system framework of CampM-CVPM

all source data of enterprisersquos production sales and researchand so on DBS provides support for DWS

Data warehouse server (DWS) is composed of ware-house management system (WMS) data warehouse (DW)and analysis tools For better centralizing and simplifyingpart work of DWS and also for more efficiently achievingthe function of DWS analysis tools here do not includeonline analytical processing and data mining Data ware-house mainly stores the data from DBS by the extractingtransforming and loading The data set is subject-orientedintegrated stable and at different time including historicaldata of enterprisersquos all products current data and integrateddata DWS offers support for ODS

Online analytical processing and data mining server(ODS) is independent of DWS During its runtime a largeamount of data extract fromDWS toODS builds a temporarywarehouse then ODS can find out a deeper level of assistantdecision informationODS ismainly responsible for updatingand maintaining preexisting productsrsquo information findingthe relationship between the loss rate of customers (LR)and overall customer satisfaction (OCS) and the relationshipbetween word of mouth (WM) and OCS ODS offers supportfor MS

Model server (MS) based on single userrsquos model baseandmodel-base management system increases network pro-tocol communication protocols security mechanisms andother server functions MS mainly stores all mathematicalmodels and data processing models necessary for CampM-CVPMrsquos threemodules mathematicalmodels includes all thealgorithms and equations such as the similarity calculationequations in CBR while data processing models are usedfor selecting sorting and summarizing the data The modelsare implemented during MS runtime meanwhile the datainformation and knowledge on other severs can also becalled up as required

Knowledge server (KS) upgrades single userrsquos knowledgebase management system knowledge base and inferenceengine and furthermore adds network protocol communi-cation protocols concurrency control security mechanismsand so forth KS is generally responsible for storing a largenumber of production rule knowledge and fact knowledgesuch as the threshold in CBR of matched similarity

42The CBRModule As discussed in the introduction CBRin CampM-CVPM can help decision makers to match out thenew product case which is most similar to the preexistingproduct cases from historical enterprisersquos data [45] Thepreexisting productrsquos parameters are able to use for the newproduct In Chanrsquos model these parameters can be acquiredonly by expertsrsquo evaluation including CS (the cost of thegoods) MC (the marketing cost) RC (remarketing cost) PC(the number of potential customers) RP (the retail price ofthe product) 119864

1(overall product attractiveness) 119864

2(overall

customer satisfaction) 1198644(marketing effectiveness) and 119864

5

(remarketing effectiveness)The key techniques in CBR are the indexation for

case representation and the similarity evaluation [46] andcan efficiently match out the most similar preexisting casefrom the case-base with the new product Therefore fromthese two aspects the section below will illustrate howCBR helps CampM-CVPM obtain the above parameters to bepredicted

421 Case Representation There are many approaches forcase representation and the knowledge expression methodin artificial intelligence An overwhelming majority of casesin intelligent case-base use the frame representation Inreality identification of different products can be doneby their structural features (including specific functions


Knowledge of product developmentRetail price of the product (RP)

The number of potential customers (PC)

Remarketing approach

Remarketing effectiveness (E5)

Overall customer satisfaction (E2)Overall product attractiveness (E1)

Marketing effectiveness (E4)

The cost of the goods (CS)

Unit 1

The marketing cost (MC)The remarketing cost (RC)

Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s

Condition-set attributeDecision-set attribute

Figure 3 The frame representation results for the product case

appearance and quality and implementation techniques)So the frame representation can obviously better meet theproductrsquos characteristics Because the frame representationexpresses the same thoughts and ideas with CBR this studyuses the framework representation to express the productcase in CBR module (see Figure 3) The dotted boxesrepresent the condition-set attributes and the solid boxesrepresent the decision-set attributes (ie the parameters tobe predicted) CBR will automatically work out the similarityof the condition-set attributes between new and preexistingproducts and then output the decision-set attributes whichhave maximum similarity as the corresponding parametersfor new product

422 Similarity Evaluation There are many algorithmsto evaluate the similarity of the productrsquos condition-setattributes Gu et al [47] propose the FRAWOwhich has obvi-ous advantages over the traditional approaches in retrievalefficiency and the quality of retrieval results Thereforemaking use of FRAWO CampM-CVPM can realize similarityevaluation between the products Specific processes are asfollows

Preexisting product cases are stored in the product case-base Assuming that the product case-base has 119898 cases thecase (119894)mdash119883

119894has 119899 attributes values such as 119883

1198941 1198831198942 119883

119894119899

and the values of 119899 condition-set attributes for the targetnew product case 119866 are 119866

1 1198662 119866

119899 then the condition-

set attributes matrix 119863(119883119894119895) can be obtained by (1) And

the average of each columnof119863(119883119894119895) is calculated by (2) with

the intermediate variable being (3)

119863(119883119894119895)

= [119883

119866]

=

119899 conditionminus set attributes⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞

[[[[[[

[

11988311

11988312

sdot sdot sdot 1198831119899

1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661

1198662sdot sdot sdot 119866

119899

]]]]]]

]

119898 cases in case base+ target case 119866

(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)

The normalized effectiveness matrix 119863(1198831015840119894119895) is expressed in

(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)


The similarity between the preexisting product case (119894) andthe target new product case is determined by (6)

sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840

119895are symbol properties

exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816

max (119894) minusmin (119894)) when 1198831015840

119894119895and 1198661015840

119895are certainly numeric properties

119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840

119895fuzzy properties

(6)

119886 stands for the surface of corresponding subordinatefunctions and 119886(1198831015840

119894119895cap 1198661015840

119895) stands for the intersection of two

fuzzy surfaces119908119895is the matched weight of the similarity for the

condition-set attributes and its initial value is set by expertsbased on their experience When using the correspondingparameters for new product the strategy for adjusting theweight is for those preexisting product cases whose similarityexceeds the threshold set by experts if the CampM-CVPMrsquospredictive result is correct the system must increase theweights of condition-set attributes which are of the samevalue as the target new product case and every increscentis Δ119894119896

119888 119896119888stands for the times that the preexisting product

case is properly matched while Δ119894 is the adjustment range ofweights set by expertsrsquo experience Of course Δ119894 can also beadjusted as actual needs

43 MAHP Module As mentioned earlier MAHP in CampM-CVPM can calculate the weight of some influencing factorsfor different levels of customers through expertsrsquo preferencematrix and then work out the gap between the effectivenessof this influencing factor and overall average effectivenessFinally the effectiveness of this influencing factor for differentlevels of customers can be obtained The specific proceduresare as follows

The overall average effectiveness of these five influencingfactors for new product is set as 119864

1(overall product attrac-

tiveness OPA) 1198642(overall customer satisfaction OCS) 119864

3

(word of mouth WOM) 1198644(marketing effectiveness MA)

and 1198645(remarketing effectiveness RA) The weight for the

influencing factor (119888) for level (119894) customers is expressed as119908119894

where 119894 = 1 2 5 represents potential first-time regularfrequent and loyal customers and 119888 = 1 2 5 representsthe five influencing factors OPA OCS WOM MA and RATotally 119896 experts participate in the evaluation according toTable 1 the expert (119901) (119901 = 1 2 119896) should determine theinfluencing gap 120575(119901)

119894119895between level (119894) and level (119895) customers

of the influencing factor (119888)

Table 1 The influencing gap 120575(119901)119894119895

on different levels of customers

120575(119901)

119894119895 Definition

0 Influence of factor (119888) on customer (119894) is the same ascustomer (119895)

2 Influence of factor (119888) on customer (119894) is a little bigger thancustomer (119895)

4 Influence of factor (119888) on customer (119894) is a bit bigger thancustomer (119895)

6 Influence of factor (119888) on customer (119894) is much bigger thancustomer (119895)

8 Influence of factor (119888) on customer (119894) is rather bigger thancustomer (119895)

The preference matrix for the expert (119901) is expressed in(7) and the relationship between 120575(119901)

119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888

is stated in (8) The constrain is determined by (9)Consider

119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)

Using a logarithmic least squares estimation the influencingweight from the expert (119901) for level (119894) customers can bepredicted by (10) The influencing weight of all 119896 expertsfor level (119894) customers is expressed mathematically by (11)


First time customers Active customersRetention rate

Design

Performance

Packaging

Competitiveness

Quality

Overall product attractiveness

Marketing effectiveness

Word of mouth

Design importance

Attractiveness of product design

Performanceimportance

Attractiveness of product performance

Packaging importance

Attractiveness of product packaging

Competitivenessimportance

Attractiveness of product

competitiveness

Quality importanceAttractiveness of

product quality

Total active customers

Loss rate

Marketing approach 1 Marketing

approach 2 Marketing approach 3

Remarketing approach 1



Remarketing effectiveness

Customersatisfaction

Potential customers

Acquisition rate

Figure 4 The stimulation figure of customer purchasing

By normalization the weight of the influencing factor (119888) forlevel (119894) customers is determined by (12) Consider

119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)

As in Chanrsquos model different levels of customers have thesame effectiveness of the influencing factor and every levelrsquoseffectiveness of the influencing factor (119888) is 119864

1198885 After divid-

ing by MAHP the allocated effectiveness of the influencingfactor (119888) for level (119894) customers is 119908

119894119888sdot 119864119888 Therefore the gap

of between the effectiveness of the influencing factor (119888) andthe overall average of level (119894) customers is stated in (13) IfΔ119864119894119888lt 0 the effectiveness does not reach the overall average

if Δ119864119894119888gt 0 it exceeds the overall average if Δ119864

119894119888= 0 it

exactly equates to the overall average 119863(119864119894119888) stands for the

effectiveness matrix to each influencing factor for differentlevels of customers and then 119864

119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)

44 The Predictive Module for NCLV The main functionsof this module are to predict the new product customervalue by stimulating customer purchasing and inputting theparameters those are the output of CBR andMAHPmodules(see Figure 4) The parameters to be stimulated include CSMC RC PC RP and119863(119864

119894119888)

The stimulation of customer purchasing used in thepredictive module for NCLV (net customer lifetime value)has the same basic idea in line with Chanrsquos model proposedby system dynamics But when calculating CLV (customerlifetime value) byMarkovmodel the randomly dynamic cus-tomersrsquo transition probability replaces the static probabilityCorrespondingly the relationships between some nodes areadjusted appropriately for better logicality understandabilityand practicability The following is the specific process of thesimulation

Following Chanrsquos study NCLV (net customer lifetimevalue) is defined as the sum of the current lifetime values ofall customers where customers lifetime value refers to thepresent value of future profit from a customer

As mentioned earlier new productrsquos customers aredivided into five levels potential first-time regular frequentand loyal customers In Chanrsquos model the second to thefourth level customer is regarded as a whole that is activecustomer the customersrsquo increase is derived from the secondlevel Within Chanrsquos formula the customersrsquo retention whichrefers to the increment from the second level to the thirdfourth and fifth level used the numbers of the third fourthand the fifth level from the input port But apparently forone specific level the increase in purchasing frequency andtransiting quantity to next level must be affected by the levelitself rather than the next levelHence on one hand this studyfollows the calculation equations in Chanrsquos model to ran-domly obtain the ldquoretention raterdquo (RR) and the ldquoacquisitionraterdquo (AR) (see (16)) On the other hand the relationships andcustomers number must be made appropriate adjustments

Obviously the influence factors for different customerrsquoslevels are not the same The potential customers of the prod-uct make purchasing on the basis of OPA ME and WOMThe first-time and regular customers are influenced by OPAWOM RE and OCS The frequent customers are affectedby OPA RE and OCS However the loyal customers only


The customersrsquo levels The influencing factors

C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)

Figure 5The relationships between customersrsquo levels and influenc-ing factors

Potential customersC1t

First-time customersC2t

Regular customersC3t

Frequent customersC4t

Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers

Figure 6 The relationship between different levels

trust enterprisersquos certain product because of their completeallegiance and will not look for other alternatives So ingeneral the loyal customers are not affected by any factors (seeFigure 5)

In CampM-CVPM the customer purchasing behavior fornew product is divided into three statues The first statueis that the customers will no longer buy this new productbecause of dissatisfaction thereby the first level customersmean the lost ones who will become the potential customersThe second statue is that the customers who are very satisfiedwith the new product will increase purchasing frequencyfor the next time and thus come into the next level Thethird statue is that the customers who neither like nor dislikethe new product will keep the former purchasing frequencyMost notably the lost customers who are not satisfied withthis new product do not mean that they will never buy itin future they may be affected by other various factors andpurchase this new product again The relationship betweendifferent levels is shown as in Figure 6

119862119904119905denotes the number of customers in level (119904) at time

(119905) AR119904119905

means the acquisition rate of the customers whotransit to the next level in level (119904) at time (119905)119860

119904119905indicates the

transition number of customers in level (119904) at time (119905) LR119904119905

implies the loss rate of customers in level (119904) at time (119905) 119871119904119905

describes the transition number of customers in level (119904) attime (119905) The relationships between 119862

119904119905 AR119904119905 LR119904119905 and 119871

119904119905

are expressed in (15)

119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)

The relationships between the five influencing factorsthe ldquoacquisition raterdquo (AR) and the ldquoloss raterdquo (LR) can beobserved from Figure 7

It is easy to know that word of mouth (1198643) and LR

are mainly decided by overall customer satisfaction (1198642)

Therefore the relationships between 1198642and 119864

3 LR and 119864

2

can be acquired through data mining of historical productSo once the value of 119864

2is worked out by CBR similarity

matching then the value of 1198643and LR could be generated

automatically The relationships between the unity matrix119863(119864119894119888) and the acquisition rate (AR) are expressed in (16)

AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)

Assumptions are as follows one customers buy newproducts under noncontractual relationship That is to saycustomers purchasing is entirely based on their own prefer-ences two all customers make purchasing decisions at thesame time point and the time interval (119889

119905) is equivalent

Hence 119862119904119905 the number of customers at level (119904) at time

(119905) are determined by (17)

1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate

Remarketingeffectiveness

Overall customersatisfaction

Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t

Figure 7 The relationship between influencing factors and AR or LR

1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)

Set 119863 as the current discount rate by the randomly dynamictransition probability (18) is applied to calculate the NCLV

NCLV = (infin

sum

119905=0

[1198622119905times (RP minusMC minus CS)

+ (1198623119905+ 1198624119905) times (RP minus RC minus CS)

+1198625119905times (RP minus CS)]) times ((1 + 119863)119905)

minus1

(18)

45 CampM-CVPMrsquos Characteristics and Application Scope

451 CampM-CVPMrsquos Characteristics To overcome the abovethree shortcomings in Chanrsquos model this study proposesa CBR-based and MAHP-based customer value predictionmodel for new product development (CampM-CVPM) Thefollowing are the new modelrsquos characteristics

(1) CampM-CVPM evaluates the quality of new productdevelopment by using the net customer lifetime value(NCLV) which is in line with the customer-orientedmarket tendency Therefore it has an advantage overformers studies in the prediction perspective

(2) CampM-CVPM considers the time value of profitsbringing by customers Hence it would work out

the net present customer value of new product devel-opment

(3) CampM-CVPM stimulates the customer purchasingbehavior with fully considering all the three recog-nized influencing factors for new product develop-ment which are product customer and market

(4) CampM-CVPM reduces the expertsrsquo workload by usingCBR so that removes the unnecessary evaluation ofeach influencing factor on each product

(5) CampM-CVPM shortens the time to predict the cus-tomer value for new product development simulta-neously improving the evaluationrsquos agility

(6) CampM-CVPM by using MAHP distinguishes theeffectiveness of each influencing factor for differentlevels of customers

(7) CampM-CVPM uses dynamic customersrsquo transitionprobability to simulate customer purchasing behav-ior which can make the model more realistic andauthenticity

(8) CampM-CVPM by using CBR avoids the samemistakein the developing process as low as possible andit can also promote the retention and heritance forenterprisesrsquo tacit knowledge to prevent possible lossescaused by a mature technology leaving the companyAlso CampM-CVPM can provide references for thenew product development which is conducive to theexistence and improvement of development level

(9) CampM-CVPM takes the marketing cost into accountto evaluate the customer value and this is also help-ful for knowledge exchanging and sharing betweenenterprises design and marketing department

(10) CampM-CVPM expands the system usersrsquo scope so thatthe ordinarily nonexpert developer could also predict


the new productrsquos customer value to better satisfy userneeds

(11) CampM-CVPM by using CBR can solve the newproblems as human thinking mode with the aid ofthe formerly accumulated knowledge and experienceBasically it can eliminate bottlenecks in retrievaldue to knowledge and information index growthfurthermore the previous case is relatively easy to becollected

452 The Application Scope of CampM-CVPM AlthoughCampM-CVPMhasmany advantages over former studies thereare still some limitations for the application scope of CampM-CVPM

(1) There is a premised assumption for CBR applying toCampM-CVPM that is similar problems have similarsolutions So new product development to be pre-dicted must have certain similarity with the cases incase-base For example food products in case-basecannot be used to predict commodity products Andit may not make sense even if the case to be predictedand the case in case-base belong to one same type Forexample if predicting electric toothbrush by ordinarytoothbrush the effectmay not be ideal and vice versa

(2) The new product customers are divided into fivelevels potential first-time regular frequent and loyalcustomers However this division may not be appliedto some products such as household appliances thatthe customer will not buy another in the short termSo the new product customers might be divided intotwo or three levels as well At the same time thoseproducts which own five-level customers usuallybelong to FMCG (fast moving consumer goods)

(3) If the case-base has insufficient product cases thepredictive results may be not accurate enough soCampM-CVPM is more suitable for those enterprisesthat already have a lot of mature products

(4) From the foregoing sufficient mature cases providea strong guarantee for the good predictive resultsConsequently building the original case-base is ahuge project that needs to consume certain humanmaterial and financial resources Enterprisersquos deci-sion whether to use CampM-CVPMor not is dependingon its own situation

(5) CBR based on the incremental learning has anautomatic-learning mechanism with the formerlyaccumulated knowledge and experience But seenin another way because of unconditionally passivelearning each reserved case in the case-base may eas-ily lead to an unmanageable state Correspondinglythe system will run with a low efficiency and theretrieval cost is going up In conclusion CBR shouldtry active learning in the face of a large number ofsamples

(6) The Markov model that is applied to calculate cus-tomer lifetime value in CampM-CVPM is the most

flexible model within the present studies but it stillconstitutes some limitations For example during thetransaction between customers and enterprises eachinterval time is same and fixed

5 Simulation Experiment and Analysis

In this section to verify CampM-CVPMrsquos validity a simulationexperiment is conducted to predict the NCLV for some regu-lar toothbrush The computer auxiliary software is MATLABR2010a Firstly this section describes the reasons why wechoose regular toothbrush as the simulation case Secondlyit gives out the experimentrsquos general design including thepurposes preparations and producers At last it analyzes theexperimentrsquos results in detail

51 The Simulation Case In order to make the experimentmore intuitive and understandable and according to CampM-CVPMrsquos assumption a new regular toothbrush is chosen asour experimental case Here are the reasons (1) to someextent toothbrush has become the necessities of life (2)toothbrush belongs to FMCG (fast moving consumer goods)because of its regular replacement (3) there are two maintypes in the market regular and electric toothbrush andregular toothbrush gains relatively larger market share due tothe price factor and peoplersquos habits

52 The Experimentrsquos General Design

521 The Experimentrsquos Purpose CampM-CVPMrsquos first majoradvantage lies in reducing the expertsrsquo workload during newproduct development With the introduction of CBR theparameters of the running system are decided by the mostsimilar historical productrsquos data automatically provided bysystem rather than totally depending on expertsrsquo evaluationIf and only if the similarity is less than expertsrsquo settingthreshold experts just do the evaluation The above advan-tage combining with the latter two advantages improvesthe accuracy of the system Thus verifying the validity ofCampM-CVPM is equal to verifying the reduction of expertsrsquoworkload and the increment of system accuracy compared toChanrsquos model

However on the other hand the accuracy for the systemproposed by Chan is strongly subjective because it is morelikely to rely on the expertsrsquo ability In reality the accuracy ofthe running systemmay be differentwhen one expert predictsdifferent products or different experts predict one productTherefore one or a few experiments are meaningless and farfrom enough and it cannot distinguish the good developmentfrom the bad For another handwhen runningCampM-CVPMthe system accuracy can be verified very well with comparingthe predictive value and the actual value for product customervalue

The purpose of this experiment is to test that (a) CampM-CVPM can reduce the expertsrsquo workload when predictingnewproduct customer value for a regular toothbrush and that(b) the deviation from the predictive to the actual value canbe acceptable in reasonable range


522 The Experimentrsquos Preparations

Experiment Tool This study uses MATLAB that has strongprocessing power and ease of use as computer auxiliary toolto verify the CampM-CVPMrsquos validity

Case Representation According to productrsquos structural fea-tures of the regular toothbrush the frame representation canbe expressed as Figure 8

Data Preparation CBR is an automatic machine-learningtechnology and it can get the overall average of new productrsquosinfluencing factors and stimulate other running modelsrsquoparameters by analyzing the inherent rule from historicalproductsrsquo data so the experiment results seldom rely onauthentication of the product data

According to productrsquos structural features and theirinherent relationships this experiment randomly generates70 toothbrush cases as the classic case-base In order toenhance the resultrsquos trustworthiness the study does strictvalidity design Except randomly generating experimentaldata the data for automatic learning is separated from thedata for testing the CampM-CVPMrsquos running accuracy 60 casesof the 70 classic cases are served as a case-base (enterprisersquospreexisting products) and the other 10 cases are served as testcase (new products) 10 cases as a batch and the case-base canbe constructed by inputting six batches and each batch mustdo once validity verification after the input is complete

For the sake of the experiment objectivity this studyinvites three professors and twodoctors fromBusiness Schoolof Central SouthUniversity as the expert teamThey compareall levels of customersrsquo importance of each influencing factorand then provide 25 importance comparison tables Fromthose tables the effectiveness matrix119863(119864

119894119888) can be got

Things to note are as follows

(1) The initial value of matching weight for eachcondition-set attributersquos similarity is set to 1119899 where119899 refers to the number of product casesrsquo condition-set attributes namely each condition-set attributehas equal contribution to similarity evaluation at thebeginning

(2) To adjust theweight to the best as soon as possible theweight adjustment Δ119894 is set to 002 for the first threebatches and 001 for the last three batches

(3) The cases are increasing along with more and moreexperiments and the weights of each decision-setvalue have constant adjustment The later on theCampM-CVPM going the larger similarity matchedby CBR module is In order to increase the validityfor each experiment comparing results the detectedthreshold is set to 04 for the first three experimentsand 085 for the last three experiments

(4) The accuracy of CampM-CVPM depends on the NCLVdeviation rate of the predictive result for new producttest case Because CampM-CVPM has used the mostsimilar case in case-base as the test case the NCLVdeviation rate measures how far the predictive NCLVvalue from the actual value for the test case is

(5) The criteria based on experience to judge the consis-tency of similarity matched results are that the NCLVdeviation rate is 3 at most

(6) 119889119905is set to 3 months because the dentist advises that

toothbrush should be changed every three monthsAnd the potential customersrsquo number is set to 8million

(7) Per unit of expertsrsquo workload equals to the expertsworkload to predict the parameters CS MC RC PCRP119864111986421198644 and119864

5 when running theChanrsquosmodel

one time

53 The Experimentrsquos Procedures

(1) 10 test cases are numbered sequentially from 1 to 10and six batches for constructing the case-base arenumbered sequentially from 1 to 6

(2) The initial value of matching weight for eachcondition-set attributersquos similarity in CBR module isset to 1119899 where 119899 refers to the number of productcasesrsquo condition-set attributes

(3) The expert team is responsible for comparing all levelsof customersrsquo importance of each influencing factorsand then provides their preference matrix Finally theeffectiveness matrix 119863(119864

119894119888) can be calculated by (8)ndash

(10)(4) The first batch cases are inputted into case-base while

doing the first experiment For each inputted case thematching weight of similarity during each condition-set attribute in CBR module must be adjusted oncewhere Δ119894 is set to 02

(5) After inputting 10 test cases the following data mustbe recorded for each case (a) the highest similarityafter each test case matching all the cases (b) thepredictive NCLV deviation rate for each case (c)the number of new produce development by CampM-CVPMrsquos prediction with the deviation rate limited to5most and (d) the expertsrsquo workloadunits inCampM-CVPM when the cases in case-base are not sufficient

(6) The average for NCLV deviation rate refers tothe reduced percentage for the expertsrsquo workloadAssume that the expertsrsquo workload in Chanrsquos model is10 units for 10 text cases so the reduced percentage =(10 minus the expertsrsquo workload units in one CampM-CVPMexperiment)10

(7) Inputting the next batch the matching weight ofsimilarity during each condition-set attribute in CBRmodule must be adjusted once where Δ119894 is set to02 for the former three experiments and 01 for thelatter ones Repeat steps (4)ndash(6) to finish all the sixexperiments

54 The Experimentrsquos Results

541 The Maximum Similarity The similarity comparisonresults of six tests are shown in Figure 9 The abscissa


Package approach

Retail price of the product (RP)


Re-marketing approach





Knowledge of product development

The color of toothbrush hair


Marketing approach

The material of toothbrush hairThe row number of toothbrush hairThe arrangement style of toothbrush hairThe height of toothbrush hairExchangeable toothbrush headThe shape of toothbrush headThe theme colors of toothbrush handThe shape of toothbrush handThe materials of toothbrush handThe antiskid design of toothbrush handThe length of toothbrush handThe angle of toothbrush hand

(Approach 1 approach 2 and approach 3)

(Curving and straight)

(Square elliptical shape diamond shape)(Nonreplaceable plug-in and embedded)

(Two rows three rows and four rows)

Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )

(9sim13mm)(Crenate full-contact )

(White + blue white + green )

(Design1 design 2 )(80sim110mm)

A X

X re

gula

r too

thbr

ush

prod

uct c

ase

(PP PBT cordura nylon )

Figure 8 The frame representation of a regular toothbrush

represents experiment sequence and the ordinate representsthe value of the similarityThepoints on the dotted linesmeanthe maximum similarity after each test case matching all thecases and the points on the solid lines mean the average forall the maximum similarity

Along with the increasing cases the matching weightof similarity during each condition-set attribute in CBRmodule has been constantly adjusted and obviously remainsrising When Δ119894 is fixed the increase range is progressivelydecreasing for both the former three experiments and thelatter three More specifically the increase range for the thirdand the sixth experiment is inconspicuous The result isfurther evidence of the validity that Δ119894 is adjusted from 02 to01 The maximum similarity for the sixth experiment trendstoward 09 and it illustrates that a certain amount of cases hasbeen included in the case-base On one hand each case couldretrieve the case very similar to itself on the other hand 60cases selected in the case-base are feasible

542 The NCLV Deviation Rate In Figure 10 the abscissarepresents experiment sequence and the ordinate representsthe NCLV deviation rate The points on dotted lines meanthe ratio from the NCLV actual value to the predictive valuefor each test case in CampM-CVPMrsquos every experiment andthe points on solid lines mean the average for overall NCLVdeviation rates

The cases are increasing along with more and moreexperiments but the NCLV deviation rate remains decliningobviously as well as the average of overall NCLV deviationrates which are slowly falling and going to 0 at last Forone thing it demonstrates that six experiments are feasiblefor another CampM-CVPMrsquos capability to correctly predicttest casesrsquo NCLV is progressively enhancing Hence CampM-CVPM can always better play its unique advantage in pre-dicting new produce development for those enterprises thathave owned plenty of mature products

543 The CampM-CVPMrsquos Validity From a qualitative angle(see Figures 9 and 10) CampM-CVPMrsquos capability to match outthe most similar and accurate case is gradually strengthenedwith the casesrsquo incensement and the similarityrsquos matchingweight adjustment From a quantitative angle Table 2 canbetter illustrate the above advantage by the related data

Along with continues case numberrsquos growth and theweightrsquos adjustment even the detected thresholds for thelatter three experiments clearly improve better than theformer three CampM-CVPMrsquos ability to effectively reduce theexpertsrsquo workload and actually predict customer value is stillimproving (see Table 2) It is also suggested that CampM-CVPMalways plays its unique advantage in predicting new producedevelopment for those enterprises that have owned plenty ofmature products


Table 2 The verified results of the validity of CampM-CVPM

Tests The threshold The productsnumber

Expertsrsquo evaluationtime

The reduction forexpertsrsquo workload

()

The average ofoverall NCLV

deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests

The mean of NCLV deviation rateThe 10th testing productThe 9th testing productThe 8th testing productThe 7th testing productThe 6th testing product

The 5th testing productThe 4th testing productThe 3rd testing productThe 2nd testing productThe 1st testing product

The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6

Figure 9 The similarity comparison results of six tests

To further clarify the accuracy that CampM-CVPM canpredict the new product customer value from all six experi-ments this study chooses the 4th test case with themaximumsimilarity and the 10th test case with the minimum similarityand then compares their NCLVrsquos predictive and actual values(see Figures 11 and 12) The dotted lines mean the predictivenumber of all levels customers and the solid lines mean theactual number

This study compares the NCLV predictive value with theactual values only for the first 24 time points (3 months as atime point and 24 points equal to 6 years) The comparisonresults for only 24 time points can be more intuitive andfurthermore it is not need to focus on the NCLVrsquos valueduring all product life cycle From Figures 11 and 12 it iseasy to know that although there is a small deviation in

The meanThe 10th testing productThe 9th testing productThe 8th testing productThe 7th testing productThe 6th testing product


Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)

Figure 10The comparison results of the NCLV deviation rate of sixtests

predicting customersrsquo number it is still accurate in predictingNCLVrsquos value That is CampM-CVPMrsquos ability to actuallypredict customer value for new product development can bewell-trusted

544 Other Results Because of limited samples and sim-ulated experiments the accuracy to predictive customersrsquonumber in CampM-CVPM is not obvious at present But withafter further analysis of Figure 13 the potential customeris active and the first-time customer is rapidly increasingwhen the previous year for selling new product Moreoverboth of them start falling one year later Consequentlyenterprises must lay stress on marketing competence fornew product which can obtain much more customer value


0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

The real value of C1


The real value of C3 + C4 + C5

The real value of NCLVThe predicted value of C1

The predicted value of C2

The predicted value of C3 + C4 + C5

The predicted value of NCLV

times105

2

18

16

14

12

1

08

06

04

02

0

Figure 11 The comparison results for the test case with maximumsimilarity

More specifically it is more suitable for enterprises to doadvertisement sales exhibition internet marketing and soon While one year later the first-time customers start toreduce along with that the active customers begin to increaseNamely enterprise strategy should pay more focus on retain-ing customers and provide more remarketing activities suchas affiliate campaigns

Besides Figure 13 also makes clear that the active cus-tomersrsquo number has reached maximum at the 15th timepoint (around the 375th year) This indicates that the targetmarket is already saturated which means that customers willnot increase their purchase frequency and be not attractedanymore therefore seldom new potential customers willcome in In such a situation enterprises should developmorenew products to attract and retain customers On the otherhand this is also the reason why constant developing newproduct is the key of enterprisersquos core competence

55 The Experimentrsquos Summary Notwithstanding if thisstudy can get a lot of specific enterprisesrsquo historical data andcompare the accuracy with Chanrsquos model many times thevalidity of CampM-CVPM can be verified more thorough Butunder the condition of the limited human material andfinancial resources the randomly generated data to verify theCampM-CVPMrsquos validity is feasible in fact








0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0

Figure 12 The comparison results for the test case with minimumsimilarity

Also this study can be treated as a periodically funda-mental work on the current condition Here are the tworeasons (a) the system running results seldom rely on thedatarsquos authenticity because CBR can automatically obtain thevalues of the influencing factors through historical data and(b) in order to enhance the resultrsquos trustworthiness the studydoes strict validity design not only randomly generatingexperiment data according to the product features but alsoseparating the data for automatic learning from the data usedfor testing the running accuracy of CampM-CVPM

6 Conclusions

To sum up better than Chanrsquos model CampM-CVPM distin-guishes the effectiveness of the influencing factors for differ-ent levels of customers and thinks about the randomvariationin a certain range and meanwhile reduces the expertsrsquoworkload to investigate and evaluate the required parameterswith the lack of historical data for new productrsquos customervalue Besides better than Chanrsquos model CampM-CVPM hasthe following four functions (a) it expands the system usersrsquoscope so that the ordinarily nonexpert developers can alsoevaluate the new productrsquos customer value Meanwhile thesame mistakes can be possibly prevented from happening toimprove the accuracy of problem solution (b) it promotesthe retention and heritance for enterprisersquos tacit knowledge


8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15

(a) The actual customer number of the fourth test case

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104

(b) The predictive customer number of the fourth test case

Figure 13 The tendency figure of the actual and predictive customers number of the test case (4)

(c) it outputs the similar product developmentrsquos experiencefor decision-maker reference which can help to enhance thedevelopers skill and also gather enterprise marketing teamrsquosknowledge to promote information sharing (d) it eliminatesbottlenecks for knowledge retrieval

In conclusion superior to Chanrsquos model CampM-CVPMcan better satisfy user needs and urge the decision progressmore scientification procedures and automation

The purpose of CampM-CVPM as a periodically funda-mental work has been achieved Future research could gatherlarger andmore complex practical samplesMeanwhile morework is needed to obtain the specific enterprisersquos historicaldata to test the validity of CampM-CVPM and compare rep-etitiously with Chanrsquos model However the divisions of fivelevels customers in CampM-CVPM may not apply to someproducts expect FMCG (fast moving consumer goods) TheMarkov model in CampM-CVPM which is the most flexiblemodel within the present models used to calculate customerlifetime value still constitutes some limitation such as thateach interval time is same and fixed

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

The authors would like to express their sincere thanks tothe Central South University for its financial support of this

research work under project that is supported by the Innova-tion Group Project National Natural Science Foundation ofChina (Grant no 71221061)

References

[1] D F Cox and R E Good ldquoHow to build a marketinginformation systemrdquoHarvard Business Review vol 45 no 3 pp145ndash154 1967

[2] J I N Zhi L I U Lin and J I N Ying Software RequirementsEngineering Principles and Methods Science Press BeijingChina 2008

[3] L Xu Z Li S Li and F Tang ldquoA decision support system forproduct design in concurrent engineeringrdquo Decision SupportSystems vol 42 no 4 pp 2029ndash2042 2007

[4] B Besharati S Azarm and P K Kannan ldquoA decision supportsystem for product design selection a generalized purchasemodeling approachrdquoDecision Support Systems vol 42 no 1 pp333ndash350 2006

[5] G Alexouda ldquoA user-friendly marketing decision support sys-tem for the product line design using evolutionary algorithmsrdquoDecision Support Systems vol 38 no 4 pp 495ndash509 2005

[6] N F Matsatsinis and Y Siskos ldquoMARKEX an intelligentdecision support system for product development decisionsrdquoEuropean Journal of Operational Research vol 113 no 2 pp336ndash354 1999

[7] G Hu and B Bidanda ldquoModeling sustainable product lifecycledecision support systemsrdquo International Journal of ProductionEconomics vol 122 no 1 pp 366ndash375 2009

[8] C Kahraman T Ertay and G Buyukozkan ldquoA fuzzy optimiza-tion model for QFD planning process using analytic networkapproachrdquo European Journal of Operational Research vol 171no 2 pp 390ndash411 2006


[9] M R Solomon G W Marshall and E W Stuart MarketingReal People Choices Pearson Education Upper Saddle River NJUSA 6th edition 2010

[10] L Xu Z Li S Li and F Tang ldquoA polychromatic sets approach tothe conceptual design ofmachine toolsrdquo International Journal ofProduction Research vol 43 no 12 pp 2397ndash2421 2005

[11] A Herrmann F Huber and C Braunstein ldquoMarket-drivenproduct and service design Bridging the gap between customerneeds quality management and customer satisfactionrdquo Inter-national Journal of Production Economics vol 66 no 1 pp 77ndash96 2000

[12] X Liu W J Zhang Y L Tu and R Jiang ldquoAn analyti-cal approach to customer requirement satisfaction in designspecification developmentrdquo IEEE Transactions on EngineeringManagement vol 55 no 1 pp 94ndash102 2008

[13] A SMelissa StrategicManagement of Technological InnovationInternational Edition McGraw-Hill New York NY USA 2005

[14] M A Schilling Strategic Management of Technological Innova-tion McGraw-HillIrwin New York NY USA 2005

[15] M J Cooper ldquoAn evaluation system for project selectionrdquoResearch Management vol 21 no 4 pp 29ndash33 1978

[16] UDeBrentani andCDroge ldquoDeterminants of the newproductscreening decision a structural model analysisrdquo InternationalJournal of Research in Marketing vol 5 no 2 pp 91ndash106 1988

[17] C Lin and C Chen ldquoNew product gono-go evaluation at thefront end a fuzzy linguistic approachrdquo IEEE Transactions onEngineering Management vol 51 no 2 pp 197ndash207 2004

[18] J P Peter and J C Olston Consumer Behavior and MarketingStrategy McGraw-Hill Irwin New York NY USA 8th edition2008

[19] S L Chan and W H Ip ldquoA dynamic decision support systemto predict the value of customer for new product developmentrdquoDecision Support Systems vol 52 no 1 pp 178ndash188 2011

[20] C K Riesbeck and R C Schank Inside Case-Based ReasoningPsychology Press 2013

[21] C S Park and I A Han ldquoA case-based reasoning withthe feature weights derived by analytic hierarchy process forbankruptcy predictionrdquo Expert Systems with Applications vol23 no 3 pp 255ndash264 2002

[22] S Craw ldquoCase-based reasoningrdquo in Encyclopedia of MachineLearning pp 147ndash154 Springer 2010

[23] H Tseng C Chang and S Chang ldquoApplying case-basedreasoning for product configuration in mass customizationenvironmentsrdquo Expert Systems with Applications vol 29 no 4pp 913ndash925 2005

[24] C Ching-Chin A I Ka IengW Ling-Ling and K Ling-ChiehldquoDesigning a decision-support system for new product salesforecastingrdquo Expert Systems with Applications vol 37 no 2 pp1654ndash1665 2010

[25] A Aamodt and E Plaza ldquoCase-based reasoning foundationalissues methodological variations and system approachesrdquo AICommunications vol 7 no 1 pp 39ndash59 1994

[26] M H Goker and T Roth-Berghofer ldquoDevelopment and uti-lization of the case-based help-desk support system HOMERrdquoEngineering Applications of Artificial Intelligence vol 12 no 6pp 665ndash680 1999

[27] H-C Chang L Dong F X Liu and W F Lu ldquoIndexingand retrieval in machining process planning using case-basedreasoningrdquoArtificial Intelligence in Engineering vol 14 no 1 pp1ndash13 2000

[28] G Schmidt ldquoCase-based reasoning for production schedulingrdquoInternational Journal of Production Economics vol 56-57 pp537ndash546 1998

[29] CM Vong T P Leung and P KWong ldquoCase-based reasoningand adaptation in hydraulic production machine designrdquo Engi-neering Applications of Artificial Intelligence vol 15 no 6 pp567ndash585 2002

[30] A Heylighen and H Neuckermans ldquoA case base of case-baseddesign tools for architecturerdquo Computer Aided Design vol 33no 14 pp 1111ndash1122 2001

[31] K L Choy W B Lee and V Lo ldquoDesign of an intelligentsupplier relationship management system a hybrid case basedneural network approachrdquo Expert Systems with Applicationsvol 24 no 2 pp 225ndash237 2003

[32] T W Liao Z M Zhang and C R Mount ldquoCase-based rea-soning system for identifying failure mechanismsrdquo EngineeringApplications of Artificial Intelligence vol 13 no 2 pp 199ndash2132000

[33] B S Yang T Han and Y Kim ldquoIntegration of ART-Kohonenneural network and case-based reasoning for intelligent faultdiagnosisrdquo Expert Systems with Applications vol 26 no 3 pp387ndash395 2004

[34] H C W Lau C W Y Wong I K Hui and K F PunldquoDesign and implementation of an integrated knowledge sys-temrdquo Knowledge-Based Systems vol 16 no 2 pp 69ndash76 2003

[35] S L Wang and S H Hsu ldquoA web-based CBR knowledgemanagement system for PC troubleshootingrdquo InternationalJournal of Advanced Manufacturing Technology vol 23 no 7-8 pp 532ndash540 2004

[36] C Hsu C Chiu and P Hsu ldquoPredicting information systemsoutsourcing success using a hierarchical design of case-basedreasoningrdquo Expert Systems with Applications vol 26 no 3 pp435ndash441 2004

[37] S W Changchien and M Lin ldquoDesign and implementationof a case-based reasoning system for marketing plansrdquo ExpertSystems with Applications vol 28 no 1 pp 43ndash53 2005

[38] C Chiu ldquoA case-based customer classification approach fordirect marketingrdquo Expert Systems with Applications vol 22 no2 pp 163ndash168 2002

[39] R Belecheanu K S Pawar R J Barson B Bredehorst andF Weber ldquoThe application of case based reasoning to decisionsupport in new product developmentrdquo Integrated Manufactur-ing Systems vol 14 no 1 pp 36ndash45 2003

[40] P Kotler Marketing Management Analysis Planning Imple-mentation and Control Prentice Hall International EditionsNew York NY USA 1997

[41] R Ramanathan ldquoStochastic decision making using multiplica-tive AHPrdquo European Journal of Operational Research vol 97 no3 pp 543ndash549 1997

[42] R C van den Honert and F A Lootsma ldquoGroup preferenceaggregation in the multiplicative AHP the model of the groupdecision process and pareto optimalityrdquo European Journal ofOperational Research vol 96 no 2 pp 363ndash370 1997

[43] D L Olson G Fliedner and K Currie ldquoComparison ofthe REMBRANDT system with analytic hierarchy processrdquoEuropean Journal of Operational Research vol 82 no 3 pp 522ndash539 1995

[44] W Chen and J LiaoDecision Support System and DevelopmentTsinghua University Press Beijing China 3rd edition 2008

[45] Q Shao W Weng X Zhen et al ldquoResearch on assistantdecision-making method with urban fire case baserdquo ChinaSafety Science Journal vol 19 no 1 pp 113ndash117 2009


[46] P Buta ldquoMining for financial knowledge with CBRrdquo AI Expertvol 9 no 2 pp 34ndash41 1994

[47] D Gu X Li C Liang et al ldquoResearch on case retrievalwith weight optimizing and its applicationrdquo Journal of SystemsEngineering vol 24 no 6 pp 764ndash768 2009

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



of customers is not yet distinguished (3) Different levels ofcustomers have the same transition probability which is staticbut dynastic To overcome the above three deficiencies inChanrsquos model this study proposes a CBR-based and MAHP-based customer value prediction model for new productdevelopment (CampM-CVPM) and its advantages are as fol-lows (1) Case based reasoning (CBR) can reduce expertsrsquoevaluationworkload andwork time which couldmake up fordeficiency one (2) Multiplicative analytic hierarchy process(MAHP) uses the actual effectiveness of the influencing factorrather than the overall average in stimulation which couldmake up for deficiency two (3) The dynamic customersrsquotransition probability is more close to reality which couldmake up for deficiency three

This section has given the general background to thestudy Section 2 discusses the literature on new productdevelopment The methodology for the proposed CampM-CVPM is presented in Section 3 Section 4 introduces themodelrsquos framework and its three modules A stimulationexperiment is conducted in Section 5 Some concludingremarks are offered in Section 6

2 Literature Review

21 The Present State of the Research In recent years manyconventional and mature models for new product devel-opment have been proposed Xu et al use fuzzy set andutility and theory to evaluate the typical phases duringnew productrsquos whole lifecycle [3] Besharati et al mea-sure new product development according to the size ofcustomer expected utility that is market demand basedon customersrsquo preference designersrsquo experience preferenceand uncertainty of product designing characteristics [4]Alexouda applies algorithm and enumeration method innew product development and then combines the optimalproduct from the perspective of customer utility maximiza-tion and at last returns market sharersquos maximization [5]Matsatsinis and Siskos discuss market share maximizationby historical data (economic or customer survey data) [6]Hu and Bidanda utilize Markov to build sustainable greenproduct lifecyclersquos evolution system and then make decisionsfor new product development from the maximization of thepresent discount value during the whole product lifecycle[7]

The following three areas which cover various influenc-ing factors have been identified as significant and requisite innew product development (a) product attributes specified bydesigners [3 8ndash10] (b) customer requirements and satisfac-tion [11ndash14] and (c) marketing competence [15ndash18]

However none of the currently available models consid-ers all of these areas concurrently Only Chan and Ip [19]integrates three recognized factors and takes the time valueof money into consideration so a new comprehensive modelis proposed on new product development Within Chanrsquosmodel the involved influencing factors are classified intofive parts overall product attractiveness (OPA) overall cus-tomer satisfaction (OCS) word ofmouth (WOM)marketingapproach (MA) and remarketing approach (RA) Based onpurchasing frequency new product customers are divided

into five levels potential customers first-time customersregular customers frequent customers and loyal customers

While the system is running the overall average effec-tiveness of the above five influencing factor is evaluated byexperts or historical data Then the customer purchasingmodel is established by system dynamics and the customerlifetime value (CLV) can be predicted byMarkovmodel Andat last this model could evaluate new product development

22 Analysis of the Research Status Relative to previousresearch Chanrsquosmodel has incomparable superioritywhetheron evaluation perspective and method or the influencingfactor But in order to better meet user needs after furtheranalyzing with the evaluationmethods used in Chanrsquos modelit can easily find out three specific deficiencies which need tobe solved urgently

Deficiency (1) The expertsrsquo workload must be reduced byauxiliary software To be specific whenChanrsquosmodel predictscustomer lifetime value of newproduct the parametersrsquo effec-tiveness such as production cost is basically obtained with theconcerted efforts of experts However there is no availabledata for the product that does not yet enter the marketTherefore it is hardly feasible to investigate and evaluate thoseparameters one by one This is because firstly it will imposelarge workloads to experts and last a long period of time toevaluate each parameter Secondly system developers maynot have necessarily adequate ability to accurately evaluateeach parameter In addition the subjectivity from expertscould easily lead to certain deviations

Deficiency (2) For improving the accuracy of evaluation it isnecessary to distinguish the effectiveness of the influencingfactors for different levels of customers Chan thinks that dif-ferent levels of customers have the same effectiveness of oneinfluencing factor that equals to overall average effectivenessof all levels of customers However in real life the high-frequency purchasing customers often have a relatively highrecognition to the product during their purchasing Whenthe productrsquos overall average effectiveness has reached acertain value itmay not attract the low-frequency purchasingcustomers but it can attract the high-frequency ones

Deficiency (3) Dynamic customersrsquo transition probability ismuchmore realisticWhen stimulating the purchasing Chanuses original nonzero probability as all levels of customersrsquotransition probability Namely the level of (119894) customersrsquotransition probability equals the retained customersrsquo numberat time (119894 minus 1) divided the customersrsquo number at time (119894 minus 2)and it is obvious that the probability is irrelevant to time SoChan believes that different levels of customers have the sametransition probability that is static but dynastic However thereal system is filled with dynamics complexity and productperiodicity As time passed each level of customersrsquo transitionprobability will change within a certain range

This study takes the advantages of CBR and MAHP andintroduces the dynamic transition probability Based on whatis discussed above CampM-CVPM is proposed In responseto the three deficiencies from Chanrsquos model this study puts






3 Study Approach





New case

Case base

Retrieve

Reus

e

Revise

Retain

Similar case

Confirmed solution

Suggestedsolution

















4 CampM-CVPM










Developer

Expe

rt CBR

MAHP

NCLV prediction

New

pro

duct

desig

n

Dec

ision

-set

attr

ibut

e

E1E2

E4E5

D(Eic)

New

pro

duct

NCL

V an

d cu

stom

ernu

mbe

r-te

nden

cy g

raph

CampM-CVPM

Client i

Client j

Model server

Network

(MS)


(ODS)

Databaseserver(DBS)

Knowledgeserver(KS)

Datawarehouse

server (DWS)

CSMCRCPC









2(overall


5












Unit 1


Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s







1198941 1198831198942 119883

119894119899


1 1198662 119866





119863(119883119894119895)

= [119883

119866]

=


[[[[[[

[

11988311

11988312


1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661


119899

]]]]]]

]


(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)


(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)



sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








3 Study Approach





New case

Case base

Retrieve

Reus

e

Revise

Retain

Similar case

Confirmed solution

Suggestedsolution

















4 CampM-CVPM










Developer

Expe

rt CBR

MAHP

NCLV prediction

New

pro

duct

desig

n

Dec

ision

-set

attr

ibut

e

E1E2

E4E5

D(Eic)

New

pro

duct

NCL

V an

d cu

stom

ernu

mbe

r-te

nden

cy g

raph

CampM-CVPM

Client i

Client j

Model server

Network

(MS)


(ODS)

Databaseserver(DBS)

Knowledgeserver(KS)

Datawarehouse

server (DWS)

CSMCRCPC









2(overall


5












Unit 1


Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s







1198941 1198831198942 119883

119894119899


1 1198662 119866





119863(119883119894119895)

= [119883

119866]

=


[[[[[[

[

11988311

11988312


1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661


119899

]]]]]]

]


(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)


(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)



sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in












4 CampM-CVPM










Developer

Expe

rt CBR

MAHP

NCLV prediction

New

pro

duct

desig

n

Dec

ision

-set

attr

ibut

e

E1E2

E4E5

D(Eic)

New

pro

duct

NCL

V an

d cu

stom

ernu

mbe

r-te

nden

cy g

raph

CampM-CVPM

Client i

Client j

Model server

Network

(MS)


(ODS)

Databaseserver(DBS)

Knowledgeserver(KS)

Datawarehouse

server (DWS)

CSMCRCPC









2(overall


5












Unit 1


Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s







1198941 1198831198942 119883

119894119899


1 1198662 119866





119863(119883119894119895)

= [119883

119866]

=


[[[[[[

[

11988311

11988312


1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661


119899

]]]]]]

]


(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)


(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)



sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Developer

Expe

rt CBR

MAHP

NCLV prediction

New

pro

duct

desig

n

Dec

ision

-set

attr

ibut

e

E1E2

E4E5

D(Eic)

New

pro

duct

NCL

V an

d cu

stom

ernu

mbe

r-te

nden

cy g

raph

CampM-CVPM

Client i

Client j

Model server

Network

(MS)


(ODS)

Databaseserver(DBS)

Knowledgeserver(KS)

Datawarehouse

server (DWS)

CSMCRCPC









2(overall


5












Unit 1


Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s







1198941 1198831198942 119883

119894119899


1 1198662 119866





119863(119883119894119895)

= [119883

119866]

=


[[[[[[

[

11988311

11988312


1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661


119899

]]]]]]

]


(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)


(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)



sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in











Unit 1


Marketing approach

Unit n

Unit 1

Unit n

A ca

se o

f pro

duct

s

Prod

ucts

struc

ture

Part 2

Part 1

Influ

enci

ngfa

ctor

s







1198941 1198831198942 119883

119894119899


1 1198662 119866





119863(119883119894119895)

= [119883

119866]

=


[[[[[[

[

11988311

11988312


1198831198971

1198831198972sdot sdot sdot 119883

119897119899

d

1198831198981

1198831198982

sdot sdot sdot 119883119898119899

1198661


119899

]]]]]]

]


(1)

119862119895=

sum119898+1

119894=1119883119894119895

119898 + 1 (2)

119872119894119895=

119883119894119895minus119862119895

119862119895

(3)


(4) and1198831015840119894119895is stated in (5)

119863(1198831015840

119894119895) = [

1198831015840

119894

1198661015840] (4)

1198831015840

119894119895=

1 minus exp (minus119872119894119895)

1 + exp (minus119872119894119895)

1198831015840

119894119895isin [minus1 1] (5)



sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





sim (119883119894 119866) = sim (1198831015840

119894 1198661015840) = 1 minus dis (1198831015840

119894 1198661015840) = 1 minus radic

119899

sum

119895=1

119908119895sdot 119889 (119883

1015840

119894119895 1198661015840

119895)

119889 (1198831015840

119894119895 1198661015840

119895) = 1 minus sim (1198831015840

119894119895 1198661015840

119895)

sim (1198831015840119894119895 1198661015840

119895) =

1 119883

1015840

119894119895= 1198661015840

119895

0 1198831015840

119894119895= 1198661015840

119895

when 1198831015840119894119895and 1198661015840


exp(minus100381610038161003816100381610038161198831015840

119894119895minus 1198661015840

119895

10038161003816100381610038161003816


119894119895and 1198661015840


119886 (1198831015840

119894119895cap 1198661015840

119895)

119886 (1198831015840

119894119895) + 119886 (119866

1015840

119895) minus 119886 (119883

1015840

119894119895cap 1198661015840

119895)

when 1198831015840119894119895and 1198661015840


(6)


119894119895cap 1198661015840










3









120575(119901)








119894119895and the weights 119908(119901)

119894119888

and 119908(119901)119895119888


119863(119901)

119888= (120575(119901)

119894119895)119888 (7)

119908(119901)

119894119888

119908(119901)

119895119888

= 119890minus lnradic2120575(119901)

119894119895 (8)

5

prod

119894=1

119908(119901)

119894119888= 1 (9)




Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





Design

Performance

Packaging

Competitiveness

Quality



Word of mouth

Design importance








competitiveness


product quality


Loss rate








Potential customers

Acquisition rate



119908(119901)

119894119888= exp( lnradic2

5

5

sum

119895=1

120575(119901)

119894119895) (10)

119908lowast

119894119888= exp( lnradic2

5119896

119896

sum

119901=1

5

sum

119894=1

120575(119901)

119894119895) (11)

119908119894119888=

119908lowast

119894119888

sum5

119895=1119908lowast

119895119888

119894 = 1 2 5 (12)







119894119888= 0 it



119894119888is easy to get by (14)

Δ119864119894119888= 119908119894119888sdot 119864119888minus119864119888

5 (13)

119864119894119888= 119864119888+ Δ119864119894119888 (14)


119894119888)







C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





C1t (potential)

C2t (first-time)

C3t (regular)

C4t (frequent)

C5t (loyal)

E1t (OPA)

E2t (ME)

E3t (WOM)

E4t (RE)

E5t (OCS)






Loyal customersC5t

A4t

L3t

L2t

L4t

L5t

A3t

A2t

A1t

Active customers





(119905) AR119904119905






119904119905 AR119904119905 LR119904119905 and 119871

119904119905


119862119904119905times AR119904119905= 119860119904119905

119862119904119905times LR119904119905= 119871119904119905

(15)





3 LR and 119864

2





AR1119905= random min (119864

11 11986413 11986414)

max (11986411 11986413 11986414)

AR2119905= random min (119864

21 11986422 11986423 11986425)

max (11986421 11986422 11986423 11986425)

AR3119905= random min (119864

31 11986432 11986433 11986435)

max (11986431 11986432 11986433 11986435)

AR4119905= random min (119864

41 11986442 11986443 11986445)

max (11986441 11986442 11986443 11986445)

(16)





1198621119905=

1198621119905minus119889119905

+ (1198712119905minus119889119905

+ 1198713119905minus119889119905

+ 1198714119905minus119889119905

+1198715119905minus119889119905

minus 1198601119905minus119889119905

) 119905 = 0

PC 119905 = 0

1198622119905=

1198622119905minus119889119905

+ (1198601119905minus119889119905

minus 1198712119905minus119889119905

minus 1198602119905minus119889119905

) 119905 = 0

0 119905 = 0


Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Potential customers

Active customers

Reacquisition rate



Acquisition rate



Word of mouth

Loss rate

First-timecustomers

C1 t

C2 t

AR1 t

AR2minus4 t

LR2minus5 t

E2

E3

E1

E5

E4

C3 t + C4 t + C5 t


1198623119905=

1198623119905minus119889119905

+ (1198602119905minus119889119905

minus 1198713119905minus119889119905

minus 1198603119905minus119889119905

) 119905 = 0

0 119905 = 0

1198624119905=

1198624119905minus119889119905

+ (1198603119905minus119889119905

minus 1198714119905minus119889119905

minus 1198604119905minus119889119905

) 119905 = 0

0 119905 = 0

1198625119905=

1198625119905minus119889119905

+ (1198604119905minus119889119905

minus 1198715119905minus119889119905

) 119905 = 0

0 119905 = 0

(17)


NCLV = (infin

sum

119905=0




minus1

(18)







































119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




























119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










119894119888) can be got











one time














Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Package approach











Marketing approach






Toot

hbru

shha

irTo

othb

rush

hand

Toot

hbru

shhe

adIn

fluen

cing

fact

ors




(Red white blue )

(00 170sim200)

(ABS PP PC )




A X

X re

gula

r too

thbr

ush

prod

uct c

ase














()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








()


deviation rate ()1 04 2 8 20 42322 04 6 4 60 27823 04 10 0 100 26524 085 6 4 60 18155 085 8 2 80 15696 085 10 0 100 121

Times of tests



The d

evia

tion

rate

of N

CLV(

)

1

09

08

07

06

05

04

03

021 15 2 25 3 35 4 45 5 55 6






Times of tests1

140

120

100

80

60

40

20

015 2 25 3 35 4 45 5 55 6

The d

evia

tion

rate

of N

CLV(

)





0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber








times105

2

18

16

14

12

1

08

06

04

02

0












0 5 10 15 20 25

NCL

Vc

usto

mer

num

ber

times104

14

12

10

8

6

4

2

0



6 Conclusions



8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




8

7

6

5

4

3

2

1

0

times104

C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

0 5 10 15


C1

C2

C1 + C2 + C3

Time

Custo

mer

num

ber

8

7

6

5

4

3

2

1

00 5 10 15

times104








Acknowledgments



References

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



















































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in













Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



Research Article A CBR-Based and MAHP-Based Customer Value ...downloads.hindawi.com/journals/tswj/2014/459765.pdf · Chan s model, this study proposes a CBR-based and MAHP-based customer

Documents