文档库

最新最全的文档下载
当前位置:文档库 > Abstract Comparison of product bundling strategies on different online shopping behaviors

Abstract Comparison of product bundling strategies on different online shopping behaviors

Comparison of product bundling strategies on di?erent online

shopping behaviors

Tzyy-Ching Yang a ,Hsiangchu Lai

b,*

a

Department of Computer Science and Information Management,Providence University,200Chung-Chi Road,Taichung 433,Taiwan

b

Department of Information Management,National Sun Yat-sen University,70Lien-Hai Road,Kaohsiung 804,Taiwan

Received 2May 2005;received in revised form 3November 2005;accepted 27April 2006

Available online 12June 2006

Abstract

Bundling is a very popular sales-promotion tool,in which a critical issue is to decide what products should be sold together in order to improve sales.Traditionally,this decision is based on the order data collected from the points of sale.However,Internet marketing now allows marketers to e?ciently collect not only order data but also browsing and shopping-cart data,which provide marketers with infor-mation on the consumers’decision-making processes,rather than only the ?nal shopping decisions.The present study aimed to deter-mine the value of this newly available information by comparing the performance of decision-making on product bundling based on three types of data on online shopping behaviors.The results from a ?eld experiment reveal that signi?cantly better decisions are made on the bundling of products when browsing and shopping-cart data are integrated than when only order data or browsing data are used.ó2006Elsevier B.V.All rights reserved.

Keywords:Online behavior;Product bundling;Shopping cart;Market basket analysis;Association rules

1.Introduction

A better understanding of customers allows better mar-keting strategies to be designed.The Internet provides mar-keters with much more data on customers,and consequently has brought marketing management into a new age [41].In addition to collecting order data through the point of sale (POS),as in the pre-Internet age,on the Internet the entire shopping procedure of any customer can be recorded,including not only what has been ordered,but also what has been clicked or browsed and what has been moved in or moved out from the shopping cart,and the timings of all of these processes.

Various studies have demonstrated the usefulness of data on shopping processes and outcomes [21,22].The essential di?erence between order data and other data on online shopping behaviors is that the former indicates the

?nal shopping decisions while the latter provides informa-tion on customer behavior whilst making shopping deci-sions.Moreover,a major di?erence between browsing and shopping-cart data is that the information embedded in the latter is more closely related to the ?nal shopping decision [5].Data on customer browsing behavior have been explored in many studies,such as to ?nd path travel patterns [6,12],website tra?c,the products being browsed the most,the hot area of a web page,and the customer pro-?les based on these browsing data [30,31].However,very few studies have investigated data related to customer shopping carts [15,32].

The usefulness of data on customer online shopping behaviors to learning about customers has resulted in data mining becoming an important and popular technique for exploring customer pro?les,shopping behaviors,and other aspects of online shopping,because the associated data sets are often huge.For example,Chen et al.[11]integrated customer behavioral variables,demographic variables,and a transaction database to establish a method for min-ing changes in customer behavior.Jiao and Zhang [25]

1567-4223/$-see front matter ó2006Elsevier B.V.All rights reserved.doi:10.1016/j.elerap.2006.04.006

*

Corresponding author.Tel.:+88675254776;fax:+88675254799.E-mail addresses:tcyang@http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.html.tw (T.-C.Yang),hclai@http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.html.tw (http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli).

Abstract Comparison of product bundling strategies on different online shopping behaviors

http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.html/locate/ecra

Electronic Commerce Research and Applications 5(2006)

Abstract Comparison of product bundling strategies on different online shopping behaviors

Abstract Comparison of product bundling strategies on different online shopping behaviors

295–304

attempted to develop an explicit decision support to improve product portfolio identi?cation by applying asso-ciation-rule mining.Further,Rygielski et al.[44]pointed out that organizations must be aware of the trade-o?s asso-ciated with choosing suitable data-mining software for use in customer relationship management(CRM).

Several sales promotion tools are used in the market-place,such as free samples,coupons,premiums,bundle pricing(bundling),and cross-promotions[45].Bundling is a very common practice that involves combining two or more products or services and selling them at a set price [20,47].Examples of bundles are opera season tickets (e.g.,tickets to various events sold together)and Internet services(e.g.,a combination of Web access,e-mail,person-alized content,and an Internet search program)[46].The bundled price is almost always lower than the cost of buy-ing all the products separately.Estelami[18]found that consumers save an average of about8%when purchasing a bundle of items relative to buying the items individually. The challenge is how to choose the appropriate products to be bundled in order to achieve the expected promotion per-formance,such as by creating new markets or increasing customer loyalty,sales,or pro?ts[37].

Previous research has investigated product bundling from di?erent viewpoints.Determining the products to be bundled so as to maximize performance is the most critical issue in formulating bundling strategies.Searching for associations in the market basket has been previously used to determining appropriate bundles.Further,Munger and Grewal[36]examined the e?ects of the bundling format and framing of promotional discounts on perceived qual-ity,price acceptability,perceived value and subsequent purchase intentions.Chakravarti et al.[10]examined the e?ects of the presentation of a multicomponent product bundle on customer evaluation and choices,as well as the underlying processing e?ects.Stremersch and Tellis[46] showed that a company that exploits opportunities o?ered by bundling will enjoy increased market share and pro?ts. However,none of the previous research studies have con-sidered product bundling based on data on online shopping behaviors,which represent newly available data from which customer pro?les can be elucidated.

The purpose of this study was to assess the value of data on online shopping behaviors in making decisions on prod-uct bundling by examining the performance of product bundling strategies based on di?erent data sets related to online shopping behaviors.The remainder of this paper is organized as follows:Section2surveys the related litera-ture on product bundling,market basket analysis,market-ing implications of data on online shopping behaviors,and association rules;Section3proposes three product bun-dling strategies based on di?erent types of data on online shopping behaviors;Section4compares the performance of these three product bundling strategies based on data collected from a?eld experiment;and the paper ends with the conclusions drawn in Section5,which also discusses some of the limitations of the study.2.Literature review

2.1.Product bundling

Product bundling is a pervasive selling strategy in mar-kets,examples of which include sporting and cultural orga-nizations o?ering season tickets,restaurants providing complete dinners,and retail stores o?ering discounts to a customer buying more than one product.How to bundle products in order to maximize pro?ts under di?erent mar-ket environments has been a popular research issue in the economics?eld[1,14,16,17,34,38].However,in this study we focused on the marketing aspects of product bundling. Stremersch and Tellis[46]identi?ed two key dimensions in classifying bundling strategies:focus and form.The focus of bundling can be either the price or the product,while the form of bundling can be none,pure,or mixed.In the paper,they also provide managers with a framework with which to understand and choose bundling strategies.

Janiszewski and Cunha[24]indicated that the impact of a price discount on the perceived attractiveness of a bundle depends on the type of product that is being discounted. They also concluded that reference dependence and prod-uct importance independently contribute to the e?ects of price discounts.Agarwal and Chatterjee[2]examined the decision di?culties experienced by customers when select-ing from a menu of bundles.They found that larger bun-dles make decisions more di?cult;more specialized services in the competing bundles increases the decision dif-?culty for small,but not large,bundles;and that the choice di?culty is greater for bundles that are more similar.

2.2.Market basket analysis

Market basket analysis refers to investigating the com-position of the basket of products purchased by a house-hold during a single shopping experience[43].Retailers have long been interested in learning about the cross-cate-gory purchase behavior of their customers[35],since such information makes it easier for the retailer to decide whether to group products by brand or by product type, for example[9].The market basket choice refers to the decision process in which a consumer selects items from several product categories during the same shopping expe-rience[43].

Therefore,basket analysis can provide the distribution of shoppers’purchases based on di?erent viewpoints,such as the product itself,the product category,the shopper’s background,or the average purchases per shopper[26]. Such distribution information will aid decisions on aspects such as planning and designing advertising,sales promo-tions,store layout,and product placement[8].In addition, basket data contain important information about the struc-ture of brand preferences both within and across product categories[42],and hence a richer picture of customer behavior and better decisions on the bundling of products may result from identifying the associations between

296T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304

product purchases at the POS[40].The other way to?nd such associations is to apply association-rule algorithms to both browsing and shopping-cart data.

2.3.Marketing implications of data on online shopping behaviors

Generally,basket analysis is applied to data on the com-position of the basket of purchased products–that is,on the content of the?nal order.Advances in information technology have radically changed the methods used to col-lect consumer data[19].For example,computerized check-outs generate almost immediate feedback about the pro?tability of brands,product groups,and the e?ects of marketing activities such as in-store promotions and weekly advertising[26].In the Internet age,it is possible to collect not only order data at the checkout but also data on the consumer’s entire shopping behavior at a website.

Kotler[29]proposed a model of the successive sets involved in the consumer decision-making process,which revealed that consumers’decision-making processes involve?ve successive sets:total set,awareness set,consid-eration set,choice set,and?nal decision.As shown in Fig.1,the total set includes all brands available to a con-sumer.However,the customer is likely to be aware of only a subset of these brands(i.e.,the awareness set),with only some of these brands(i.e.,the consideration set)ful?lling the consumer’s initial buying criteria.The collection of more information by the customer will result in only a few brands remaining acceptable,which together form the choice set.Finally,the customer chooses one item from the choice set.

Through the information search process,the products foremost in the consumer’s mind will change from the total set to the awareness set,consideration set,and choice set, with the?nal decision then being made.Whether a product proceeds through each stage and reaches the?nal choice depends on the information available to the customer and his or her thought processes at each stage of the decision process.If marketers are provided with informa-tion about the thought processes that are important at the awareness,consideration,and choice sets to the cus-tomers’decision-making processes,they can understand how the?nal choice results from the successive sets,which should result in better marketing decisions being made.

Collecting data on online shopping behaviors is one way to understand what a customer is interested in at each step and the possible thoughts underlying this process.Tradi-tionally,the only data collected through the POS is the order data,which re?ects the?nal shopping decision rather than the previous sets as outlined in Fig.1.Although knowing what consumers have bought is useful to market-ers,this does not allow them to learn why consumers have made a particular purchase decision and not bought some other item.

In the Internet age,it is relatively easy to collect data related to the decision process such as browsing and shop-ping-cart data.The browsing data may provide informa-tion on the thought processes that lead from the awareness set to the consideration set and the choice set, and the shopping-cart data may provide more information on the thought processes from the consideration set to the choice set,and then to making the?nal decision.It should be noted that shopping-cart data may di?er from the order data,since items in the shopping cart can be changed at any time before the purchase is?nalized.

Information on the stages closer to the?nal shopping decision are especially valuable to building a more accurate customer pro?le.Whilst shopping-cart data is considered to provide more reliable information about why the con-sumers buy or do not buy certain products,even the brows-ing data provides useful information on the customers’interest that cannot be found in the shopping-cart or order data.For example,the changes that occur from the aware-ness set to the consideration set,and then to the choice set might be due to the customer’s preferences,budget,avail-able shopping time,and exposure to marketing promotion programs,for example.In addition,the patterns of the

Abstract Comparison of product bundling strategies on different online shopping behaviors

T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304297

di?erent successive sets can di?er between customers(e.g., a product in a customer’s choice set might only be in another customer’s consideration set).

2.4.Association rules

The problem of searching for association rules in large databases using data mining has been widely studied [3,4,33].A very popular application area is to?nd associa-tions by analyzing the market basket of products pur-chased in a single shopping experience[13,43].The goal is to discover buying patterns,such as two or more items that are often bought together.The past research has eluci-dated techniques for improving the performance of algo-rithms for discovering association rules in large databases of sales information.

Agrawal et al.[4]?rst introduced the problem of?nding association rules.An association rule can be expressed by the form X)Y when c%of transactions that contain X also contain Y.We call c the con?dence of the association rule,where c%=Prob(Y j X).The rule X)Y has support s, where s%=Prob(X\Y).An example outcome of associa-tion-rule mining is determining that‘‘90%of customers who buy A and B will also buy C’’in a transaction data-base;in this case the con?dence of the rule is90%.Another parameter is the support of an item set,such as{A,B,C}, which is de?ned as the percentage of times that the item set is contained in all of the transactions[27].Therefore, the problem can be de?ned by generating all association rules that have support greater than the user-speci?ed min-imum support[3].

Apriori is the most popular algorithm(Table1)used to ?nd association rules.This algorithm constructs a candi-date set of large(kà1)-item sets,counts the number of occurrences of each candidate item set,and then deter-mines large k-item sets based on the minimum support in each iteration[12].Some algorithms representing revisions of the Apriori have been proposed,such as AprioriTid, AprioriHybrid[3],OCD[33],DHP[39],and SETM[23].

3.Di?erent product bundling strategies

Based on Fig.1,we can see there are three di?erent types of data on online shopping behaviors:browsing data,shop-ping-cart data,and order data.Fig.1also displays the rela-tionship between the online shopping behaviors and the successive sets proposed by Kotler[29].Traditional basket analysis is based on ordered products only;that is,the?nal shopping decision.Because the preferences underlying the decision process may be hidden,the other types of data on online shopping behaviors such as browsing and shop-ping-cart data should be considered when determining the customer pro?le and thereafter the optimal marketing strategy.

Therefore,in addition to searching for associations in the order data,we propose another two bundling strategies based on the other types of data on online shopping behav-iors that are available:(1)that based on the browsing data only;and(2)that based on both browsing and shopping-cart data.The general method?rst requires the collection period of the data set and the required minimum support to be set.The data belonging to the same customer are then merged,and the Apriori algorithm is applied to?nd asso-ciations that ful?ll the minimum support.The customer’s shopping is merged(rather than using a single shopping record)in order to obtain a more robust result,since a sin-gle shopping record may not be representative of the cus-tomer http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmlparing the performance of these di?erent strategies will reveal if the new proposed strategies are better than the traditional one(based on order data only).In the following sections,we describe the strategies and their advantages and disadvantages.

3.1.Strategy based on order data only

Before the rise of the Internet it was only possible for order data to be collected e?ciently,and hence it was very common for traditional marketers to make bundling deci-sions based on these data only.In order to improve the robustness of product bundling,we suggest searching for associations based on all orders of a customer.First,all of the order data belonging to the same customer are merged,and then the algorithm is applied to the reorga-nized data set to determine the product associations.The process is summarized in Table2.

The weakness of the strategy in Table2is that the result will not be robust if insu?cient order data are available. Further,the order data directly imply only the purchase decision rather than the underlying decision process,and therefore cannot be applied to one-time shopping such as that of large electric appliances.In addition,a customer’s

Table1

Algorithm Apriori and the Apriori–gen function[3]

Algorithm Apriori

(1)L1={large1–itemsets};

(2)for(k=2;L kà1?u;k++)do begin

(3)C k=apriori-gen(L kà1);//New candidates

(4)for all transactions t2D do begin

(5)C t=subset(C k,t);//Candidates contained in t

(6)forall candidates c2C t do

(7) c.count++;

(8)end

(9)L k={c2C k j c.count P minsup}

(10)end

(11)Answer=¨k L k;

Apriori–gen function

(12)insert into C k

(13)select p.item1,p.item2,...,p.item kà1,q.item kà1

from L kà1,p,L kà1,q

where p.item1=q.item1,...,p.item kà2=q.item kà2,p.item kà1

(14)forall itemsets c2C k do

(15)forall(kà1)subsets s of c do

(16)if(s2L kà1)then delete c from C k;

Notation:(1)L k:set of large k-item sets.(2)C k:set of candidate k-item

sets.(3)minsup:minimum support.

298T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304

?nal order may also depend on factors such as his or her budget,preferences,available shopping time,and the other available substitute products.For example,a customer may browse a product on the Internet but not buy it, instead buying the product later at a physical store;in this

case the Internet order data do not re?ect the customer’s interest.However,the advantage of this bundling strategy is that it is easier to collect more comprehensive informa-tion about the customer’s shopping behavior.

3.2.Strategy based on browsing data only

The second proposed product bundling strategy is based on the browsing data so as to overcome the possibility of the order data not re?ecting the hidden preferences of cus-tomers.As discussed in Section2,browsing data may include more information about the shopping process,in terms of the customer’s consideration set or choice set (but not what is actually bought by the customer).Hence, customers are interested in these products but may still choose not to buy them due to reasons such as high price or low product quality,or a low customer demand.Bun-dling might allow these products to reach the customer’s ?nal stage of the shopping process.Table3lists the process of determining product bundling based on browsing data.

The weakness of the strategy in Table3is that the browsing behavior might be in?uenced by the design of the website,such as what is contained within the hot area and how the hyperlinks are arranged.That is,a product may be browsed the most due to its place in the website lay-out rather than the customer’s preferences.One way to reduce this interference is to delete data from the data set whose browsing time is less than a minimum threshold, as in Table3.One advantage of this strategy is that the col-lected data set will be much larger,and hence will be more applicable to data-mining techniques.3.3.Strategy based on both browsing and shopping-cart data

The above problems of the order data only re?ecting the ?nal purchase preferences of a few consumers and the brows-ing data being easily in?uenced by the design of the websites are overcome by using a strategy based on both browsing and shopping-cart data(Table4).This strategy involves ?nding the large item sets based on the browsing data?rst, and then examining whether the support of each of these large item sets exceeds the minimum support requirement based on the shopping cart.

The advantage of this integrated method is that it can re?ect the behavior of more consumers and the hidden preferences of the customers by using browsing data instead of order data to?nd the large item sets.Moreover, it also can avoid the in?uence of the website layout of product catalogs by considering the shopping-cart data as well.In addition,the associations resulting from the strat-egy are supported both by the browsing behavior and by the items placed in the shopping cart,and therefore they more accurately re?ect the customers’potential shopping preferences.

4.Experimental design

4.1.Data collection

For the purpose of examining and comparing the per-formance of the three product bundling strategies,in the present study the data on the online shopping behaviors of customers were collected from the website of a publisher specializing in information technology and electronic com-merce books.The site Web server was internet information server(IIS)running on Windows NT,with SQL Server as its database.ASP was adopted to develop the online mem-bership functions of the publisher’s website and dynamic shopping homepages in order to collect customers’back-grounds,shopping-cart data,and order data.In addition, the log?les were implemented in the W3C Extension for-mat,and several cookie items were created to record all the browsing behavior of every member.The customers’browsing data could be extracted from the log?les.

Table2

Product bundling strategy based on order data

Step1:Let marketers set the period of analysis(P)and the minimum support(S)

Step2:Select data collected during P

Step3:Merge the order data belonging to the same customer

Step4:Apply an association-rule algorithm to determine the large item sets whose support is larger than S

Table3

Product bundling strategy based on browsing data

Step1:Let marketers set the period of analysis(P),minimum support(S), and minimum browsing time(T)

Step2:Select browsing data collected during P.If the browsing time of the data is less than T,the data are removed

Step3:Merge the browsing data belonging to the same customer

Step4:Apply an association-rule algorithm to determine the large item sets whose support is larger than S Table4

Product bundling strategy based on both browsing and shopping-cart data Step1:Let marketers set the period of analysis(P),minimum support(S), and minimum browsing time(T)

Step2:Select browsing data collected during P.If the browsing time of the data is not greater than T or the page is not linked to the shopping cart, the data are removed

Step3:Merge the browsing data belonging to the same customer

Step4:Apply an association-rule algorithm to determine the large item sets whose support is larger than S

Step5:Select shopping-cart data collected during P,and then merge the data belonging to the same customer

Step6:Calculate the support based on the shopping-cart data for each of the large item sets found in Step4,and choose the large item sets whose support is larger than S

T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304299

Before the?eld experiment,the online shopping behav-iors of customers were recorded by the aforementioned technologies for six months,during which the publisher published136books in14categories.A total of1500cus-tomers joined up as members,among whom77.5%were male,68.4%were educated at least to university level, and53.7%were students(the remaining customers were all employed).

There were24,316browsing sessions during the period of online data collection.Removing data without login information resulted in2836valid sessions belonging to 1472members.There were197orders,involving459books bought by168members.The shopping-cart data com-

prised719valid records belonging to447members.

4.2.Operation of product bundling strategies

The following sections discuss the product bundling decisions that resulted from the three strategies based on the collected data on online shopping behaviors.

4.2.1.Strategy based on order data only

Merging the order data belonging to the same cus-tomer revealed that95customers purchased more than one book.We set the minimum support as5(%),indicating that the chance of any pair of products being purchased by the same customer was greater than5%.We then applied the Apriori algorithm in Table1to?nd product associations;the top-ten product associations with higher support are listed in Table5,which indicates,for example, that17.9%of the customers purchased both book L001 and book L06.

4.2.2.Strategy based on browsing data only

Any records associated with a browsing time of less than 5s were?rst removed,which left1134sessions.Merging the browsing data belonging to the same customer resulted in517valid browsing sessions.Again,the Apriori algo-rithm was applied to?nd the product association with the minimum support set as5(%)(i.e.,the chance of any pair of products being browsed by the same customer was greater than5%).The top-ten product associations are summarized in Table6,which indicates,for example,that18%of the customers browsed both book L001and book S07.

4.2.3.Strategy based on both browsing and shopping-cart data

First,the Apriori algorithm was used to?nd product associations based on browsing data with the minimum support set as5(%).In this case,the data cleaning process deleted not only the records associated with a browsing time of less than5s but also those records relating to books that did not link to shopping cart.After data cleaning,232 valid browsing sessions were used to?nd product associa-tions.We then checked whether or not either the product of each two-product association based on the browsing data appeared in the shopping-cart data.Table7presents the results in the order of the probability of both of the two products in a product association appearing in the shopping cart simultaneously.Taking the?rst pair as example,the chance of both B18and B2being browsed by the same customer is8.2%,while the chance of both being placed in this customer’s shopping cart is6.9%.

4.2.4.Selection of the product bundling strategies

Table8summarizes the product associations extracted by applying the above three strategies.After discussions with the manager of the publishing company,we decided

Table5

Product associations based on order data

Product item1Product item2Support(%) L001L0617.9

B15B1613.7

B18B2110.5

L001T00110.5

L06T0019.5

B15B189.5

B19B219.5

B15B218.4

B09B158.4

B14B157.4Table6

Product associations based on browsing data

Product item1Product item2Support(%) L001S0718.0

L06S0716.8

B21S0711.6

L001L0610.8

S07T0028.9

S07T0017.9

B20S077.9

B18S077.4

N26S077.4

L002S077.4

Table7

Product associations based on both browsing and shopping-cart data Based on browsing data Probability of product associations

appearing in a consumer’s shopping

cart

Product

item1

Product

item2

Support

(%)

None

(%)

One product

(%)

Both

products(%) B18B218.278.414.7 6.9

B19B21 5.686.28.7 5.0

L001L0615.986.710.1 3.2

B20B2111.683.913.3 2.8

B18B207.880.717.0 2.3

B21B22 6.983.913.8 2.3

L001L002 6.989.010.10.9

B21SB01 5.685.313.80.9

B19B20 5.689.010.10.9

L002L3038.294.0 5.50.5

300T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304

to choose two sets of product associations for each strategy to serve as the product bundling strategies in the experi-ment.In order to reduce any interference,the selected product associations extracted from di?erent online data should have the same ranking.In addition,books that had been promoted recently or sold out were removed from the candidate list.Finally,the product associations in bold-face font in Table8were selected as the targets of product bundles in the experiment.The selected product bundles are all ranked at2nd and5th for the three di?erent bun-dling strategies.

We found that one book appeared in both of the two chosen bundles for each of the di?erent strategies.We decided to add a new bundling that included three products comprising the union of both product bundles selected from each strategy.Finally,the nine bundling strategies listed in Table9were selected for inclusion in a?eld exper-iment on the website of the publisher.

4.3.Implementation

After the product bundling had been decided,a?eld experiment was implemented.We collaborated with the publisher to provide three di?erent pricing bundles on the

Table8

Summary of association rules

Priority Based on order data only Based on browsing data only Based on both browsing&shopping-cart data 1(L001,L06)1(L001,S07)(B18,B21)

2(B15,B16)(L06,S07)(B19,B21)

3(B18,B21)or(L001,T001)1(B21,S07)(L001,L06)1

4(B18,B21)or(L001,T001)1(L001,L06)1(B20,B21)

5(L06,T001)1or(B15,B18)or(B19,B21)(S07,T002)2(B18,B20)or(B21,B22)

6(L06,T001)1or(B15,B18)or(B19,B21)(S07,T001)or(B20,S07)(B18,B20)or(B21,B22)

7(L06,T001)1or(B15,B18)or(B19,B21)(S07,T001)or(B20,S07)(L001,L002)or(B21,SB01)or(B19,B20)

8(B15,B21)or(B09,B15)(B20,S07)or(N26,S07)or(L002,S07)(L001,L002)or(B21,SB01)or(B19,B20)

9(B15,B21)or(B09,B15)(B20,S07)or(N26,S07)or(L002,S07)(L001,L002)or(B21,SB01)or(B19,B20)

10(B14,B15)(B20,S07)or(N26,S07)or(L002,S07)(L002,L303)1

Notes.1.L001,L002,L06,L303,L306,and T001had been promoted just prior to the experiment,so these rules were removed from the candidate list.

2.T002was sold out,so this rule was also removed.

Table9

Nine bundling strategies selected for the?eld experiment

Product bundling strategies Discounted bundling strategies

Based on order data only(B15,B16),(B15,B18),(B15,B16,B18)

Based on browsing data only(L06,S07),(S07,T001),(L06,S07,T001)

Based on both browsing and

shopping-cart data

(B19,B21),(B21,B22),(B19,B21,

Abstract Comparison of product bundling strategies on different online shopping behaviors

B22)

Fig.2.Webpage showing the pricing of product bundles.

T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304301

publisher’s website(Fig.2)that o?ered discounts of approximately25%.This represents a type of?eld experi-ment because the experiment was embedded in the normal functioning of the publisher’s website,and customers did not know that it formed part of an experiment.All of the online shopping behaviors were recorded.

5.Data analysis

Before the?eld experiment,the publisher had provided bundling strategies for6months that were based on the manager’s expertise.During this period there were means of 2.7customers, 2.8orders,and nine books sold per month.Nine bundling strategies were promoted for1 month in our?eld experiment,during which there were 22customers,33orders,and78books sold,which indicates a clear improvement in the performance.

We chose the number of purchased books to be the mea-surement variable;that is,more books being purchased due to the discounted bundling strategies indicates a better per-formance.The mean and standard deviation values for the number of purchased books for each method are summa-rized in Table10.Two-way ANOVA was used to deter-mine whether the performance of applying both browsing and shopping-cart data to deciding product bundling was signi?cantly better than that of the other two strategies that applied one type of data only.The results are summarized in Table11.

In the experiment,every customer was allowed to pur-chase any product bundles extracted from the three strate-gies.Therefore,the sources of variation included both the preferences di?ering between users and the di?erent strate-gies.The results indicated that the purchasing of the dis-counted bundles was not signi?cantly a?ected by the users’preferences(F=0.55,p=0.929)but was signi?-cantly a?ected by the three product bundling strategies (F=3.78,p=0.031).

We next used Sche?e’s multiple-comparison test to esti-mate the95%simultaneous con?dence intervals of the dif-ferent strategies in order to compare the means for each pair of strategies.Table12indicates that the strategy based on both browsing and shopping-cart data performed signif-icantly better than the strategy based on order data only and the strategy based on browsing data only.Note that the performance did not di?er signi?cantly between the strategy based on order data only and the strategy based on browsing data only.

6.Conclusions

The Internet makes it easy to collect both browsing and shopping-cart data,in addition to traditionally collected order data.This study evaluated the value of this newly available information by comparing the performance of making decisions on product bundling based on di?erent types of data on online shopping behaviors.A?eld exper-iment revealed that the usefulness in making decisions on product bundling based on both browsing and shopping-cart data was signi?cantly higher than that when using order data only or browsing data only.

This study was subject to some limitations.The book-store chosen for performing the?eld experiment specializes in selling books on technology and electronic commerce, and sells a relatively small number of books.Further,a book essentially represents a type of one-time purchase and the?eld experiment was implemented for only1 month,resulting in a small total number of orders.Many users browsed the website without logging in,and hence their browsing and shopping-cart data could not be col-lected.Finally,the customers’online behavior would be impacted by several other factors such as the website design,the popularity of the publisher,and the maturity of electronic commerce.

There are some promising directions for future research–including several possible bene?ts of bundling

Table10

Means and standard deviations of purchasing for the discounted bundling

strategies

Strategy Mean Standard deviation

Based on order data0.818 1.296

Based on browsing data0.591 1.008

Based on both browsing

and shopping-cart data

2.136 2.765

Table11

Results of two-way ANOVA for the discounted bundling strategies

Source Sum of squares DF Mean square F p

Users47.1521 2.250.550.929

Methods30.64215.32 3.780.031*

Error170.0342 4.05

Total247.8265

Table12

Simultaneous95%con?dence intervals for the di?erent product bundling strategies

Con?dence interval(i–j)Strategy j

Strategy i Based on order data Based on browsing data Based on both browsing and shopping-cart data Based on order data0.227±1.089à1.318±1.089*

Based on browsing dataà0.227±1.089à1.545±1.089*

Based on both browsing and shopping-cart data 1.318±1.089* 1.545±1.089*

302T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304

such as promoting complimentary or new products–which were not considered in our research.How bundle pricing in?uences sales performance is another promising issue for future research,as is determining better data-mining techniques.Finally,a major challenge for Internet retailers is the high percentage of consumers who aban-don their virtual shopping carts without?nalizing their orders[28],and hence another worthwhile research topic is how best to interact with potential buyers so as to maximize the probability of their shopping carts being ?nalized as orders[7].

Acknowledgement

Hsiangchu Lai thanks the Taiwan National Science Council for the funding provided under Grant No. NSC89-2416-H-110-029.

References

[1]W.J.Adams,J.L.Yellen,Commodity bundling and the burden of

monopoly,The Quarterly Journal of Economics90(2)(1976)475–498.

[2]M.K.Agarwal,S.Chatterjee,Complexity,uniqueness,and similarity

in between-bundle choice,Journal of Product and Brand Manage-ment12(6)(2003)358–376.

[3]R.Agrawal,R.Srikant,Fast algorithms for mining association rules,

in:Proceedings of the20th VLDB Conference,Santiago,Chile,1994.

[4]R.Agrawal,T.Imielinski, A.Swami,Mining association rules

between sets of items in large databases,in:Proceedings of the1993 ACM SIGMOD International Conference on Management of Data, Washington,DC,1993.

[5]J.Balla,G.Desai,J.Fenner,Understanding e-commerce,Inform13

(8)(1999)28–30.

[6]R.Barrett,P.P.Maglio,D.C.Kellem,Autonomous interface agents,

in:Proceedings of the Conference on Human Factors in Computer Systems,Atlanta,GA,1997.

[7]R.Bayan,How business owners can prevent online shopping failures,

Link-up17(4)(2000)32–33.

[8]T.J.Blischok,Every transaction tells a story,Chain Store Age

Executive71(3)(1995)50–62.

[9]R.R.Burke,Virtual shopping:breakthrough in marketing research,

Journal of Product Innovation Management13(6)(1996)558–559.

[10]D.Chakravarti,R.Krish,P.Paul,J.Srivastava,Partitioned

presentation of multicomponent bundle prices:evaluation,choice and underlying processing e?ects,Journal of Consumer Psychology 12(3)(2002)215–229.

[11]M.-C.Chen,A.-L.Chiu,H.-H.Chang,Mining changes in customer

behavior in retail marketing,Expert Systems with Applications28(4) (2005)773–781.

[12]M.-S.Chen,J.Han,P.S.Yu,Data mining:an overview from a

database perspective,IEEE Transactions on Knowledge and Data Engineering8(6)(1996)866–883.

[13]Y.-L.Chen,K.Tang,R.-J.Shen,Y.-H.Hu,Market basket analysis

in a multiple store environment,Decision Support Systems40(2) (2005)339–354.

[14]R.Coase,J.L.Yellen,The problem of social cost,Journal of Law and

Economics3(1)(1960)1–44.

[15]G.Dalton,S.Gallagher,Online data’s?ne line,InformationWeek27

(1999)18–20.

[16]R.E.Dansby, C.Conrad,Commodity bundling,The American

Economic Review74(2)(1984)377–381.

[17]H.Demsetz,The cost of transacting,Quarterly Journal of Economics

82(1968)33–53.[18]H.Estelami,Consumer savings in complementary product bundles,

Journal of Marketing Theory and Practice7(3)(1999)107–114. [19]J.L.Gogan,The web’s impact on selling techniques:historical

perspective and early observations,International Journal of Elec-tronic Commerce1(2)(1997)89–108.

[20]J.P.Guiltinan,The price bundling of services:a normative frame-

work,Journal of Marketing51(2)(1987)74–85.

[21]P.J.Haynes,M.M.Helms,A.R.Casavant,Creating a value-added

customer database:improving marketing management decisions, Marketing Intelligence&Planning10(1)(1992)16–21.

[22]D.L.Ho?man,T.P.Novak,Marketing in hypermedia computer-

mediated environments:conceptual foundations,Journal of Market-ing60(3)(1996)50–68.

[23]M.Houtsma,A.Swami,Set-oriented mining for association rules in

relational databases,in:Proceedings of the1995International Conference on Data Engineering,Taipei,Taiwan,1995.

[24]C.Janiszewski,M.Cunha Jr.,The in?uence of price discount framing

on the evaluation of a product bundle,Journal of Consumer Research 30(4)(2004)534–546.

[25]J.Jiao,Y.Zhang,Product portfolio identi?cation based on

association rule mining,Computer-Aided Design37(2)(2005) 149–172.

[26]C.-R.Julander,Basket analysis:a new way of analysing scanner data,

International Journal of Retail&Distribution Management20(7) (1992)10–18.

[27]M.Kitsuregawa,M.Toyoda,I.Pramudiono,Web community

mining and web log mining:commodity cluster based execution,in: Proceedings of the13th Australasian Conference on Database Technologies,Melbourne,Victoria,Australia,2002.

[28]G.Koloszyc,Abandoned‘shopping carts’pose major challenge for

internet retailers,Stores81(7)(1999)41–44.

[29]P.Kotler,Marketing Management–Analysis,Planning,Implemen-

tation,and Control,Prentice-Hall,NJ,1997.

[30]http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli,An object-oriented architecture for intelligent virtual recep-

tionist,International Journal of Electronic Commerce4(3)(2000)69–

86.

[31]http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli,T.-C.Yang,A system architecture for intelligent browsing on

the web,Decision Support Systems28(3)(2000)219–239.

[32]http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli,T.-C.Yang,Applying shopping cart data to web customers

clustering,in:Proceedings of the Western Decision Sciences Institute 30th Annual Meeting(WDSI2001),Vancouver,Canada,2001. [33]H.Mannila,H.Toivonen,A.I.Verkamo,E?cient algorithms for

discovering association rules,in:Proc.AAAI Workshop Knowledge Discovery in Databases(KDD’94),1994.

[34]R.P.McAfee,J.McMillan,M.D.Whinston,Multiproduct monop-

oly,commodity bundling,and correlation of values,The Quarterly Journal of Economics104(2)(1989)371–383.

[35]http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmld,T.Reutterer,An improved collaborative?ltering approach

for predicting cross-category purchases based on binary market basket data,Journal of Retailing and Consumer Services10(3)(2003) 123–133.

[36]J.L.Munger,D.Grewal,The e?ects of alternative price promo-

tional methods on consumers’product evaluations and purchase intentions,Journal of Product and Brand Management10(3) (2001)185–197.

[37]A.Ovans,Make a bundle bundling,Harvard Business Review75(6)

(1997)18–20.

[38]T.R.Palfrey,Bundling decisions by a multiproduct monopolist with

incomplete information,Econometrica51(2)(1983)463–483. [39]J.S.Park,M.-S.Chen,P.S.Yu.,An e?ective hash-based algorithm for

mining association rules,in:Proceedings of the1995ACM SIGMOD International Conference on Management of Data,San Jose,CA, 1995.

[40]P.R.Peacock,Data mining in marketing:Part I,Marketing

Management6(4)(1998)8–18.

[41]J.Reedy,S.Schullo,K.Zimmerman,Electronic Marketing:Inte-

grating Electronic Resources into the Marketing Process,The Dryden Press,Harcourt College Publishers,TX,2003.

T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304303

[42]G.J.Russell,W.A.Kamakura,Modeling multiple category brand

preference with household basket data,Journal of Retailing73(4) (1997)439–461.

[43]G.J.Russell,A.Petersen,Analysis of cross category dependence in

market basket selection,Journal of Retailing76(3)(2000)367–392.

[44]C.Rygielski,J.-C.Wang,D.C.Yen,Data mining techniques for

customer relationship management,Technology in Society24(4) (2002)483–502.[45]J.Strauss,A.I.El-Ansary,R.Frost,E-Marketing,Prentice Hall,NJ,

2003.

[46]S.Stremersch,G.J.Tellis,Strategic bundling of products and prices:a

new synthesis for marketing,Journal of Marketing66(1)(2002)55–

72.

[47]M.S.Yadav,K.B.Monroe,How buyers perceive savings in a bundle

price:an examination of a bundle’s transaction value,Journal of Marketing Research30(3)(1993)350–358.

304T.-C.Yang,http://www.wendangku.net/doc/2e34f6264b35eefdc8d333ed.htmli/Electronic Commerce Research and Applications5(2006)295–304