The hare and the tortoise: do earlier adopters of online channels purchase more?

(1)

channels purchase more?

Citation for published version (APA):

Li, J., Konus, U., Pauwels, K. H., & Langerak, F. (2015). The hare and the tortoise: do earlier adopters of online channels purchase more? Journal of Retailing, 91(2), 289-308. https://doi.org/10.1016/j.jretai.2015.01.001

DOI:

10.1016/j.jretai.2015.01.001

Document status and date: Published: 01/01/2015 Document Version:

Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers) Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website.

• The final author version and the galley proof are versions of the publication after peer review.

• The final published version features the final layout of the paper including the volume, issue and page numbers.

Link to publication

General rights

Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain

• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please follow below link for the End User Agreement:

www.tue.nl/taverne

Take down policy

If you believe that this document breaches copyright please contact us at:

openaccess@tue.nl

(2)

JournalofRetailingxxx(xxx,2015)xxx–xxx

The

Hare

and

the

Tortoise:

Do

Earlier

Adopters

of

Online

Channels

Purchase

More?

Jing

Li

a,∗

,

Umut

Konus¸

b,1

,

Koen

Pauwels

c,2

,

Fred

Langerak

a,3

a_Eindhoven_University_of_Technology,_Innovation,_Technology_{Entrepreneurship}_&_Marketing_Group,_School_of_Industrial_Engineering,_The_Netherlands b_Faculty_of_Economics_and_Business,_University_of_Amsterdam,_The_Netherlands

c_Ozyegin_University,_Faculty_of_Economics_and_{Administrative}_Sciences,_Turkey

Abstract

Earlieradoptersofaproductorservicetendtobemorevaluablethanlateradopters.Doesthisempiricalgeneralizationequallyapplytoearlier adoptersofamultichannelretailer’snewonlinechanneltoo?Thisstudysegmentscustomersonthebasisoftheirresponsestoanewonlinechannel andinvestigatestheeffectsoftheironlinechanneladoptiononpurchasevolumesacrosssegments.Thedatacover12.5yearsofpurchasehistory andindividualtransactionsatalargemultichannelFrenchretailerofnaturalhealthproducts.Contrarytoconventionalwisdom,itisnotinnovators orearlyadopters,butratherthelatemajoritysegmentthatpurchasesmorethantheothersegments,bothbeforeandafteronlineadoption.Adoption ofthefirm’snewonlinechanneldoesnotinfluencepurchasevolumesofheavyshoppersegments(latemajorityandinnovators),whereaslight shoppersegmentstendtoincreasetheirpurchasesafteradoptingthisnewchannel.

Keywords:E-commerce;Channels;Segmentation;Onlineretailing;Customerrelationshipmanagement;Channeladoption

Areearlieradopterskeytomarketingsuccess?Whenitcomes totheadoptionofnewproductsandservices,researchshowsthat earlieradopterspurchaseanduseproductsmoreoftenandare greatlyinfluencedbymediapromotions(GoldsmithandFlynn 1992;Mahajan,Muller,andSrivastava1990).Theyalsomaybe moreprofitablethanlateadopters, becausefirmsoftencharge apremiumpriceinthe earlyphasesof aproduct’slifecycle. Furthermore,earlieradoptershavecriticalinfluencesonuptake

∗_{Corresponding}_author_at:_Innovation,_Technology_{Entrepreneurship}_&

Mar-keting Group, School of Industrial Engineering, Eindhoven University of Technology,P.O.Box513,5600MB,Eindhoven,TheNetherlands.Tel.:+31 402475693;fax:+31402468054.

E-mail addresses: jing.li@tue.nl (J. Li), u.konus@uva.nl (U. Konus¸),

koen.pauwels@ozyegin.edu.tr(K.Pauwels),f.langerak@tue.nl(F.Langerak).

1_Address:_Faculty_of_Economics_and_Business,_University_of_Amsterdam,

PlantageMuidergracht12,1018TVAmsterdam,TheNetherlands.

2_Address:_Faculty_of_Economics_and_{Administrative}_Sciences,_Ozyegin

Uni-versity,NisantepeMah.OrmanSok.34794Cekmekoy,Istanbul,Turkey.

3_Address:_Innovation, _Technology_{Entrepreneurship}_& _Marketing_Group,

SchoolofIndustrialEngineering,EindhovenUniversityofTechnology,P.O. Box513,5600MB,Eindhoven,TheNetherlands.

decisionsby lateradopters, because theyspreadthe attitudes or satisfaction theydevelop toward the innovation (Mahajan, Muller,andWind2000).Inconsideringbothfinancialandsocial effects,Hogan,Lemon,andLibai(2003)emphasizethattheloss of anearlieradoptercostsafirmmuch morethanthe lossof alateradopter.Bytargetingearlieradopters,firmscanensure fasterreturnsontheirinvestmentsandtakeadvantageofsocial spillovereffectsfordiffusingnewproducts.

However,areearlieradoptersalsocritical tothesuccessof a newly introduced marketing channel? Driven by the Inter-netandmobiletechnology,retailersincreasinglyintroducenew onlinechannelstosupplementexistingchannels,retainexisting customers, and acquire new customers. Existing offline cus-tomersadopttheretailer’snewonlinechannelatdifferenttime periodsandpurchasethroughmultiplechannels;theresulting multichannel shoppers spendmorethan single-channel shop-pers(Ansari,Mela,andNeslin2008;Neslinetal.2006;Thomas andSullivan2005).Somestudiessuggestthat customerswho arefastertoadoptanew(online)channel exhibitgreater pur-chase frequencyand transaction volume before the adoption (Venkatesan, Kumar, and Ravishanker 2007; Xue, Hitt, and

http://dx.doi.org/10.1016/j.jretai.2015.01.001

(3)

Chen2011); no study has however investigated the different behaviorsorfeaturesofcustomergroupsthatadoptaretailer’s newchannelearlieror laterthanothercustomersthough.For example,doinnovatorsor earlyadoptersof newonline chan-nels purchase more than the majoritysegments or laggards? Canwe distinguishamongsegments thatadopt newchannels at different periods? Identifying the most valuable customer groupsandunderstandingtheircharacteristicscouldhelp retail-ersallocatetheirlimitedmarketingresourcesmoreeffectively acrosscustomersegmentsandthusimproveoverallprofits.So we investigate the monetary contributions andcharacteristics ofdifferentcustomersegments,identifiedonthebasisoftheir adoptiondurationofnewlyintroducedonlinechannelsandtheir purchaseamountspriortothatadoption.

In addition, we investigate the extent towhich customers change their purchase volumes due to online channel adop-tion.Plentyofstudiesinvestigatetheeffectsofonlinechannel adoption or use on customer shopping behaviors over time (e.g.,Ansari,Mela,andNeslin2008;CampbellandFrei2009; Gensler,Leeflang,andSkiera2012).Farlessresearchexplores itseffectsonconsumerbehaviorsacrossdifferentsegments,with thenotableexceptionofPauwelsetal.(2011),whoinvestigate theinfluenceofaninformationalwebsite.Weseektoextendthis literaturestreambyempiricallyinvestigating theeffectsofan onlinetransactionalchannelonpurchasesbyvarioussegments thatadoptthechannelatdifferenttimes.Iftheeffectsvaryacross segments,firmsshoulddifferentiatetheirmultichannelstrategies accordingly.Thusweinvestigatetwokeyresearchquestions: 1. Doearlier adoptersof aretailer’s online channelpurchase

morethanotheradoptersegments,identifiedonthebasisof theiradoptiondurationofnewlyintroducedonlinechannels andpurchaseamountspriortotheonlineadoption? 2. Howdoescustomeradoptionoftheretailer’sonline

transac-tionalchannelsaffectpurchasevolumesofdifferentcustomer segments,identifiedbyadoptionduration?

We rely on latent class cluster analysis (LCCA) to seg-mentcustomersaccordingtotheironlineadoptiondurationand purchase amountsbefore adoption,then profilethe identified segmentsusingvariouscovariatesrelatedtotheirdemographics andshoppingbehaviorsafteradoption(VermuntandMagidson 2005).To estimate theimpact of online channel adoption on customerbehavior, wecontrol for thepotential effectof cus-tomerself-selection(Boehm 2008;Campbell andFrei2009). Thusinthesecondstep,weemployapropensityscorematching (PSM)techniquetodetermineamatchedofflinecustomergroup foreachonlineadoptersegment(Dehejia2005;Rosenbaumand Rubin1985).Finally,foreachsegment,weapplyaTypeIITobit modeltoinvestigatetheimpactofonlinechanneladoptionon monthlypurchaseincidenceandmonetaryvaluepertransaction (ordersize)overtime.Tosupplementthismodel,weundertake adifference-in-differenceanalysis(DID)toexaminechangesin purchasevolumeandfrequency,oneyearaftertheadoptionof theonlinechannel(CampbellandFrei2009).

Withtheseapproaches,ourresearchrevealsthatthe heavi-est shoppers are neither innovators nor early adopters of a

newonlinechannelbutratherthelatemajoritysegment.Most research oncustomers’ adoption of newproducts or services focuses onthe contributions of earlieradopter segments;our studyrevealsthatlateradopters(latemajority)canbethemost valuablecustomergroup,bothbeforeandaftertheonline chan-neladoption.Inaddition,wedemonstratetheeffectsofonline channel adoption on purchase volumes across different seg-ments, whichcanhelpfirmspredicttheconsequencesoftheir online channel introduction more precisely and identify key challenges for different customer segments. Considering that our results show that purchases by heavy shopper segments (i.e.,latemajorityandinnovator)areunaffectedbytheir adop-tion ofonline channels,whereascustomersinothersegments (i.e.,earlyadopter,earlymajority,andlaggard)tendtoincrease theirpurchasevolumesafteradopting,retailersshouldconsider developingdifferentstrategiestoaddresssegment-specific chal-lenges.

ConceptualDevelopment

The two-partconceptualframeworkof thisstudyinFig.1 features (1) customer segmentation on the basis of customer heterogeneity (leftside) and(2) theeffects of online channel adoptiononpurchasevolumesacrossdifferentsegments(right side).

IdentifyingCustomerSegments

Increasing varietyof marketingchannels allowscustomers toadoptnewchannelsandbecomemultichannelshoppers.For retailers,multichannelcustomersegmentation,whichsegments customersaccordingtotheirshoppingbehaviorsacrossmultiple channels, offersaneffective methodfordesigning multichan-nel marketing strategies (Neslinet al.2006).The underlying logic is that customers self-select into channels that invoke different costs,relatedto time,travel, shopping,andso forth (Anderson, Day, and Rangan 1997); in addition, psycholog-ical and economic attitudes, together with expected benefits andcosts,affectchannelpreferencesanduses(Konus,Verhoef, and Neslin2008).Forexample, Thomas andSullivan(2005) identify two customer segments—multichannel shoppers and store-onlyshoppers—andcitetheimpactsofprice,product cat-egory,distance,marketingspending,andpreviouspurchaseson channel choices.Konus, Verhoef, andNeslin(2008)segment customers by channel choices across multiple phases of the shoppingprocess.Differentfrompreviousstudies,weidentify customersegmentsbasedontwoindicators:adoptionduration andpurchaseamountbeforeonlinechanneladoption.

Adoption duration. Customers adopt product and service innovations at different times after launch(Mahajan, Muller, andSrivastava1990;Rogers2003).Dependingonhowquickly the adoption takes place, Rogers (2003)classifies consumers into five groups:innovator,early adopter,early majority,late majority,andlaggard.Thesesegmentsdifferintheir demograph-ics, psychographics,socialclass,andlifestyles(Gatignonand Robertson1985;Rogers2003).Forexample,earlyadopterstend tohavehigherincomeandstatusoccupations,moreeducation,a

(4)

• Online adoption duration • Purchase amount before online

channel adoption

Customer segmentation based on customers’ responses to online channel introduction

Indicators

• Age • Gender

Covariates (active)

Customer

Segments Purchase Volume

Online Channel

Adoption Self-Selection

Effects of online channel adoption in different segments

Control Variables • Demographic characteristics • Previous purchase • Seasonality • Competition • Economic climate

Fig.1.ConceptualFrameworkofOnlineChannelAdoption.

sociallyforwardattitude,andmoreexperiencewithother tech-nicalproducts(Mahajan,Muller,andSrivastava1990;Rogers 2003).Innovatorstendtoberisk-taking,impulsive,dominant, inner-directed,flexible,andventuresome(FoxallandGoldsmith 1988; Goldsmith and Flynn 1992). Groups that adopt inno-vations at different times might have distinct shopping and behavioral patterns too, such that earlier adopters use new products more frequently but only for their basic functions (GoldsmithandFlynn1992;Mahajan,Muller,andSrivastava 1990). Similarly, earlier adopters of new e-services exhibit higherserviceusagelevelsthanlateadopters(PrinsandVerhoef 2007).

Purchasesbeforeonlinechanneladoption.Weareprimarily interested in comparing segments’ purchases that are repre-sented by the purchase amounts—monetary contributionsby customers (Campbell and Frei 2009; Gensler, Leeflang, and Skiera2012).Moretransactionswithafirmmightenhance cus-tomertrustmorequickly(MorganandHunt1994)andshorten thetimebeforethecustomeradoptsthefirm’snewchannel. Cus-tomerexpendituresalsocontributetobehavioralloyalty,which acceleratesadoptionspeed(DemoulinandZidda2009). Empir-ical evidence shows that customers who adopt transactional channelsfasteralsoexhibitgreatertransactionalfrequencyprior totheiradoption (Venkatesan,Kumar, andRavishanker2007; Xue,Hitt,andChen2011).Therefore:

H1. Segmentswithhigherpriorpurchaseamountsadoptnew

(online)channelsfaster.

Oneofthecrucialaspectsofthesegmentationframeworkisto exploretheimpactofcovariatesonthemembershipof segmen-tationandtoprofilefeaturesofidentifiedsegmentsaccordingto thesecovariates.

Covariates.Weincludecustomerdemographics,suchasage and gender, in our framework as covariates that can affect the segmentation membership. Such demographic variables influencechanneladoptionduration(Venkatesan,Kumar, and

Ravishanker 2007; Xue, Hitt, and Chen 2011), and chan-nel choice (Ansari, Mela, and Neslin 2008; Inman, Shankar, and Ferraro 2004). For example, Venkatesan, Kumar, and Ravishanker (2007)find that male customersare morelikely to adopt additional channels faster, but their income lev-els do not affect channel adoption. Xue, Hitt, and Chen (2011) identify a curvilinear relationship between age and online channel adoption speed: younger customers likely exhibit quickeradoption.Becausetheeffects of demographic controls on behaviors often are insignificant or inconsistent (Konus, Verhoef, and Neslin 2008), we do not formulate a formal hypothesis but rather include age and gender as covariates.

Movingfrom predictingchanneladoption topredictingits consequences,wenextdiscusswhetherandhowonline chan-nel adoptionimpacts customerspending for differentadopter segments.

EffectsofOnlineChannelAdoptiononPurchaseVolumesof DifferentCustomerSegments

Extensivemultichannelmanagementstudiesinvestigatethe effectsofonlinechanneladoptionandusageoncustomer behav-iors and firm performances over time. Some studies employ aggregated, firm-level data; for example, Geyskens, Gielens, andDekimpe(2002)determinethataddinganInternetchannel acceleratesstockmarketreturns,andLeeandGrewal(2004)find similarresultsinthecompactdisccategory.Anotherresearch streamfocusesondisaggregateddata,relatedtoindividual cus-tomerpanels.Gensler,Leeflang,andSkiera(2012)revealthat the use of online channels increases customer revenue. The researchofBoehm(2008)indicatesastrongpositiveimpactof onlinechanneluseoncustomerretention.Althoughmoststudies suggestthatonlinechanneladoptionandusepromotecustomer demand,Ansari,Mela,andNeslin(2008)findthatonlineusage isnegativelyassociatedwithlong-termpurchasefrequency.

(5)

Despiterichresearchontheconsequencesofonlinechannel additions, few studies investigate the effects of online chan-nel adoption byconsideringthe impactof different customer segments(e.g.,Pauwelsetal.2011).Asiswellestablishedin marketing, customer heterogeneity critically affects customer responses toafirm’smultichannelstrategies (e.g., Kushwaha andShankar2013;ThomasandSullivan2005).Therefore,we expectthattheeffectsofonlinetransactionchanneladoptionon customerpurchasesmightvaryacrosscustomersegmentsthat differintheirpurchasevolumespriortoonlinechanneladoption. Toformulateourhypotheses,weclarifypreciselywhyweexpect customerstoaltertheirshoppingbehaviorsinresponsetoonline channelintroduction,accordingtotwoopposingmechanisms: intrinsicbenefitsandmarketingcommunications.

Intrinsic beneﬁts. Customers change their behaviors after onlinechanneladoption,becauseof thebenefitstheyperceive from online shopping. The online shopping makes it easier forcustomerstosearchforinformationandcompareproducts (Ariely2000).Therefore,customersperceivegreater informa-tion control than they would if they relied solely on offline channels (Gensler,Leeflang, andSkiera 2012).Greater infor-mation control likely leads to higher customer satisfaction and higher repurchase rates (Meuter et al. 2000; Mittal and Kamakura 2001). Moreover,online channels offer customers greater convenienceand accessibility,through constant avail-abilityandinteractivity,theconvenienceofbuyingfromhome, andenhancedaccesstopersonalizedoffers(Gensler,Leeflang, andSkiera2012;Montoya-Weiss,Voss,andGrewal2003;Wolk andSkiera 2009). Finally, shoppingonline can reduce trans-action costs, including the costs of search, travel, time, and physics (Chircu and Mahajan 2006; Varadarajan and Yadav 2002),thoughthesecostsalsodependoncustomer heterogene-ity(Chintagunta,Chu,andCebollada2012).Becauseofthese benefits,customer’overallpurchasevolumesfromonlineand offlinechannelslikelyincreaseaftertheyadoptafirm’sonline channel(CampbellandFrei2009;Xue,Hitt,andChen2011).

Segments of heavy shoppers may perceive fewer benefits of onlineshopping thanlightshopper segments though. Cus-tomers’ perceptions of the usefulness and use of innovative technology(e.g.,Internetchannel) dependontheir preference forthestatusquo(Falketal.2007;Limayem,Hirt,andCheung 2007),andthehabitualbehaviorformsthroughmultiple repe-titionsofdecisions(Aarts,Verplanken,andKnippenberg1998; Orbelletal.2001).Becausefrequentinteractionswithoffline channelscultivateofflineshoppinghabits,heavyshopperslikely induceastrongerpreferenceforthesechannelsthanisthecase forlightershoppers.Falketal.(2007)notethatsatisfactionwith offlinechannelsreducestheperceivedusefulnessandenhances theperceivedriskofonlineshopping,andMontoya-Weiss,Voss, and Grewal (2003) show that positive perceptions of service qualityinanexistingchannelcaninhibitusesofanewonline channel. Moreover,according toKonus, Neslin, andVerhoef (2014),heavyshoppersarelessaffectedbychangestoits chan-nelrepertory(e.g., eliminationof acatalogchannel),possibly becausethesecustomers,whoaremorefamiliarwiththefirm’s offerings,perceivefewerchangestotheirshoppingbenefits(e.g., searchconvenience,enjoyment).Integratingthesefindings,we

predictthatheavyshopperslikelyperceiveonlineshoppingas lessusefulandbeneficialthanlightshoppers.Ifcustomer behav-iormainlyreflectstheintrinsicbenefitsofonlineshopping,we expect:

H2. Onlinefirmchanneladoptionhasmorepositiveeffectson

thepurchasevolumesoflightshoppersegmentsthanonthose ofheavyshoppersegments.

Marketingcommunications.Customersaltertheirpurchase volumes after adopting online channels, likely because they receive more marketing contacts through varied channels (KumarandVenkatesan2005;Neslinetal.2006).Ansari,Mela, and Neslin (2008) note that multichannel customers process moremarketingmessagesandrespondmorefrequentlyto mar-ketingcommunications.

The extent to which customers alter their behaviors after adopting online channels likely differs across segments, as customers respond differently to marketing communications. Existingliteraturedemonstratesthatheavyshoppersaremore responsive to advertising, price cuts, and coupons, because they cangain morefrom such promotions (Krishna, Currim, and Shoemaker 1991; Vanhuele and Drèze 2002; Zhang, Seetharaman, and Narasimhan 2012).Moreover, heavy users exhibithighershoppingdemandandcanabsorbadditional quan-tities, because they tend to have larger families and live in larger houses (Neslin, Henderson, and Quelch 1985; Zhang, Seetharaman, andNarasimhan 2012).Ifcustomerbehavioris mainlyaffectedbymarketingefforts,weproposeanalternative hypothesis:

H3. Onlinefirmchanneladoptionhasmorepositiveeffectson

thepurchasevolumesofheavyshoppersegmentsthanonthose oflightshoppersegments.

Self-selection. Customers withcertain characteristics have intrinsic preferences for a particular channel (Boehm 2008; Konus, Verhoef, and Neslin 2008). The differences in char-acteristicsbetweenonlineadoptersandofflinecustomersalso exist,suchasage,education,andpurchaselevelbeforeonline adoption (HittandFrei2002;Neslinetal.2006;Verhoefand Donkers 2005).Variousstudiesshowthat ignoringsuch self-selection biasesleads toinaccurate estimations of the effects ofonlineadoptionoruseoncustomerbehavior(Boehm2008; CampbellandFrei2009;Gensler,Leeflang,andSkiera2012). Therefore,weemployamatchingtechniquetoensureamatch inthecharacteristicsofonlineadoptersandofflinecustomers.

Control variables. Finally, we control for several factors thatcouldaffectcustomershoppingbehaviors:customer char-acteristics, previous purchases,competition,andtimefactors. Demographic characteristics includeage andgender (Ansari, Mela, andNeslin2008; Xue,Hitt, andChen 2011). Wealso consider the effect of previous purchases on current pur-chases,knownasstatedependenceorinertia(Rust,Lemon,and Zeithaml2004;Valentini,Montaguti,andNeslin2011).Because ourdataspanalongperiod,wecontrolfortheimpactoftime oncustomerspending,suchas seasonality(Ansari,Mela,and Neslin2008;Pauwelsetal.2011).Finally,wecontrolfor the effects of externalfactors,such as competitionandeconomic

(6)

climate(recession),whichcouldinfluencecustomershopping behaviorsandexperiences(VanDiepen,Donkers,andFranses 2009;Verhoefetal.2009).

DataDescription Data

WeuseddailytransactionaldatafromamultichannelFrench retailer that sells healthy and natural products. With the aid ofaFrenchmultichanneldata-warehouseandconsulting com-pany,wecollectedtransactionaldatafromcompetitors,namely 16French retailerscompeting inthe sameindustry.Ourdata setthuscontainsindividualtransactionpanels(i.e.,transaction date,purchaseamount,andtransactionchannel)fromboththe focalfirmanditscompetitors.Transactionscollectedfrom com-petitorsconstituted6.3%oftotaltransactions,whichweused to control for the effect of competition. This data set spans 12yearsandseven months,fromJanuary 2000toJuly2012. Thefocalretailerhadtwoestablishedofflinepurchasechannels (callcenterandcatalog),thenintroducedanewonlinechannel inJanuary 2001.Thus,wehad1 yearof observationprior to theonlinechannelintroductionand11.5yearsafter,asFig.2 details.

Toinvestigatetheprocessbywhichexistingofflinecustomers adoptandevolveinrelationtoanewlyintroducedonline chan-nel, we selected a random set of 3,270 customers who had purchasedfromtheretailerbeforetheonlinechannel introduc-tion.Allthesecustomersstartedpurchasingfromthefocalfirm intheyear2000.Inthisset,2,180(66.7%)customersadopted theonlinechannelbytheendofthedataperiod,whereas1,090 remainedoffline customersdid not adopt. We also used two additionalcriteriatoselectthefinalsampleforanalysis.First, so that we could examine the effects of online adoption on customerbehavior,theonlineadoptershadtohavemade pur-chasesfromthisfirmlongerthanoneyearpriortoandoneyear aftertheironlineadoption.Wethusidentifiedasampleof1,695 onlineadopters.Second,weexcludedcustomerswhoterminated their shopping relationshipwiththe firm in the earlyperiod, becauseourfocusofinterestisontheeffectofonlineadoption oncustomerrevenue,ratherthancustomerchurn.Specifically, eachselectedcustomerhadtopurchaseatleastonetimefrom thefocalfirminthelasttwoyears,whichexcluded45online adoptersand 105 offlinecustomers. The selection procedure thusyieldedasampleof1,650onlineadopterswhoadoptedthe onlinechannelbetweenJanuary2001–June2011and985offline onlycustomers.

Weprovidethedemographicdescriptionsandpurchase infor-mation(fromthefocalfirm)abouttheonlineadoptersandoffline customersinTable1.Inlinewithpreviousstudies(e.g.,Boehm

2008; Campbell and Frei 2009),on average, online adopters are younger (44 years) than offline customers (52 years). Online adopters’ annual purchase amounts are lower (162.1 Euros) thanthe yearly purchases of offline customers(191.4 Euros),whichconflictswithfindings thatindicate multichan-nel customersspendmorethanoffline-onlyor single-channel customers(Neslinetal.2006;ThomasandSullivan2005),but itisnotabnormalforthehealthandnaturalproductscategory. Theseproductlinestendtobemoreexpensive forolderthan youngershoppers,sotheolder,offlinecustomerslikelypurchase largervolumesthanyounger,onlineadopters.Annualpurchase amounts vary greatly across customers, from 13.8 to 2715.3 Eurosforonline adoptersand12.4to1486.8Eurosforoffline customers.

Outliers

Tocontrolfortheeffectsofextremeoutliers,westandardized the yearly purchase amountsfor each customerand dropped customerswithstandardscoresof4orgreater(Hairetal.2010). Thusweexcluded20onlineadoptersand7 offlinecustomers fromthedataset,yieldingfinalsamplesof1,630onlineadopters and978offlinecustomersforthemodeling.

Model-FreeEvidence

We explored the purchase volumes of different customer groups who adopted the online channel at different periods. Because the maximum adoption duration is 124 months, we equally divided this time length into three periods, thus get three groups that adopt online at different times: early adopters (duration ≤40 months), middle-period adopters (40 months<duration ≤80 months), and late adopters (duration >80months).Table2summarizestheaverageannualpurchase amountsforthesesegments.Theyearlypurchaseamountswere similaracross segments,but differentpatterns emergedwhen weseparatedtheamountspentpriortoonlinechanneladoption from theamount spentafter it. Inline withour expectations, earlyadoptersspendmorethanothersegmentsbeforeadopting; however,late adoptersgeneratemorerevenuesper yearafter the adoptionevent.Wecannotjumptoconclusionsfromthis simple analysis, butthe model-freeexploration suggeststhat variousshoppingpatternsemergeamongcustomergroupswho adoptonlinechannelsatdifferenttimes.

Methodology

Ourmethodologyconsistsofthreestepsandaseriesof mod-els.Wefirstuselatentclassclusteranalysis(LCCA)tosegment customers onthe basis of their online adoption durationand

Online channel introduction

Offline customers adopting online

July 2012 January 2000

January 2001 July 2011

(7)

Table1

Descriptivestatisticsfortwocustomergroups.

Variable Onlineadopters Offlinecustomers

M SD Min Max M SD Min Max

Age(inyears) 44 10 20 91 52 11 22 82

Gender(female) 96.7% 97.1%

Purchasesperyear 1.9 1.9 0 23.0 2.2 1.8 0 18.3

Purchaseamountperyear(inEuros) 162.1 168.2 13.8 2715.3 191.4 154.3 12.4 1486.8

Onlineadoptionduration(inmonths) 73 28 3 124

Numberofonlinepurchases 5.6 9.3 1 165

Table2

Comparisonofpurchaseamountsacrossadoptionperiods.

Variable Adoptionduration≤40months 40months<adoptionduration≤80months Adoptionduration>80months

M Min Max M Min Max M Min Max

Yearlypurchaseamount 148.1 20.8 1108.8 151.8 13 1623.6 159.2 15.3 2696.5

(125.8) (152.4) (188.8)

Yearlypurchaseamount beforeadoption

156.4 10.2 853.3 133.1 1 1309.6 145.3 5.9 2196.5

(134.2) (130.9) (179.4)

Yearlypurchaseamount afteradoption

147.3 13.2 1482.7 170 12.8 2434.4 200.5 17.1 3807.7

(147.0) (213.1) (272.7)

Notes:Valuesinbracketsdonatethestandardizeddeviation.

purchase amountbefore adoption. Next,we useapropensity scorematching(PSM)techniquefor each identifiedsegment, to control for the effect of self-selection. Finally, by apply-ingaTypeII Tobitmodel anddifference-in-difference(DID) analysis,weinvestigatetheimpactofonlinechanneladoption onpurchasevolumesofdifferentsegments.Fig.3summarizes themodelingpurpose,correspondingmethod(s),anddata for-mat. The original datafollowed anunbalanced panel format, butweconvertedthedatasetintoacross-sectionalorbalanced panelformat,dependingontherequirementofeachmodeling purpose.

LatentClassClusterAnalysis(LCCA)

TheLCCAsegmentscustomersonthebasisofonline adop-tiondurationandpurchaseamountsbeforeadoption,whilealso

consideringtheimpactofcovariatesoncustomermembership (VermuntandMagidson2005),withthefollowingmodel spec-ification: f(y_i|zact_i cov)= K x=i P(x|zact_i cov)J j=1f(yij|x) (1)

whereyidenotesasetofJresponsevariables(indicators)that measurecustomeri’sresponsetothenewonlineintroduction, andyj isaparticularindicator. Inour case,the indicatorsare adoptiondurationthatisthenumberofmonthsbetweenonline channelintroductionandadoption,andyearlypurchaseamount before adoption.Thelatent variable(x)iscategorical,withK values, which corresponds to K segments. It is unnecessary topredict apriori thenumberof segments;rather,Kis deter-minedbythemodelselectioncriteria(VermuntandMagidson 2005). Furthermore, zact_i cov indicates the vector of active

Latent class cluster analysis

Propensity score matching

Diff erence-in-Difference analysis

(a)

Type II Tobit model (b)

Modelling Method Data Format

Cross-sectional

(a) Cross-sectional (b) Balanced panel

Modelling Purpose

Step 1: Identifying customer segments

Step 2: Eliminating self-selection effects

Step 3: Investigating the effects of online channel adoption on purchase volumes

(8)

Table3

VariablemeasurementsofTypeIITobitmodel.

Variable Measurements

Postadoption =1ifacustomerhasadoptedtheonlinechannelbeforethecurrentmonth;=0otherwise

Treatedgroup =1ifacustomeristheonlineadopter;=0otherwise

Pastonlinepurchase =1ifacustomerpurchasesthroughtheonlinechannelfromthefocalfirminthelastmonth;=0otherwise Pastofflinepurchase =1ifacustomerpurchasesthroughofflinechannelsfromthefocalfirminthelastmonth;=0otherwise Purchasefromcompetitors =1ifacustomerpurchasesfromcompetitorsinthecurrentmonth;=0otherwise

Lastordersize Amountofmoneyacustomerspentonthelastpurchase

Age Ageofacustomerinthecurrentmonth

Gender =1male;=0female

Recency Numberofmonthssincethelastpurchase.

Economicrecession =1ifcurrentmonthisin2001–2003or2008–2010;=0otherwise Seasonality1:March =1ifthecurrentmonthisMarch;=0otherwise

Seasonality2:August =1ifthecurrentmonthisAugust;=0otherwise Seasonality3:April&May =1ifthecurrentmonthisAprilorMay;=0otherwise Seasonality4:June&October =1ifthecurrentmonthisJuneorOctober;=0otherwise.

covariates (age andgender) that could affect the latent vari-ablebuthavenodirectinfluenceontheresponsevariables.We alsoincludedinactivecovariates—onlineshoppingpreference4 andyearlypurchaseamountafteronlineadoption—todescribe customers’behaviorsofidentifiedsegmentsafteradoption.As consequencesofonlinechanneladoption,thesevariablesdonot affectthelatentvariableormodelestimation.Finally,f(yij|x) representstheprobabilitydistributionofcustomeri’sresponse toaparticularindicatorj,giventhatcustomeribelongsto seg-mentx, andf(y_i|z_iact cov) isthe joint probabilityfunction of customeri’sresponsetoallindicators,asinfluencedbyactive covariates.

PropensityScoreMatching(PSM)Method

Abasicapproachtotesttheeffectofonlinechannel adop-tionistomeasurethechangesinacustomer’spurchasesafter adoption(i.e.,purchaseincidenceandordersize),relativetoa controlgroupofofflinecustomerswhodonotadopt.Weused PSMmethodtofindamatchedcontrolgroupforeachidentified onlineadoptersegment,whichalsocontrolsfortheself-selection bias.ThebasicideaofPSMmethodistofindmatchedsamples (i.e.,offlinecustomers)whohavetheclosestpropensityscores tothoseofthetreatedgroup(i.e.,onlineadopters).The propen-sityscoreistheprobabilitythataunitinthefullsamplereceives thetreatment,givenasetof observedcharacteristics (Dehejia 2005).Withthismethod,wecanensurethatthedistributionof characteristics inthetreatedandmatchedgroups isthe same (RosenbaumandRubin1985).

Weusedabinarylogisticmodeltoestimatetheprobability that acustomeradoptsthe newonline channel,as afunction ofpurchase volumesprior toadoption(average monthly pur-chasefrequency,averageordersizepertransaction,oraverage monthlypurchaseamount),age(inyears),gender,andtenure(in

4 _With_a_{random-effect}_logistic_model,_we_calculate_online_shopping

prefer-ence,thatis,theprobabilitythatacustomershopsonlineversusofflineinthe periodaftertheonlineadoption.Themodelfunctiondependsontheamountsof thepreviousonlinepurchaseandofflinepurchase,cumulativenumbersofonline purchasesandofflinepurchasebeforethecurrentpurchase,age,andgender.

months).Becausecustomersinthesamesegmentcouldadopt theonline channelindifferentperiods,itisdifficultto antici-pateadoptiondurationforofflinecustomersatthismoment.We insteadcalculatedpreviouspurchasevolumesintheperiodprior totheadoptionoftheearliestadopterinasegment.Thecontrol groupcomesfromthedatapoolof the978offlinecustomers, followingtherulesofone-to-onematchingwithoutreplacement. Specifically,foreachonlineadopter,wechoseamatchedoffline customerwhohasthe closestestimatedpropensity score.We also setupacalipertoguaranteethat theabsolute difference between thepropensity of an online adopterandits matched offlinecustomerisless thanacertainthreshold. Witha com-monsupportrestriction,werequiredallcustomerstoliewithin a regionof common support (Heckman,Ichimura, andTodd 1997).Thisapproachexcludesonlineadopterswithpropensity scoressmaller(larger)thantheminimum(maximum)valueof thepropensityscoresofthecontrols.

Difference-in-Difference(DID)Analysis

Usingthematchedsamples,we testedtheeffectsofonline channeladoptionintwocomplementaryways.ADIDanalysis comparesthechangesofcustomerbehaviorsbeforeandafterthe adoption eventbetweentreated(adopters)andcontrol groups (CampbellandFrei2009).Thus,wemeasuredchangesinterms oftotalpurchaseamount,totalpurchasefrequency,offline pur-chaseamount,andofflinepurchasefrequencybetweenoneyear priortoandoneyearaftertheadoptionofonlinechannels.The onlineadoptiondurationofanofflinecustomerequalsthe adop-tiondurationofthe matchedonlineadopter.Ifthechangesin performances differ statistically between the group of online adoptersandtheirmatchedofflinecustomers,weconcludethat onlineadoption significantlyaffectscustomerpurchases.This simpleDIDmethodprovidesusefulinformationabouttheeffect ofonlineadoptiononbehavioralchanges,butitmaynotcontrol adequatelyfordifferentialpostadoptiontrendsbetweenonline adopters and the control group that result from factors that emergeovertime(e.g.,changesinpreviouspurchaseamounts, economy)(CampbellandFrei2009).

(9)

TypeIITobitModel

TocomplementourDIDanalysis,weused aTypeIITobit specificationtoestimatetheeffectsofonlinechanneladoption onpurchaseincidenceandordersizeovertime.Weassumethat acustomerfirstdecideswhethertopurchasefromthefocalfirm, andthendecideshowmuchtospend(i.e.,ordersize)(Ansari, Mela,andNeslin2008).Inthistwo-stepmodelingapproach,we firstemployedabinomialprobitmodelwithrandomeffectsto determinewhetheracustomerpurchasesfromthefocalfirmin thecurrentmonth—adummyvariablerepresentedbyPit.Then, conditionalonapurchasefromthefocalfirminagivenmonth, wedesignedaregressionmodeltodeterminetheaverageorder size per transaction, denoted byQit inour model.Similar to the DIDanalysis, wetestedthe effects ofonline adoption on customerbehavioracrosssegments,usingthefollowingmodel specifications:

Pit=Purchase, ifPit∗>0; NoPurchase, ifPit∗≤0 (2) P_ii∗=β0+β1Postit+β2Postit×Treatedi+β3Genderi

+Xitγ+vit (3)

Qit=Q∗it, ifPit∗>0; unobserved, ifPit∗≤0 (4) Q∗_it=δ0+δ1Postit+δ2Postit×Treatedi+δ3Genderi

+Xitθ+μit (5)

whereP_it∗ referstothe latentutilitythat customeripurchases fromthefocalfirminmontht,andQ∗_it isthelatentutilityof theordersizefromthefocalfirminmontht.Inaddition,Postit isthekeyexplanatoryvariable,equalto1fortheperiodafter customeriadoptstheonline channelinmonthtandto0 oth-erwise.Itscoefficients(β1andδ1)captureanychangesinthe purchasesforthecontrolgroupinthepostadoptionperiod.The dummyvariableTreatediis1ifcustomeriistheonlineadopter and0 otherwise. The interaction betweenTreatedi andPostit measuresthe differenceintheresponsevariablesbetweenthe treatedandcontrolgroupafteradoption,thusrevealingtheeffect ofonlineadoptiononcustomerbehavior.Xitrepresentsavector of time-varyingcontrol variables,including age, severalstate dependentvariables,5purchasefromcompetitors,recency,and seasonality.Tomitigateseasonalinfluences, weadoptAnsari, Mela, andNeslin (2008) method:(1) selectmonthly dummy variablesthatsignificantlyaffectmonthlypurchasefrequency, then(2)combinethemonthdummieswhoseparametersdonot significantlydiffer.Wetherebyidentify fourseasonality indi-cators:March,August,April&May,andJune&October.We alsoconsiderthe effectof theeconomic climate.Inline with

5 _{State-dependent}_variables_are_lagged_variables_that_are_defined_differently

ineachequation.InEq.(3),whichexaminespurchaseincidence,the state-dependentvariablesaretwodummyvariablesthatindicatewhetheracustomer shoppedthroughtheonlinechanneloranofflinechannelinthelastmonth.In Eq.(4),whichestimatesordersize,itistheordersizeoflastpurchase.

0 10000 20000 30000 40000 50000 60000 1 2 3 4 5 6 7 8 9 10 BIC AIC AIC3

Fig.4.GraphsofModelSelectionCriterion.

theperiodsofeconomicrecession(Mostaghimi2004;Ohanian 2010),weobservedramaticdeclinesinthetotalnumberof trans-actionsbetween2001and2003andbetween2008and2010.A dummyvariable(Economicrecession)identifiestheseyears.We summarizethemeasurementsofabovevariablesinTable3.

EstimationResults ResultsofLatentClassClusterAnalysis

Model selection.WestartbypresentingtheLCCAresults, whichweestimatedbyapplyingsolutionswithdifferent num-bersofsegments.FollowingKonus,Verhoef,andNeslin(2008), our model selection procedure relies on a set of statistical criteria:theBayesianinformationcriterion(BIC),Akaike infor-mationcriterion(AIC),andAkaikeinformationcriterionwith apenaltyfactorof three(AIC3),togetherwiththe interpreta-tionofthederivedsegments.Amongthestatisticalcriteria,we primarilyrelyontheBIC,becauseitismoreeffectivefor deter-miningthecorrectnumberofsegmentsforLCCAthanareother criteria(VermuntandMagidson2005;Zhang2004).

Weprovide the graphsfor BIC,AIC,andAIC3 inFig.4. The valuesofthethreecriteriaare closeandkeepdecreasing with moresegments,but Fig.4 alsosuggests thatthe graphs oftheseindexesbecomeflatafterfivesegments.Theestimated results areconsistentacrossmodels withmorethanfour seg-ments;increasingthenumberofsegmentsmostlyenhancesthe complexityoftheinterpretation.Therefore,wechosethemodel with five segments, tobalancethe fit criteria andachieve an intuitiveinterpretation.

Modelproﬁle.Theresultsindicateaclearsplitofcustomer segments on the basis of their online adoption duration and purchase amount prior to online adoption. Table 4 contains descriptivestatisticsforeveryidentifiedsegment.Because aver-age adoption duration differs across the five segments—20 months,47months,72months,92months,and101months—we apply Rogers’s (2003)segmentation terminologyandreferto the identified segments as innovators (181 customers), early adopters(311customers),earlymajority(511customers),late majority(170customers),andlaggards(457customers), respec-tively.

(10)

Table4

Profilesofsegments.

Label Innovator(N=181,11.10%) Earlyadopter(N=311,19.08%) Earlymajority(N=511,31.35%)

M SD MIN MAX M SD MIN MAX M SD MIN MAX

Indicators

Adoptionduration 20 6 3 30 47 8 31 65 72 8 56 87

Yearlypurchaseamount beforeonlineadoption

165.66 142.76 13.26 853.33 133.64 116.33 10.21 732.44 100.46 60.32 1.02 287.56

ActiveCovariates

Age 44 8 21 68 44 10 20 76 43 9 20 70

Gender(male) 4.97% 5.79% 2.74%

InactiveCovariates Yearlypurchaseamountafter

onlineadoption

139.86 102.57 15.19 710.65 144.05 117.88 12.79 730.2 140.41 115 13.78 794.21

Onlineshoppingpreference 0.622 0.104 0.205 0.857 0.596 0.124 0.124 0.843 0.589 0.103 0.199 0.805

Label Latemajority(N=170,10.43%) Laggard(N=457,28.04%)

M SD MIN MAX M SD MIN MAX

Indicators

Adoptionduration 92 13 64 121 101 10 85 124

Yearlypurchaseamount beforeonlineadoption

342.81 159.53 21.03 1041.3 80.63 46.91 5.94 207.44

ActiveCovariates

Age 51 11 24 91 43 9 22 88

Gender(male) 1.76% 2.19%

InactiveCovariates Yearlypurchaseamountafter

onlineadoption

331.15 223.34 28.67 1333.7 144.45 135.59 17.14 1088.94

Onlineshoppingpreference 0.285 0.149 0.004 0.654 0.589 0.092 0.279 0.792

The link betweenadoption duration and purchaseamount beforeadoptiondiffersfromourexpectation:priortothe adop-tionofonlinechannels,thelatemajoritysegmentspent342.81 Eurosperyear,morethananyothersegments.Therefore,in con-trastwithH1,thesegmentexhibitingthemostintensiveshopping behaviorisnottheearlieradoptersbutratherthelatemajority. Theresults forotherfoursegments instead matchour expec-tations,suchthatinnovators(165.66Euros)andearlyadopters (133.64)spendmorethantheearlymajority(100.46Euros)or laggards(80.63Euros)priortotheiradoptionof online chan-nels.

Aftertheadoptionofonlinechannels,mostsegmentsmake morepurchases,thoughinnovatorsandthelatemajorityreduce theirspendingslightly,from165.66to139.86Eurosandfrom 342.81to331.15Eurosperyear,respectively.Withrespectto onlineshoppingpreferences,thelatemajoritysegmentexhibits the lowestpreference (.285) for shopping online, rather than thelaggards. The averageage of membersof the late major-ityisapproximately51 years,olderthanotherfoursegments whoseaverageagesrangebetween43and44years.The late majoritysegmentalso is least likelyto includemen (1.76%) comparedwiththeother segments.Forothersegments,there are greater proportions of men in earlier adopter segments compared withlater adoptersegments, which is inline with previous studies (e.g., Venkatesan, Kumar, and Ravishanker 2007).

Parameter estimation.Table5 containstheparameter esti-mationsfor theindicatorsandactivecovariatesintheLCCA. Two indicators are statistically significant (p<.001) in most segments,suggestingthattheyeffectivelyclusterthecustomer segments.

With respecttothe covariates,theWald testindicatesthat the age(p<.001) andgender (p<.05) coefficientsdiffer sig-nificantly across segments. Age has a positive effect on the probabilityofbeing inthe latemajority(.051,p<.001)buta negative effecton the likelihood of being in othersegments: earlyadopter(−.010,p<.05),earlymajority(−.017,p<.01), orlaggard(−.022,p<.001).Malecustomersaremorelikelyto beearlyadopters(.330,p<.05);however,genderdoesnotaffect membershipinothersegments.Theseresultsareconsistentwith findingsinTable4.

ResultsofPropensityScoreMatchingMethod

Ourmatchingtechniqueseekstolinkanonlineadopterina segmentwithanofflinecustomerwhohasasimilarpropensity toadopt the online channel, so that we canaccount for self-selection.Weused alogistic modeltocalculate acustomer’s propensity to adopt the online channel. In each segment, we chosethepredictorvariablesthatgeneratedthebestmodelfit, andinTable6,wepresenttheestimationsoftheparametersfor fivesegments.

(11)

Table5

Parameterestimationoflatentclassclustermodel.

Innovator Earlyadopter Earlymajority Latemajority Laggard Wald p-Value(Wald) Indicators

Adoptionduration −1.075*** −0.193*** 0.226*** 0.476*** 0.566*** 6144.474 0.000

Yearlypurchaseamountbeforeadoption 10.8 −24.2*** −56.6*** 146.0*** −75.9*** 284.686 0.000

ActiveCovariates Age −0.002 −0.010* _−0.017** _0.051*** _−0.022*** _44.2644 _0.000 Gender(male=1) 0.232 0.330* _−0.055 _−0.343 _−0.164 _9.5435 _0.049 * _Significant_at_.05. **_Significant_at_.01. ***_Significant_at_.001.

Asweexplainedinthemethodologysection,weemployeda commonsupportrestrictionandsetourcaliperto.01toestablish theminimumdifferenceallowedwithrespecttotheestimated propensitiesbetweenanonlineadopterandthematchedoffline customer.Thisrestrictionexcluded1(.55%)ofinnovators,29 (9.32%)ofearlyadopters,84(16.44%)oftheearlymajority,9 (5.29%)ofthelatemajority,and113(24.73%)oflaggards.

To qualify the performance of our matching procedure, we firstcheckedifthe differencesincustomercharacteristics remainedstatisticallysignificantaftermatching,usingat-test, thencomputed the percentage of bias reduction (Rosenbaum and Rubin1985) (see Table 7). The reduction in bias repre-sentsthe difference inthe mean of aparticular characteristic between two matchedgroups after matching, minusthe dif-ference before matching (Rosenbaum and Rubin 1985). The percentageof biasreductionwas substantialformost charac-teristics.Onlythemetricsofpreviousordersizewerenegative forcertainsegments,suggestingthatthetwogroupsbecameless comparableonthisfactoraftermatching.However,anincrease inbiasforthisvariablewouldnotaffectoverallmatching per-formance,becausethedifferencesbetweenonlineadoptersand matchedofflinecustomerswerenotsignificantforallcustomer characteristicsaftermatching.Therefore,thesampleswere com-parableaftermatching,andweeliminatedself-selectionbiasin relationtotheselectedcharacteristicswithourPSMmethod.

ResultsofDifference-in-DifferenceAnalysis

Table8 containstheresultsoftheDIDanalysis,whichare estimatesofthedifferencesinthemeanofpurchaseactivities, aggregatedintheone-yearperiodspriortoandaftertheadoption

of online channels. Fortheearly adopter,early majority,and laggardsegments,totalannualpurchaseamountsand frequen-cies significantly increaseafter online channel adoption, and thedifferencemetricsaresignificantlylargerthanthechanges for thematched offlinecustomers. Theseresults suggest that onlinechanneladoptionispositivelyassociatedwithpurchase volumesinthesesegments.Moreover,thechangesintheoffline purchaseamountsandfrequenciesafteronlinechanneladoption areessentiallynullinthesesegments,suggestingthatincreases incustomerspending derivefromadditionaldemand through thenewonlinechannel,ratherthansubstitutingforpurchasesin existingofflinechannels.However,thechangesinthepurchase volumesintheinnovatorandlatemajoritysegmentsdonot dif-fersignificantlyfromthevariationofpurchasesinthecontrol groups. Innovators increasetheir purchasessignificantly after adopting online(markedby* inTable8),buttheseincreased amounts do not significantlydiffer from thosein the control group.Furthermore,bothsegmentsreducetheirofflinepurchase amountsandfrequenciesafteradoptingonlinechannels.Thus, onlinechanneladoptionhasnoeffectonthepurchaseamountsor frequenciesofinnovatorsandthelatemajority.Thus,theresults oftheDIDanalysissupportH2,whichsuggeststhatonline chan-nel adoptionexertsmorepositiveeffectsonthepurchase vol-umesoflightshoppersegmentsthanofheavyshoppersegments.

ResultsofTypeIITobitModel

We employed the Type II Tobit model to investigate the effects of online channel adoption onmonthlypurchase inci-denceandordersizepertransactionacrossdifferentsegments over time. Customers from different segments adopt online Table6

Parameterestimationofpropensityscoremodel.

Innovator Earlyadopter Earlymajority Latemajority Laggard

Age −0.088 −0.067 −0.079 −0.008 −0.080

Age2 _−0.003 _0.000 _−0.001 _0.000 _−0.001

Gender(male=1) 0.664 0.667 0.056 −0.331 −0.061

Tenure −0.013 0.038 −0.017 −0.008 0.015

Previousmonthlypurchaseamount – – −0.085 – –

Previousmonthlypurchasefrequency −0.884 −2.702 – 3.463 −14.417

Previousordersizepertransaction −0.001 0.003 – 0.008 −0.002

Constant −1.301 −2.670 1.334 −1.987 −0.402

(12)

Table7

Significanceofdifferenceandreductioninbiasaftermatching.

Innovator Earlyadopter Earlymajority Latemajority Laggard

Significanceofdifference

Age 0.548 0.712 0.423 0.722 0.771

Gender 0.793 0.664 1.000 1.000 1.000

Tenure 0.557 0.717 0.520 0.843 0.991

Previousmonthlypurchaseamount – – 0.204 – –

Previousmonthlypurchasefrequency 0.432 0.869 – 0.647 0.500

Previousordersizepertransaction 0.689 0.356 – 0.185 0.680

Reductioninbias(%)

Age 92.8 96.1 94.4 38.8 97.8

Gender 72.3 74.9 100.0 100.0 100.0

Tenure 7.5 40.6 65.6 51.7 98.8

Previousmonthlypurchaseamount – – 91.9 – –

Previousmonthlypurchasefrequency 63.5 96.8 – 93.9 97.3

Previousordersizepertransaction −69.3 21.1 – 13.5 −196.6

channelsatdifferenttimes,sotheperiodpriortoandafteronline adoptionvariesgreatly.Forafaircomparisonacrosssegments, wetestedthemodelsusingthesamelengthoftime(oneyear) priortoandafteronlineadoption.Therefore,thetesteddata con-tain24-monthobservationsfor eachcustomer.Wepresentthe resultsofthepurchaseincidencemodelinTable9andtheorder sizemodelinTable10.Parameterestimatesoftheinteraction betweenthepostadoption periodandthetreatedgroupreveal significant and positive effects on purchase incidence (.157, p<.001)andordersize(11.776,p<.01)amongtheearly major-ity;thesecustomersincreasetheir monthlypurchasevolumes afteradoptingonlinechannels,relativetothecontrolgroup, con-sistentwiththeDIDanalysis.However,theinteractiveeffectsare notsignificantfortheothersegments,suggestingonline chan-neladoptionhasnoimpactonthemonthlypurchasesofthese segments.Thefindings relatedtoearlyadoptersandlaggards understandablydifferfromthoseintheDIDanalysisthatreveals positiveeffectsofonlineadoption,becausethechangesin pur-chasevolumesduetoonlinechannel adoptionlikelyaremore exaggeratedin aDIDanalysis than aTobit model. The DID analysismeasureschangesintheyearlypurchasevolumeafter adoption,whereastheTobitmodelevaluateschangesinmonthly purchasevolumesovertime.Theinsignificanteffectsofonline channeladoptionfortheinnovatorandlatemajoritysegments insteadareconsistentwiththeDIDanalysis,whichconfirmsour predictioninH2butiscontrarytoH3.Thesecombinedfindings indicatethatcustomerbehaviorisdrivenpredominantlybythe intrinsicbenefitsofonlineshopping.

Forthecontrolvariables,wefindthatpurchasesfromonline channelsinthepreviousmonthexertpositiveeffectsonpurchase incidence in most segments, with the exception of innova-tors.Offlinepurchasesinthepreviousmonthenhancepurchase probabilitiesamongtheearlyadopter(.110,p<.05)andearly majority(.190,p<.001) segments. The ordersize of the last transactionrelatespositivelytocurrentordersizeinallsegments. Purchasefromcompetitorsinthecurrentmonthaffectsthe pur-chase incidence of the early adopters(.430, p<.05) and the latemajority(.321,p<.01),suggestinghighercategorydemand. Ageonlyaffectsthepurchaseincidenceofearlyadopters(.005, p<.05)andwefindnosignificantgendereffects.Furthermore,

recencyhassignificant,negativeeffectsonpurchaseincidence inallsegments,whichmayreflectafeatureofthebeautyand healthy category, for which purchase frequency is relatively lowerthaninmostconsumergoodsindustries(Inman,Shankar, andFerraro2004).Inourstudy,customerspurchasefromthe firm twice per year on average. Because the average period betweenpurchasesislong,itmightbedifficultforcustomersto recallthefirmorbrandfromwhichtheyboughtpreviously,and theirpurchasepatternscouldbeinterruptedeasily.Therefore,the longerthetimesincetheirlastpurchase,thelesslikelycustomers maybe topurchase from thefocal firm.Economicrecession relatesnegativelytopurchaseprobability,butthiseffectisonly significantforearlyadopters(−.155,p<.01).Asforseasonality, wefindthatcustomersinallsegmentsincreasetheirpurchase frequenciesinMarch,butthisfactordoesnotaffecttheamount spentpertransaction.

RobustnessChecks

Severaladditionalanalysesenableustotesttherobustness oftheestimatedeffects.First,weexaminedtheeffectsofonline adoptiononpurchaseincidenceandordersizeinlongerperiods: twoyearspriortoandafteronlinechanneladoption(fouryears total)andthreeyearspriortoandafteronlinechanneladoption (six yearstotal). We repeatedthe DID andTobit II analyses but only for the early adopter and early majority segments; the dataperiods for the othersegments were tooshort either before(i.e.,innovator)orafter(i.e.,latemajorityandlaggard) theonlineadoptiondate(resultsinAppendixA).Althoughearly adopterspurchasemoreinthepostadoptionperiod,thevariation intheirpurchasefrequencyandordersizepertransactiondonot significantly differfromthe changes exhibitedby thecontrol groupineitherthefour-orsix-yeartimewindows.Consistent withourinitialanalysis,online channeladoptionsignificantly increasespurchaseincidenceandordersizeintheearly major-ity segment, relative to the control group, in the four-year period. In the six-year time window, the purchase incidence change isnotsignificantfor theearly majorityversuscontrol group of offline customers. However, the order size increase issignificantlylargerthanthatdisplayedbythecontrolgroup.

(13)

Table8 DIDanalysis.

Onlineadopters Offlinecustomers

Before After Change Before After Change

Innovator

Totalpurchaseamount(in Euros)

122.76 146.48 23.72* 122.48 135.54 13.07

(160.96) (183.11) (13.59) (176.52) (173.43) (221.79)

Totalpurchasefrequency 1.32 1.63 0.32* _1.31 _1.38 _0.07

(1.63) (1.85) (1.90) (1.87) (1.54) (2.11)

Offlinepurchaseamount (inEuros) 122.76 64.83 −57.93* (160.96) (116.39) (163.82) Offlinepurchase frequency 1.32 0.72 −0.60* (1.63) (1.12) (1.57) Earlyadopter

113.28 158.83 45.54*,# 140.70 153.54 12.85

(155.29) (217.52) (213.71) (176.64) (174.82) (195.51)

Totalpurchasefrequency 1.23 1.96 0.74*,# _1.58 _1.83 _0.25

(1.65) (2.75) (2.66) (1.86) (2.17) (2.14)

Offlinepurchaseamount (inEuros) 113.28 95.99 −17.29 (155.29) (161.08) (176.30) Offlinepurchase frequency 1.23 1.15 −0.08 (1.65) (2.02) (2.04) Earlymajority

94.82 183.54 88.72*,# _125.01 _126.73 _1.72

(133.37) (259.05) (254.96) (192.67) (184.99) (217.68)

(1.55) (2.98) (2.85) (2.24) (2.08) (2.31)

Offlinepurchaseamount (inEuros) 94.82 106.64 11.82 (133.37) (204.17) (215.39) Offlinepurchase frequency 1.13 1.14 0.01 (1.55) (2.31) (2.33) Latemajority

383.55 347.01 −36.53 246.75 255.52 8.78

(326.14) (308.74) (354.04) (292.69) (394.04) (406.89)

Totalpurchasefrequency 4.37 4.20 −0.16 2.88 2.89 0.02

(3.61) (3.78) (3.93) (3.28) (3.71) (3.58)

Offlinepurchaseamount (inEuros) 383.55 268.55 −115.00* (326.14) (284.85) (338.02) Offlinepurchase frequency 4.37 3.20 −1.16* (3.61) (3.39) (3.62) Laggard

71.22 133.26 62.04*,# _98.05 _101.38 _3.33

(124.30) (264.82) (276.16) (170.77) (157.32) (168.38)

(1.37) (2.56) (2.63) (1.82) (1.88) (1.96)

Offlinepurchaseamount (inEuros) 71.22 69.68 −1.54 (124.30) (153.80) (168.43) Offlinepurchase frequency 0.83 0.85 0.02 (1.37) (1.50) (1.60)

Notes:Thistableprovidesthemeans,withthestandarddeviationsinbrackets.

* _{Significantly}_different_from₀_at_least_at_the_10%_level.

# _The_change_in_the_variable_for_online_adopters_is_{significantly}_different_from_the_change_for_offline_customers_(control_group)_at_least_at_the_10%_level. Thus,across varioustest periods,the early majoritysegment

increasesitsmonthlypurchaseamountafteradoptingtheonline channel.

Second, we tested whether our models are sensitive to extremevalues byincluding the outliersthat we deleted pre-viously,thenrepeatingthe modeling process(detailed results areinAppendixB).Someminordifferencesarose,butthe esti-matedresultsof thefulldatasetareconsistentwithourmain findings.

Third,wecheckedtheresultsrelatedtothelatemajority seg-ment,becausethestandardizeddeviationsofpurchaseamounts beforeandafteronlineadoptionweremuchlargerthaninthe othergroups.Toeliminatetheinfluenceofextremevalues,we excludedthe5%customerswiththegreatestpurchaseamounts andthe5%customerswiththelowestpurchaseamountsbefore or after online adoption. With these two selection rules, we dropped15customersintotal,thenreplicatedtheanalyses.The resultsoftheDIDanalysisandTypeIITobitmodelbothsuggest

(14)

Table9

Purchaseincidencemodel(24months).

Variable Innovator Earlyadopter Earlymajority Latemajority Laggard

Postadoption 0.063 0.076 0.070* −0.084 0.087*

Postadoption×Treatedgroup −0.032 −0.032 0.157*** 0.021 0.041

Pastonlinepurchase 0.070 0.199** 0.142* 0.158* 0.321***

Pastofflinepurchase −0.009 0.110* 0.190*** 0.039 0.079

Purchasefromcompetitors 0.028 0.430* _0.232 _0.321** _0.177

Age 0.006 0.005* _−0.002 _0.005 _0.003 Gender −0.062 0.035 0.092 0.044 −0.158 Recency −0.016*** _−0.008*** _−0.006*** _−0.022*** _−0.009*** Economicrecession −0.131 −0.155** _−0.060 _−0.045 _−0.029 Seasonality1:March 0.303*** _0.298*** _0.291*** _0.331*** _0.230*** Seasonality2:August −0.070 −0.242*** _−0.136** _−0.090 _−0.032

Seasonality3:April&May 0.080 0.009 0.060 −0.026 0.048

Seasonality4:June&October 0.122* _−0.029 _0.036 _0.092 _0.083

Constant −1.464*** _−1.491*** _−1.372*** _−0.993*** _−1.610***

* _Significant_at_.05. ** _Significant_at_.01. ***_Significant_at_.001.

thatonlinechanneladoptionhasnoeffectoncustomerpurchase volumesinthissegment(seeAppendixC),whichconfirmsour previousfindings.

DiscussionandImplications

Wesegmented customers on the basis of online adoption durationandpurchase amounts beforethe adoption. Wealso investigatedtheeffectsofonlinechanneladoptiononcustomer purchasesacrossmultiplesegmentsovertime.Forthe discus-sionweaddressthetworesearchquestionsthatmotivatedour study.

DoEarlierAdoptersofaRetailer’sOnlineChannel PurchaseMore?

Briefly, no. Our results instead reveal that customers in the latemajority segmentpurchase morethan the other seg-ments,bothbeforeandaftertheyadoptthenewonlinechannel.

Previous literature describes later adopters as having less income,lowereducationlevels,andlessinvolvementinanewly adoptednewproductorservice(Mahajan,Muller,andSrivastava 1990; Prins and Verhoef 2007; Rogers 2003). Our research suggests additionalfeatures that differentiatethem from oth-ers.Customersinthislatemajoritysegmentexhibitthelowest online shoppingpreferenceandare morelikelytobewomen andolderthan thoseinothersegments. Yetthe latemajority still is themost valuable segment. Weexplainwhywith two sub-questions.

Whyareheavyshoppersthelatemajorityintheiradoptionof onlinechannels? Themultichannelenvironmenthelpsanswer thisquestion.Inthisstudy,thefirm’sexistingcustomers grad-uallyadoptedanewlyintroducedonlinechannel,so theyhad purchasedthroughtheretailer’sofflinechannels(catalog, tele-phone) prior toadopting the online channel.Heavy shoppers boughtwithhigherfrequencyandvolumethroughtheseoffline channels,whichmightsuggesttheyperceiveofflineshopping as more convenient than do other customers. Moreover,

Table10

Ordersizemodel(24months).

Variable Innovator Earlyadopter Earlymajority Latemajority Laggard

Postadoption 0.506 −2.343 5.981 −1.635 −7.500*

Postadoption×Treatedgroup −6.724 −2.889 11.776** _−0.449 _4.053

Lastordersize 0.169*** _0.103*** _0.129*** _0.163*** _0.176***

Age −0.335 0.310 −0.036 −0.294 0.021 Gender 0.509 7.372 −3.424 −2.964 −11.730 Recency 1.687 −0.165 −0.101 −0.286 0.026 Economicrecession 17.164 −0.656 −16.856*** −5.173 5.111 Seasonality1:March −20.220 9.875 0.369 12.462 −4.397 Seasonality2:August 12.320 −3.623 −8.751 −2.499 −1.774

Seasonality3:April&May 4.734 2.965 −5.244 −6.112 0.936

Seasonality4:June&October −7.833 3.577 −8.371* 1.699 61.264

Constant 241.083 −2.209 6.576 50.318 3.238

* _Significant_at_.05. ** _Significant_at_.01. ***_Significant_at_.001.

(15)

positive shopping experiences in a channel increase channel loyalty (Ansari, Mela, and Neslin 2008), especially if cus-tomersinitiatetheirpurchaseprocessthroughofflinechannels (Dholakia,Zhao, andDholakia2005; Gensler,Dekimpe,and Skiera2007).Thus,heavyshoppersmighttendtokeepshopping throughtheirpreferred,existing,offlinechannelsanddelaytheir adoption of anewonline channel. Yet theyare notlaggards, becausetheirfrequentinteractionswiththefirmquickentherate atwhichtheydeveloptrustinitandformtheirperceptionsofthe benefitsofthisfirm’sproductsorservices(Hinde1979;Morgan and Hunt 1994). Therefore,purchase frequencyshortens the timeneededtoadoptadditionalchannels(Venkatesan,Kumar, andRavishanker2007).Furthermore,customers’expenditures cultivatetheirfirmloyalty,whichalsospeedsuptheadoption process(DemoulinandZidda2009).Facingconflicting mech-anisms, these customers do not adopt immediately after the introductionoftheonlinechannel(duetochannelloyalty),but nordotheytakethelongesttimetostartonlineshopping(due totrustandfirmloyalty).Instead,theyadopttheonlinechannel inamiddle–lateperiod.

Whydoheavyshopperspurchaselessfromtheonlinechannel afteradoptingit?Customers’channelchoicesevolveovertime, astheylearnfromprevioususageexperiences(Konus,Neslin, andVerhoef2014;Valentini,Montaguti,andNeslin2011). Cus-tomersbecomeless responsivetomarketingandlesslikelyto movetonewchannelswhentheyknowmoreaboutthefirm’s establishedchannels (Valentini,Montaguti,andNeslin2011), whichmayexplainwhycustomersinthelatemajoritysegment make few online purchases after adopting the channel. They alreadymakemorepurchasesthroughexistingofflinechannels, sotheyaremoreknowledgeableaboutofflinechannelsandless responsivetomarketingeffortsthatencourageusesofthenew onlinechannel.Empiricalevidenceaffirmsthatheavyshoppers exhibitgreaterloyaltytosaleschannelsthanlightshoppersand arelesslikelytoswitchtodifferentchannels(Gensler,Dekimpe, andSkiera2007).Thus,ourstudyconfirmsthatitremains dif-ficulttomoveheavyshoppersfromexistingsaleschannelstoa newchannel,evenaftertheyadoptthisnewchannel.

Areearlieradoptersnotvaluable?Earlieradoptersremain valuable;however,theirpurchasevolumesarelowerthanthose ofthelatemajoritysegment.Innovatorsandearlyadopterswho adoptanewonlinechannelintheearlyperiodpurchasemore thanlateadopters (earlymajorityandlaggard), prior totheir adoptionofanewonlinechannel.Thesefindingsareconsistent withourexpectations andpreviouschannel adoptionresearch (Venkatesan,Kumar,andRavishanker2007;Xue,Hitt,andChen 2011).

HowDoesOnlineChannelAdoptionAffectPurchase VolumesAcrossSegments?

Theeffectsofonlinechanneladoptiononcustomerpurchases varyacrosssegments.Theseeffectsdifferparticularlybetween theheavyandlightshoppersegments.

• Heavyshoppers.Theheavyshoppersegmentsarethe innova-torsandlatemajority,whoaretheheaviestshoppingsegments

priortotheadoptionofthenewonlinechannel.Theironline channeladoptionhasnoeffectontheir purchases,interms ofmonthlypurchase incidence,ordersize,yearly purchase amount,or yearly purchase frequency. Customers inthese twosegmentssimplymoveaproportionoftheirdemandfrom existingofflinechannelstothenewonlinechannel.Therefore, thenewonline channelcannibalizespurchasesfromoffline channelsinthesesegments.

• Lightshoppers.Customersintheearlyadopter,earlymajority, andlaggardsegmentsincreasetheiryearlypurchaseamounts andfrequenciesafteradoptingonlinechannels(DID analy-sis),butonlytheearlymajoritysegmentincreasesitsmonthly purchaseincidenceandordersizeovertime(Type IITobit analysis).AccordingtotheDIDanalysis,customersinlight shoppersegmentstendtopurchasethesameamountoffline after adopting online channels, so the additional volumes derivemainlyfromsalesinthenewonlinechannel.

Overall, heavyshopper segments areless affectedbytheir adoption of online channels thanare light shopper segments. Although customer behavior can be driven by intrinsic ben-efits and by marketing communications (Ansari, Mela, and Neslin2008;Neslinetal.2006),ourfindingssuggestthatthe benefits of online shopping represent the predominant influ-ence oncustomerpurchasesafter theyadopt onlinechannels. Heavy shoppersestablish strongerpurchasinghabitsin exist-ingofflinechannelsthanlightshoppers(Aarts,Verplanken,and Knippenberg 1998;Orbellet al.2001),so theymayperceive fewerbenefitsfromonlineshoppingthandolightshoppers(Falk etal.2007).Asaresult,thesecustomersviewtheonlinechannel asasimpleextensionofdistributionchannels,whichdoesnot affecttheiroverallshoppingdemand.Incontrast,lightshoppers perceivemorebenefitsfromonlineshoppingandconsiderthe newonlinechannelanadditionalbenefit,beyondoffline chan-nels.Theyrewardthefirmforthisextrabenefitbyincreasing their spending,mostlycoming from the online channel. Fur-thermore,customers’share-of-walletmightalsoaffectpurchase volumesbetweenlightandheavyshoppersegmentsaftertheir adoption.6Heavyshoppersaremorelikelytogiveahigh share-of-wallettothecompany.Sincecustomersonlyneedacertain amount of groceries, it is more difficult to gain extra sales from heavy buyers thanlight shoppers after online adoption. Liu(2007)supportsthisargumentandprovesthatlightbuyers purchasemorefrequentlyandbecomemoreloyaltofirmsafter adoptionofaloyaltyprogram,whereasthespendinglevelsand loyaltyofheavyshoppersdonotchangeovertime.

ManagerialImplications

Becausetheeffectofonlinechanneladoptionvariesacross segments,retailmanagersshoulddifferentiatetheirstrategiesto appealtotwospecificgroups:acombinationofearlyadopter, earlymajority,andlaggardsegments,andthenacombinationof innovatorsandlatemajority.Intheformergroup,customersare

(16)

moreresponsivetotheonlinechannelandincreasetheiroverall purchase volumes through the new channel, without reduc-ingpurchasevolumes inofflinechannels.Therefore,retailers shouldfocusonstimulatingtheironlineshoppingvolumes.For example,theycouldincreasethefrequenciesoffirm–customer interactionsthatpromoteonlinespendingbythesecustomers. Furthermore,retailersshouldactivelyworktoswitchmost pur-chasesbythesecustomerstothecost-savingonlinechannel,to reduceoverallservicecosts.Incontrast,forthelattergroupof customers(innovatorsandlatemajority),theoverallpurchases donotincreaseafteradoption.Instead,theyreplacetheiroffline purchases withonline purchases. Theseheavy customersare likelyhabitualshoppersintheretailer’sexistingoffline chan-nelsandare less likelytoview online shoppingas abenefit. Thus,insteadofpushingthemtoshoponline(i.e.,bysending moreadvertisements),retailersshouldworkonfacilitatingtheir perceptionsofthebenefitsofonlineshopping,suchasby pro-motingits convenienceor emphasizingitsother benefits.Yet retailerscannottaketheriskofignoringtheirprofitablecontacts withthesecustomersthroughexistingofflinechannels. LimitationsandFurtherResearchDirections

Thisresearchhasseverallimitationsthat provideideasfor futureresearch.First, welacked informationonthe retailer’s marketingactivitiesinthethreechannels.Therefore,wedetect theeffectsofmarketingoncustomerpurchasesonlyindirectly. Additionalresearchmayinvestigatetheeffectsofmultichannel communicationsoncustomers’behaviorsacrosssegmentsthat adoptonlineindifferentperiods.Second,wefocused on pur-chaseamountsandfrequenciesratherthanprofitability,because wecannotaccessunitproductcostsorservicecosts.Retailers use customer profitability as a key metric for evaluating the monetaryvalueoftheirindividualcustomers,sofurtherresearch couldexplorecustomerprofitabilityacrosssegments.Third,we didnotdistinguishdifferentproductcategoriesortypes,dueto datalimitations.Ascustomermultichannelshoppingbehavior andtheeffectsofonlinechanneladoptiondifferacrossproduct types(Gensler,Leeflang,andSkiera2012;Konus,Verhoef,and Neslin2008),futurestudiescouldreplicateourresearchinother categories. Fourth, our data set did not contain information aboutattitudinalorpsychographicfeaturesandofferedlimited demographic information. Including more such information couldhelpfirmsidentify andcharacterizecustomersegments. Therefore,additionalresearchshouldincludemorecovariates thatreflectcustomers’characteristicsandtheirattitudesabout multiple channels. Last but not the least, the study period spansthetimeframewhentheInternet channelbecamemore sophisticatedovertime,whichcouldhaveplayedaroleinthe observedresults.Ofcourse,thisisacharacteristicofallnewly emerging channels and thus the results can provide useful generalizations.

Executivesummary

Marketers have long held the notion that earlier adopters ofaproductorservice aremorevaluablethanlateradopters,

because theynormally generatemoreprofit andare likelyto influencetheshoppingdecisionoflateadopters.However,are earlieradoptersalsocriticaltothesuccessofanewlyintroduced marketing channel? Driven by the Internet and mobile tech-nology,retailersincreasinglyintroducenewonlinechannelsto provideinformation,sellproducts,orofferservices.Aretailer’s existingofflinecustomersadoptthenewonlinechannelat dif-ferenttimeperiodsandmaypurchasethroughmultiplechannels eventually.However,itisstillunclearabouttherevenue contri-butionsorbehaviorsofcustomergroupsthatadopttheretailer’s newchannel earlier or laterthanother customers. For exam-ple, do innovators or early adopters of new online channels purchasemorethanthemajoritysegmentsorlaggards?Ifitis true,willtheycontinuetopurchasemoreafteradoptingthenew channel?

Motivated by above questions, we segment customers on the basis of their online channel adoption duration and pur-chase amounts before adoption, and investigates the effects of their online channel adoption on purchasevolumes across segments. The datacover 12.5yearsof purchase historyand individualtransactionsatalargeFrenchretailerthatsells nat-ural health products through catalog, telephone and online channels. We employ a series of models to solve our prob-lems, including latent clusterclass analysis,propensity score matching, difference-in-difference analysis andType II Tobit model.

MainFindings

Ourresearchrevealsplentyofinterestingfindings.Contrary toconventionalwisdom,itisnotinnovatorsorearlyadopters, butratherthelatemajoritysegmentthatpurchasesmorethan theothersegments,bothbeforeandafteronlinechannel adop-tion.Inaddition,wefindtheeffectsofonlinechanneladoption onpurchase volumesacross differentsegments. Online chan-nel adoption does not influence purchase volumes of heavy shoppersegments(latemajorityandinnovators),whereaslight shopper segments(earlyadopter, earlymajority,andlaggard) tendtoincreasetheir purchasesafteradoptingthisnew chan-nel.

Implications

Thisstudythusoffersimplicationsonidentifyingheavy shop-persandhowmanagersinfluencecustomershoppingbehaviors acrossdifferentsegments.Totargetheavyshoppers,managers shouldnotsolelyfocusonearliestadopters(e.g., innovators), butmoreimportantly,theyshouldconsiderthecustomerswho adopt during the middle-late period (late majority). In addi-tion,marketershoulddifferentiatetheirstrategiestoappealto two specific adopter groups:a combination of early adopter, earlymajority,andlaggardsegments,andthenacombination of innovators and late majority. In the former group, retail managers should focus on stimulating their online shopping volumes.Forexample,theycouldincreasethefrequenciesof firm–customer interactions that promote online spending by thesecustomers.Forthelattergroupof customers,insteadof pushingthemhardtoshoponline,retailersshouldworkon facil-itatingtheirperceptionsofthebenefitsofonlineshopping,such as by promoting its high qualityor emphasizing its benefits. Yet retailers cannot take the risk of ignoring their profitable