Personalized Product Recommendations : Evidence from the Field

P oceedings o he 50 h Hawaii In e na ional Con e ence on Sys em Sciences | 2017 Pe sonalized P oduc Recommenda ions: E idence om he Field Essi Pöy y Aal o Uni e si y essi.poy y@ aal o. i Ninni Hie aniemi Aal o Uni e si y ninni.hie aniemi@ aal o. i Pe i Pa inen Uni e si y o Helsinki pe i.pa inen@ helsinki. i Juho Hama i Uni e si y o Tampe e juho.hama i@ s a .u a. i Mau i s Kap ein Tilbu g Uni e si y m.c.kap ein@ ilbu guni e si y.edu Abs ac Ta ge ing pe sonalized p oduc ecommenda ions o indi idual cus ome s has become a mains eam ac i i y in online s o es as i has been shown o inc ease click- h ough a e and sales. Howe e , as pe sonaliza ion becomes inc easingly commonplace, cus ome s may eel pe sonalized con en in usi e and he e o e no esponding o e en a oiding hem. Many s udies ha e in es iga ed ad e ising in usi eness and a oidance bu a esea ch gap on he e ec o deg ee o pe sonaliza ion on cus ome esponses based on ield e idence exis s. In his pape , 27,175 ecommenda ion displays om i e di e en online s o es a e analyzed. The esul s show ha he u he he cus ome is in he pu chasing p ocess, he mo e e ec i e pe sonaliza ion is i i is based on in o ma ion abou he p esen a he han pas b owsing session. Mo eo e , ecommenda ions in passi e o m a e mo e e ec i e han ecommenda ions in ac i e o m sugges ing he need o dispel he pe cep ion o in usi eness. 1. In oduc ion Pe sonalized elemen s ha e become an essen ial pa o online s o es. Elemen s ha can be pe sonalized include o example welcome messages, s o e layou , sales a gumen s o p oduc ecommenda ions [5, 6, 14, 15]. The aim o pe sonaliza ion is o inc ease sales h ough mo e pe suasi e, sui ed and ele an con en , and, in gene al, pe sonaliza ion has been shown o inc ease click- h ough a es [e.g., 21] and sales [e.g., 14, 15]. Howe e , pe sonalized ad e isemen s may esul in ad e isemen eac ance and ul ima ely a oidance among consume s because a ge ed ecommenda ions may be pe cei ed as oo in usi e [4, 11]. Ad e ising li e a u e has ex ensi ely s udied ad e isemen in usi eness [e.g., 12, 18] bu esea ch on in usi eness o pe sonalized online con en wi h ield da a is limi ed. Whi e and colleagues [30] and an Doo n and Hoeks a [28] ha e s udied he deg ee o pe sonaliza ion in e-mail messages and online ad e isemen s bu bo h s udies u ilize hypo he ical scena io-based da a. Field da a is pa icula ly aluable because in ac ual pu chase si ua ions consume s may URI: h p://hdl.handle.ne /10125/41626 ISBN: 978-0-9981331-0-2 CC-BY-NC-ND no always ecognize ha some con en is pe sonalized. In labo a o y expe imen s pe sonalized elemen s a e usually highligh ed wi h a ious cues and he e o e c ea e a nega i e mindse . The aim o his s udy is o p o ide an unde s anding on how he deg ee o pe sonaliza ion in p oduc ecommenda ions a ec s consume esponses in di e en s ages o buying. In addi ion, he pu pose is o measu e pe cei ed in usi eness o pe sonalized p oduc ecommenda ions. The s udy is conduc ed by using da a om i e di e en online s o es and en di e en pe sonalized p oduc ecommenda ion ad e isemen s. The decisions ega ding he ad e isemen s – hei wo ding, hei placemen , he base o hei pe sonaliza ion – we e na u al in a sense ha he esea che s did no ha e any in luence o e hem. Thus, he p esen pape con ibu es o he li e a u e on pe sonaliza ion by showing how he deg ee o pe sonaliza ion a ec s consume s’ ac ual clicking beha io in online shopping con ex . 2. Theo e ical backg ound 2.1. Online pe sonaliza ion and ecommende sys ems Cus ome in o ma ion can be used o ailo p oduc s, se ices and consump ion expe iences o i he speci ic needs and as es o cus ome s [6]. Kap ein and Pa inen [15] de ine online pe sonaliza ion as he ac o speci ically selec ing con en o indi idual cus ome s based on p ope ies o he cus ome wi h he goal o inc easing business ou comes o an e- comme ce pla o m. In p ac ice, his equi es iden i ying he cus ome , ga he ing in o ma ion abou him o he and p ocessing da a o p o ide ecommenda ions [6]. Acco ding o Chellappa and Sin [6], a ailabili y o po en ial cus ome in o ma ion is la gely a ec ed by how willing cus ome s a e o sha e hei pe sonal in o ma ion and use pe sonalized se ices. In online s o es, he e a e di e en hings ha can be pe sonalized. Lee and Pa k [17] iden i y h ee a eas o pe sonaliza ion: o e , ecogni ion and pe sonal ad ice. O e includes op ions o pe sonalizing wish lis s as well as pe sonalized ewa ds and p omo ion eminde s. Recogni ion s ands o using he cus ome ’s name, and p o iding op ions o sa e 3859 Published in P oceedings o he 50 h Hawaii In e na ional Con e ence on Sys em Sciences (HICSS 2017) (pp. 3859-3867). Uni e si y o Hawai´i a Manoa. ISBN: 978-0-9981331-0-2. h p://dx.doi.o g/10.24251/HICSS.2017.467. 3860 pe sonal and inancial in o ma ion. Pe sonal ad ice consis s o pe sonalized shopping and sea ch ea u es. On he o he hand, pe sonaliza ion can be based on a a ie y o ac o s. Van Doo n and Hoeks a [28] sugges ha online con en can be pe sonalized based on b owsing da a, pe sonal da a, and/o ansac ion da a. Online pe sonaliza ion can be a gued o inc ease in o ma ion sea ch p ocess e iciency because i aids cus ome s in making decisions and p e en s in o ma ion o e load. [3, 6]. As a consequence, pe sonaliza ion can lead o inc eased sales [3, 21]. Fo example, Pos ma and B okke [21] showed ha pe sonalized e-mail messages gene a e highe click- h ough a es han non-pe sonalized messages. Pe sonalized p oduc ecommenda ions o m one ca ego y o online pe sonaliza ion. Recommende sys ems gene a e ecommenda ions based on cus ome s’ b owsing his o y and p e iously de eloped da a sou ces [5]. The sys ems a e applied o help cus ome s in making pu chase decisions and p e en in o ma ion o e load by ma ching he cus ome ’s needs and p e e ences wi h sui able p oduc ecommenda ions [1, 22]. The e o e, ecommende sys ems o en succeed in in luencing he choices consume s make [13]. As in he case o gene al online pe sonaliza ion, he e a e a ious ways how p oduc ecommenda ions a e gene a ed. Typically, ecommenda ions a e made based on cus ome s’ exp essed p e e ences, pe sonal in o ma ion o pas beha io [2, 5, 27]. In p ac ice, his would mean o example sugges ions on wha o buy based on al eady selec ed p oduc s o on wha o he consume s exp essing simila needs ha e bough . Scha e e al. [23] p opose ha he e a e ou di e en o ms o ecommenda ions: Sugges ing p oduc s o cus ome s, p o iding pe sonalized p oduc in o ma ion, summa izing communi y opinion, and p o iding communi y c i iques. Cheung e al. [8] ca ego ize ecommende sys ems in o con en -based and collabo a i e sys ems based on he echnology ha is used. Con en -based ecommenda ions a e made based on he in e es s and p e e ences o a consume wi hou aking in o ma ion collec ed on o he consume s in o conside a ion. Collabo a i e ecommenda ions a e based on he p e e ences o o he simila consume s. 2.2. Ad e ising in usi eness, eac ance and a oidance When discussing ad e ising, sales p omo ions o o he pe suasi e communica ions, cus ome ’s pe spec i e should also be conside ed, and some imes cus ome s dislike he communica ion hey a e a ge ed wi h. Thus, ad e ising is some imes pe cei ed as in usi e. Li and colleagues [18] de ine in usi eness as “a pe cep ion o psychological consequence ha occu s when an audience’s cogni i e p ocesses a e in e up ed”. In he ad e ising con ex , ad e isemen s can be conside ed in usi e when a pe son pe cei es hem as in e up ing his o he goals. A ypical emo ional consequence o ad e isemen in usi eness is i i a ion [19]. Typical causes o inc eased in usi eness and i i a ion a e loud and dis u bing ad e isemen s o ad e isemen s ha a e placed in a dis ac ing way [18]. E-mail ma ke ing and pop-up ad e isemen s a e equen examples o in usi e online ad e ising [4, 12]. In beha io al e ms, ad e ising in usi eness can cause consume s o eac nega i ely o he ad e isemen and s a a oiding i . Acco ding o Edwa ds and colleagues [12], heo y o eac ance desc ibes he e ec he loss o eedom o a h ea ened loss o eedom has on people. I sugges s ha when aced wi h a h ea o losing eedom, eac ance c ea es a mo i a ional s a e in an indi idual o e- gaining eedom. Reac ance beha io has also been obse ed in he case pe sonalized online ad e ising [28, 30]. Ad e ising a oidance, on he o he hand, is de ined as he ac ions o media use s o in en ionally educing exposu e o ad e isemen s [26]. The e a e di e en ways ha consume s use o a oid ad e isemen s. Tele ision comme cials ha e been a popula subjec o s udy, and Clancey [10] sugges s ha he e a e h ee ways o a oiding ele ision comme cials: cogni i e a oidance (igno ing he ad), physical a oidance (lea ing oom) and mechanical a oidance (swi ching channel). These ways can also be applied o online ad e ising: igno ing he ad, closing b owse , and using p og ams ha block online ad e isemen s, such as AdBlock. Cho [9] a gues ha ad e ising a oidance in he In e ne is a esul o p e ious nega i e expe iences, pe cei ed hind ance o achie ing a goal and pe cei ed clu e o ads. A mo e ecen s udy by Baek and Mo imo o [4] sugges ha he e a e h ee de e minan s o ad e isemen a oidance: p i acy conce ns, ad e isemen i i a ion and pe cei ed pe sonaliza ion. P i acy conce ns and ad i i a ion inc ease ad e isemen a oidance whe eas inc eased pe sonaliza ion was ound o dec ease a oidance. In addi ion, p i acy conce ns a e an ex ensi e conce n among consume s as companies use hei pe sonal in o ma ion when p o iding pe sonalized online se ices [6, 27]. Pe sonalized messages may c ea e eac ance i indi iduals pe cei e hem as oo pe sonal and eel ha hey do no ha e con ol o e how hei pe sonal in o ma ion is used [4]. 3. Resea ch model and hypo heses Based on he li e a u e e iew, i is clea ha he e exis s a ade-o be ween pe sonaliza ion o online con en and eelings o i i a ion ha a e due o pe cei ed ad e isemen in usi eness. Baek and Mo imo o [4] ound ha inc eased pe sonaliza ion can dec ease ad e isemen a oidance, while Van 3861 Doo n and Hoeks a [28] ound ha highe deg ees o pe sonaliza ion inc ease pe cei ed in usi eness, which in u n a ec s buying in en ions nega i ely. Whi e and colleagues [30], on he o he hand, showed ha high deg ees o pe sonaliza ion in e-mail messages esul s in eac ance. The esul s sugges ha jus i ica ion and pe cei ed u ili y a e ac o s ha dec ease eac ance. Howe e , p e ious li e a u e has no conside ed he e ec he s age o buying migh ha e on he e ec i eness o pe sonalized online con en , o he basis on which he con en has been pe sonalized. These a e ypical a ian s in he ealm o online s o es, and mo e o en han no , hey a e no explici ly ecognized by consume s. This is a no able di e ence o p e ious esea ch ha o en uses ecipien names as one pe sonaliza ion aspec [e.g., 28]. Howe e , esea ch has no conside ed he e ec o o m o he messages has – a e consume s add essed di ec ly using ac i e o m o indi ec ly using passi e o m. Nex , we cons uc hypo heses based on hese a iables. 3.1. S age o buying Li e a u e on online pe sonaliza ion is limi ed in e ms o he e ec he s age o a cus ome ’s pu chase p ocess has on he e ec i eness o he ecommenda ions. In sales li e a u e, he poin a which a sales call is made has been seen o a ec cus ome esponse [e.g., 20, 25]. Simila ly, we belie e ha cus ome eac ions on pe sonalized p oduc ecommenda ions in online s o es a y in e ms o he s age o buying p ocess; in he beginning, a cus ome migh ha e a p oduc in mind ha he o she wan s o ind and is less esponsi e o he selle ’s ecommenda ions. La e , howe e , he immedia e need o isi he s o e has mo e likely been ul illed (e.g., ind in o ma ion abou a speci ic p oduc [11]) and he cus ome is mo e open owa ds he selle ’s sugges ions. Thus, we make a dis inc ion be ween p oduc ecommenda ions shown on he on page o an online s o e and p oduc ecommenda ions shown on pages u he in he shopping p ocess, such as ca ego y, p oduc and pu chase pages, and hypo hesize he ollowing: H1: Recommenda ions on he on page gene a e ewe clicks han ecommenda ions on la e pages. 3.2. Message o m Wa al and colleagues [29] dis inc be ween implici and explici pe sonaliza ion. The dis inc ion can be also e e ed o as passi e and ac i e message o m. A ecommenda ion using ac i e o m speaks o he cus ome explici ly by using wo dings such as “we ecommend o you”. Passi e o m e e s o ecommenda ions such as “o he s who iewed his also bough ” o “ he mos popula igh now”. Passi e o m is also o en used when ecommenda ions a e made by he company such as “picks o he day”. In p ac ice, ecommenda ions in passi e o m a e ypically based on in o ma ion on o he use s and ecommenda ions in ac i e o m on in o ma ion on he cu en use . Howe e , i is no necessa ily so, and ecommenda ions in passi e o m can be based on in o ma ion on he cu en use , and ice e sa. The assump ion on he basis o he ecommenda ion is ne e heless easily made by a consume based on he o m o he ecommenda ion. Ac i e message o m ep esen s p oduc ecommenda ions ha imply ha he ecommenda ions a e made speci ically o he cus ome . A message in passi e o m may no seem pe sonalized and does no imply ha he ecommenda ion is a sugges ion o a pa icula cus ome . Thus, ac i e message o m ep esen a highe le el o pe sonaliza ion in he eyes o he cus ome . As esea ch shows ha using he cus ome ’s name in pe sonalized ad e isemen s inc eases pe cei ed in usi eness and he eby dec eases pu chase in en ions [28, 29], we hypo hesize ha consume s espond be e o ecommenda ions in passi e a he han ac i e o m: H2: Recommenda ions in passi e o m gene a e mo e clicks han ecommenda ions in ac i e o m. 3.3. In e ac ion o s age o buying and message o m Whi e and colleagues [28] show ha click- h ough in en ions a e lowe o pe sonalized messages ha use explici cus ome da a and when he i be ween he ad e isemen and he cus ome need is low. P io esea ch also sugges s ha e-mail ad e isemen s ha do no men ion he use o cus ome in o ma ion a e pe cei ed as mo e a ac i e, while cus ome s eac nega i ely o ad e isemen s ha explici ly use pe sonal in o ma ion, such as ones name in a pe sonalized g ee ing [28, 29]. Acco ding o Wa al and colleagues [29], he nega i e eac ion is mos ly due o he conce ns o he sou ces and uses o pe sonal in o ma ion. Also Baek and Mo imo o [4] ha e shown ha oo explici ly pe sonal messages a e easily pe cei ed nega i ely by consume s. Mos consume s a e o en awa e ha p omo ions and o e s made by ma ke e s come wi h an agenda [7]. Mo eo e , p oduc ecommenda ions ha cus ome s pe cei e as i hey ha e been made o i hei needs by a company a e less a ac i e han p oduc ecommenda ions ha i hei p e e ences wi hou he company’s meaning [24]. Sela e al. [24] u he p opose ha elling consume s ha an o e is ailo ed o hem can lowe he deg ee o which consume s pe cei e he o e s as ba gains. The esea che s explain he inding by he idea o a compe i i e ela ionship be ween consume s and ma ke e s, acco ding o which a gain o ei he side is 3862 hough o come a he expense o he o he side. Thus, in his s udy, i is p oposed ha p oduc ecommenda ions using a passi e a he han ac i e message o m a e mo e e ec i e pa icula ly in la e s ages o a buying p ocess. This is because in he la e s ages he cus ome becomes mo e awa e o he selle ’s in en o pe suade he cus ome o buy. Thus, we hypo hesize: H3: Recommenda ions on la e pages gene a e mo e clicks i hey a e in passi e o m a he han ac i e o m. 3.4. In e ac ion o s age o buying and base o pe sonaliza ion In his s udy, base o pe sonaliza ion desc ibes wha in o ma ion has been used in making a p oduc ecommenda ion. Pe sonaliza ion can be based on he p esen b owsing session, pas b owsing session o i can be a andom p oduc ecommenda ion. Pas session-based p oduc ecommenda ions a e used when an online s o e has acqui ed b owsing in o ma ion om a cus ome ’s p e ious isi , and uses his da a in making a p oduc ecommenda ion he nex ime he same cus ome isi s he s o e. P esen session-based and andom p oduc ecommenda ions do no use p e iously acqui ed cus ome da a. Random p oduc ecommenda ions a e i ems selec ed by he company, and hey can be o example campaign p oduc s o he s o e’s mos popula p oduc s. P esen session-based p oduc ecommenda ions, on he o he hand, a e ecommenda ions ha a e ypically shown a e he on page and hey a e based on he cus ome ’s cu en shopping isi . These can be p oduc ecommenda ions shown o a cus ome based on an i em he cus ome is cu en ly iewing. Van Doo n and Hoeks a [28] an icipa e ha in usi eness is in luenced by he deg ee o pe sonaliza ion, and using only b owsing da a is conside ed mo e accep able han using ansac ion o o he pe sonal da a. This s udy is based on p oduc ecommenda ions ha use only b owsing da a, hus he deg ee o pe sonaliza ion is de e mined based on whe he he p oduc ecommenda ion uses his o ical b owsing da a o no . Some esea ch shows ha pe sonalized ecommenda ions based on p e ious pu chases a e pe cei ed as aluable and inc ease cus ome e en ion [1]. Howe e , as Van Doo n and Hoeks a [28] ound ha high i be ween a pe sonalized ad e isemen and a cus ome need inc eases pu chase in en ions and dec eases he nega i e e ec pe cei ed in usi eness, i is p oposed ha pe sonaliza ion ha is based on one’s p esen b owsing session c ea es a be e i be ween he ecommenda ion and he need. Simila ly, Li and colleagues [18] ound ha use ul and in o ma i e ad e isemen s a e conside ed less i i a ing and he e o e less likely o be a oided, which is p oposed o he be case in ecommenda ions based on p esen a he han pas b owsing session. E en hough ad e isemen s wi h a high i wi h cus ome needs p o ide ele an in o ma ion and he e o e usually inc ease pu chase in en ions, a high i may also inc ease pe cei ed in usi eness, and hus, pa icula ly high i can also educe he posi i e e ec o he i because i e eals o he cus ome ha pe sonal in o ma ion has been used [28]. On he o he hand, Ki e z and Simonson [16] show ha cus ome s pe cei e o e s ha i hei own needs and p e e ences as mo e aluable han o e s ha i he needs o o he cus ome s be e . Also, Whi e and colleagues [30] a gue ha jus i ied p oduc ecommenda ions inc ease pu chase in en ions, bu i he ecommenda ions a e no jus i ied, hey may lead o eac ance. We belie e ha cus ome s pe cei e p esen session-based p oduc ecommenda ions as mo e jus i ied han pas session-based p oduc ecommenda ions because hey a e mo e i ed o hei cu en need. Based on hese conside a ions, we p opose ha p oduc ecommenda ions based on a cus ome ’s cu en ac i i y ha e a highe i han p oduc ecommenda ions based on a cus ome ’s pas ac i i y. Fu he , we assume ha p oduc ecommenda ions based on a cus ome ’s p e ious ac i i y ha e a highe i han andomly chosen p oduc ecommenda ions. We he e o e hypo hesize: H4: Recommenda ions on he on page gene a e mo e clicks i hey a e based on he cus ome ’s pas isi a he han i hey a e chosen a andom. H5: Recommenda ions on he la e pages gene a e mo e clicks i hey a e based on he cus ome ’s cu en isi a he han pas isi . 4. Me hodology 4.1. Da a The esea ch da a was collec ed om i e di e en online s o es anging om June 2015 o June 2016, and i consis s o a o al o 27,175 ue displays o p oduc ecommenda ions. Fou o he online s o es ope a e in Finland, and one in he Uni ed Kingdom. The ypes o he s o es we e gene al supe ma ke , ou doo appa el and clo hing s o e, consume elec onics s o e, icke agen , and child en’s wea s o e. The da a was acqui ed om a company ha p o ides a so wa e o pe sonalize websi es, and he online s o es included in he analysis we e clien s o he company. Ten di e en ypes o p oduc ecommenda ions we e included in he da a, and hey we e ca ego ized based on hei message o m (ac i e, passi e), base o pe sonaliza ion (p esen session, pas session, andom) and page ( on page, u he pages). The 3863 ac ual p oduc s ha we e ecommended a ied be ween indi idual use s. Table 1 p esen s he di e en p oduc ecommenda ions. 4.2. P e- es A p e- es was conduc ed o in es iga e he pe cei ed in usi eness o he di e en ypes o p oduc ecommenda ions. 159 uni e si y s uden s pa icipa ed in a 3 (base o pe sonaliza ion – p esen session, pas session, andom) x 2 (message o m – passi e, ac i e) be ween-subjec s ac o ial expe imen . Based on a andom selec ion, esponden s we e sen an online su ey ha included a pic u e o an online s o e layou and one o he s udied p oduc ecommenda ion ype. Also, he e was a ex abo e he pic u e, which in oduced a scena io o a pu pose o isi he s o e. We used he look and eel o he hype ma ke ’s online s o e ha was included in he main s udy and he p oduc s in he ecommenda ions we e kep cons an ( ablewa e). The ques ionnai e consis ed o claims ega ding pe cei ed in usi eness [18], deg ee o in e es , loss o p i acy [4], and p obabili y o click. A se en-poin Like - ype scale was used in he ques ionnai e o all o he i ems anging om “s ongly disag ee” o “s ongly ag ee”. 59% o he esponden s we e male, and a e age age was 22 yea s. Nei he age no sex explained a iance o pe cei ed in usi eness. Mean sco e o pe cei ed in usi eness (measu ed on i ems “This ad e isemen is o ced”, “…is dis ac ing” and “… is in usi e”) was 3.67. An ANOVA es e eals ha bo h base o pe sonaliza ion (F = 6.256, p < .01) and message o m (F = 3.017, p < .10) had an e ec on pe cei ed in usi eness, bu no in e ac ion e ec eme ged (F = .119, p = .888). The lowes mean sco e appea ed in he ad e isemen ha ecommended p oduc s based on he cus ome ’s cu en b owsing session and s a ed in passi e o m “o he s who iewed his, iewed also” (M = 2.89, SD = 1.45, N = 28). The highes le el o in usi eness was pe cei ed in he ad e isemen ha was based on pas b owsing session and s a ed in ac i e o m “we ecommend o you” (M = 4.23, SD = 1.38, N = 27). Figu e 1 p esen s he mean sco es o he di e en ea men g oups. Figu e 1. Pe cei ed in usi eness o pe sonalized ecommenda ions Nex , esul s o he analysis o he esea ch da a is p esen ed. 4.3. Resul s To analyze he e ec o he esea ch a iables on consume s’ ac ual clicking beha io , we conduc ed chi-squa e es s and logis ic eg ession analyses. An analysis on he e ec o ac i e and passi e message o m on click- h ough a es was conduc ed. O he o al messages shown on on page, 4,521 we e passi e and 9,149 we e ac i e. 13.5% o he ecommenda ions wi h ac i e o m on he on page we e clicked while 14.9% o he ecommenda ions wi h passi e o m we e clicked. A chi-squa e es shows a s a is ically signi ican di e ence (X2 = 5,078, p < .05). In addi ion, logis ic eg ession u he demons a es ha he esul s a e s a is ically signi ican (B = -.117, Wald = 5,074, p <.05). O he messages shown a e on page, 9,004 had a passi e message o m and 4,501 an ac i e message o m. 29.7% o he p oduc ecommenda ions wi h passi e message o m shown a e he on page we e clicked, while 21.9% o he p oduc ecommenda ions wi h ac i e message o m we e clicked. The di e ence is s a is ically signi ican based on a chi- Table 1. Recommenda ions Recommenda ion Company Message o m Pe sonaliza ion base Page T ue displays Click- h ough a e % "A ecommenda ion o you" Ou doo appa el Ac i e Pas F on page 4090 18,2 "Recommended o you" Ticke agen Ac i e Pas F on page 443 28,2 "We ecommend also" Consume elec onics Ac i e P esen Pu chase page 2233 16,7 "Buy also" Ou doo appa el Ac i e P esen Pu chase page 2268 26,9 "Buy his" Childe n's wea Ac i e Random F on page 4616 7,9 "The mos iewed" Hype ma ke Passi e Pas Ca ego y page 2734 4,9 "The mos iewed" Ou doo appa el Passi e Pas Ca ego y page 1769 34,8 "The mos wan ed igh now" Ou doo appa el Passi e Random F on page 2429 23 "Picks o Feb ua y" Hype ma ke Passi e Random F on page 2092 5,6 "O he s who iewed his, also iewed" Hype ma ke Passi e P esen P oduc page 4501 42,7 3864 squa e di e ence es (X2 = 93,054, p < .000). Logis ic eg ession was conduc ed o u he alida e he esul s (B = -.527, Wald = 297,036, p < .05). The e ec o message o m on clicking beha io is shown in Figu e 2. The esul s suppo hypo heses 1–3. Figu e 2. E ec o message o m and page on click- h ough a e Nex , an analysis was conduc ed o in es iga e he e ec base o pe sonaliza ion has on clicking beha io wi h ega d o p oduc ecommenda ions shown on he on page. Pas session-based ecommenda ions on he on page we e displayed 3,662 imes, and 9.2% o he displays we e clicked. Random-based ecommenda ions on he on page we e displayed 8,097 imes, and 11.4% o hem we e clicked. A chi- squa e es shows ha he e was a s a is ically signi ican di e ence be ween pas session and andom ecommenda ions on he on page (X2 = 154,565, p < .000). Thus, i can be concluded ha pas session-based ecommenda ions a e mo e e ec i e in gene a ing clicks han andom p oduc ecommenda ions. Logis ic eg ession u he demons a ed ha he e ec o pe sonaliza ion base on clicking in en ions is s a is ically signi ican (B = - 0.616, Wald = 151,479, p < 0.05). Thus, H4 is suppo ed. A simila analysis was conduc ed wi h p oduc ecommenda ions shown a e he on page, including ca ego y, p oduc and pu chase pages. A o al o 13,505 p oduc ecommenda ions on pages o he han he on page we e iewed by cus ome s o he online s o es. O he ecommenda ions based on p esen session (N = 9,002), 32.3% gene a ed clicks while 16.7% o messages based on pas session (N = 4,503) gene a ed clicks. A chi-squa e es shows ha he e was also a s a is ically signi ican di e ence be ween p esen and pas session-based ecommenda ions on pages o he han he on page (X2 = 371,693, p < .000). The esul indica es ha p esen session-based p oduc ecommenda ions gene a e mo e clicks han pas session-based ecommenda ions on ca ego y, p oduc and pu chase pages. Logis ic eg ession was conduc ed o u he alida e he indings (B = -0.672, Wald = 1155,914, p <.05). Thus, H5 is also suppo ed. 5. Discussion 5.1. Theo e ical implica ions The esul s o he p e- es indica e ha cus ome s pe cei e p oduc ecommenda ions ha a e based on in o ma ion abou hei pas b owsing session as mo e in usi e han ecommenda ions ha a e based hei cu en b owsing ac i i y. The esul suppo s p io esea ch on p i acy and in usi eness o online ad e ising [e.g., 12, 27, 28] – using cus ome in o ma ion ha could no ha e been known based on he p esen session’s b owsing ac i i y, is hough o iola e ones p i acy. The esul s also suppo p io esea ch ha has shown ha cus ome s eac nega i ely o explici use o da a [4, 29]. In addi ion, he esul s illus a e ha cus ome s a e mo e in e es ed in p oduc ecommenda ions ha a e based on hei p esen shopping p ocess. The explana ion o his is, mos p obably, ha p oduc ecommenda ions ha a e based on he cu en shopping ac i i y o a cus ome ha e a highe i wi h he cus ome ’s cu en need. This suppo s he no ion o Whi e and colleagues [30] ha he be e jus i ied a pe sonalized message is, he mo e likely consume s a e o espond posi i ely o i . The esul s o also highligh ha a high deg ee o pe sonaliza ion does no necessa ily esul in inc eased click- h ough a es. The esea ch o Whi e and colleagues [30] and Van Doo n and Hoeks a [28] a gue ha in usi eness esul s in lowe pu chase in en ions. The esul s o he p esen s udy indica e ha ecommenda ions wi h a passi e o m gene a e mo e clicks han ecommenda ions wi h an ac i e message o m. This applies o all s ages o a cus ome ’s buying p ocess. The explana ion could be ha cus ome s pe cei e p oduc ecommenda ions wi h an ac i e message o m as mo e in usi e and o ced, esul ing in eac ance due o pe cei ed loss o eedom. Mo eo e , p oduc ecommenda ions wi h passi e message o m may be pe cei ed as unin en ionally pe sonalized o cus ome s. The a gumen o Sela and colleagues [24], which poin s ou ha cus ome s eac posi i ely o ecommenda ions ha a e unin en ionally aluable o hem, may be applied he e. Thus, cus ome s may eel ha passi e messages a e no o ced, and ind hem mo e in e es ing, pa icula ly i hey i hei needs and p e e ences. The analysis o message o m and message base was di ided in o wo ca ego ies based on he page he p oduc ecommenda ion was placed a . The esul shows ha p oduc ecommenda ions on he on page gene a ed less clicks han p oduc ecommenda ions on u he pages, p obably because consume s a e mo e open o selle ’s ecommenda ions a e hey ha e ul illed hei i s immedia e need o isi he pa icula s o e. The dis inc ion o page ca ego ies enables he possibili y o compa e he e ec i eness o p oduc ecommenda ions wi h di e en kinds o 3865 pe sonaliza ion bases. Acco ding o he analysis, pas session-based p oduc ecommenda ions gene a e mo e clicks on he on page han andomly chosen p oduc ecommenda ions. Howe e , on pages a e he on page, p oduc ecommenda ions based on he p esen session gene a e mo e clicks and pu chases han ecommenda ions based on a use ’s p e ious isi s. I can be concluded ha p esen session-based in o ma ion is mo e ele an han pas session-based in o ma ion. The indings o an Doo n and Hoeks a [28] s a e ha a high deg ee o pe sonaliza ion inc eases pu chase in en ions e en hough i also inc eases in usi eness. Howe e , he esul s o his s udy imply ha a high deg ee o pe sonaliza ion inc eases in usi eness and lowe s he e ec i eness o he ecommenda ion. Thus, he mos ecen cus ome beha io da a and passi e message o m a e mos posi i ely esponded by cus ome s. 5.2. Manage ial implica ions The esul s p o ide ools o companies o use when designing hei online pe sonaliza ion s a egies. E-comme ce companies using ecommende sys ems should ake in o conside a ion he deg ee o pe sonaliza ion hey a e applying in hei ad e isemen s and o he con en . Mo e speci ically, manage s should conside he message o m and pe sonaliza ion base o p oduc ecommenda ions. They should also emembe ha he page and s age o he buying p ocess may a ec he ype o p oduc ecommenda ion ha should be used. In gene al, p oduc ecommenda ions in he la e phases o he shopping p ocess gene a e mo e clicks han ecommenda ions on he on page. This esea ch implies ha p oduc ecommenda ions wi h a passi e message o m a e mo e e ec i e han ecommenda ions wi h ac i e message o m in all phases o he buying p ocess. Thus, online s o es should implemen p oduc ecommenda ions ha do no imply he amoun o in o ma ion known abou cus ome s. Gene alized lines such as “ he mos popula ” a e e ec i e o ms o a ge ing cus ome s wi h pe sonaliza ion wi hou c ea ing eac ance – e en i he ecommenda ion would be based on known cus ome in o ma ion. E-comme ce companies should also conside he message base hey use in making he ecommenda ions. Based on his s udy, companies should use he mos ecen in o ma ion hey ha e on hei cus ome s. Thus, in o ma ion ha has been acqui ed du ing pas isi s should be used on he on page in o de o inc ease click- h ough a es. Howe e , a e he on page, such as ca ego y, p oduc and pu chase pages, i is he mos e ec i e o use in o ma ion ha is based on he cu en shopping session o he cus ome . Thus, p oduc ecommenda ions on u he pages should e lec he choices and p e e ences he cus ome has implied on he cu en isi ins ead o pas isi s. This kind o pe sonaliza ion is also app ecia ed by he cus ome s. 5.3. Limi a ions The da a was acqui ed om a company ha p o ides a pe sonaliza ion so wa e o i s clien s. Thus, he da a is limi ed o ce ain ypes o online s o es and o ce ain ypes o p oduc ecommenda ions – he e a e many o he kinds o ecommenda ions ha could ha e di e en kind o e ec . Mo eo e , cus ome s’ clicking and buying beha io may di e be ween he s o es as he sold p oduc s and he designs o he s o es a e di e en . Addi ionally, he p oduc s sold a y in e ms o p ice, which migh a ec he e ec i eness o he ecommenda ion. Howe e , he a iance in p oduc s, s o es and p ices can also be conside ed a s eng h as he esul s p o ide be e gene alizabili y. In ega ds o he esea ch design, u u e esea ch should con ol he exposu e o ecommenda ions based on p esen and pas beha io and he eby ule ou he sel - selec ion bias ha a ec s he esul s o his s udy. A scena io-based p e- es was conduc ed wi h pa icipan s ha we e shown sc eensho s o possible p oduc ecommenda ions. Unde ideal ci cums ances, he same ques ions would ha e been posed o eal cus ome s du ing hei shopping expe ience, and all he di e en ecommenda ion ypes would ha e been conside ed. Howe e , as he esea che s had no con ol o e he decisions o he companies o had any con ac in o ma ion o o he ouchpoin o he cus ome s, such p ocedu e was no possible. In addi ion, no online s o e ha would ha e used all he di e en ypes o p oduc ecommenda ions could no be included in he s udy. The e o e, he compa ed ecommenda ions a e subjec o a numbe o uncon olled a iables. This limi a ion was alle ia ed by ca ego izing he analyzed ecommenda ions as objec i ely as possible. 6. Conclusion Online pe sonaliza ion has become a i al ma ke ing and sales ool o e-comme ce companies. P oduc ecommende sys ems, which apply consume da a in making ecommenda ions, a e a common ool o e-comme ce companies. The e ec s o p i acy issues and pe cei ed in usi eness ha e been s udied in he ad e ising li e a u e bu esea ch on he e ec o online pe sonaliza ion on ac ual clicking beha io is limi ed. Thus, he aim o his s udy was o ill his esea ch gap by u ilizing clicks eam da a om i e di e en online s o es. The esul s sugges ha pe sonalized p oduc ecommenda ions ha a e based on cus ome ’s p e ious b owsing session inc ease pe cei ed in usi eness and dec ease click- h ough a e. The analysis implies ha p oduc ecommenda ions gene a e he mos clicks when hey a e based on he 3866 mos ecen in o ma ion acqui ed o he cus ome . In addi ion, he esul s sugges ha p oduc ecommenda ions wi h passi e message o m gene a e highe click- h ough a es han ac i e message o m, which sugges he need o mi iga e he pe cep ion o in usi eness. 7. Re e ences [1] A. Acquis i, and H.R. Va ian, “Condi ioning p ices on pu chase his o y”, Ma ke ing Science, 24, 2005, pp. 367-381. [2] A. Ansa i, S. Essegaie , and R. Kohli, “In e ne ecommenda ion sys ems”, Jou nal o Ma ke ing Resea ch, 37, 2000, pp. 363-375. [3] A. Ansa i, and C.F. Mela, “E-cus omiza ion”, Jou nal o Ma ke ing Resea ch, 40, 2003, pp. 131-145. [4] T.H. Baek, and M. Mo imo o, “S ay away om me”, Jou nal o Ad e ising, 41, 2012, pp. 59-76. [5] D. Baie , and E. S übe , “Accep ance o ecommenda ions o buy in online e ailing”, Jou nal o Re ailing and Consume Se ices, 17, 2010, pp. 173-180. [6] R.K. Chellappa, and R.G. Sin, “Pe sonaliza ion e sus p i acy: An empi ical examina ion o he online consume ’s dilemma”, In o ma ion Technology and Managemen , 6, 2005, pp. 181-202. [7] A. Che ne , and G.S. Ca pen e , “The ole o ma ke e iciency ins i u ions in consume choice: A case o compensa o y in e ences”, Jou nal o Ma ke ing Re-sea ch, 38, 2001, pp. 349-361. [8] K. Cheung, J.T. Kwok, M.H. Law, and K. Tsui, “Mining cus ome p oduc a ings o pe sonalized ma ke ing”, Decision Suppo Sys ems, 35, 2003, pp. 231-243. [9] C.H. Cho, “Why do people a oid ad e ising on he in e ne ?”, Jou nal o Ad e ising, 33, 2004, pp. 89-97. [10] M. Clancey, “The ele ision audience examined”, Jou nal o Ad e ising Resea ch, 34, 1994, pp. S1-S1. [11] A.G. Close, and M. Kuka -Kinney, “Beyond buying: Mo i a ions behind consume s’ online shopping ca use”, Jou nal o Business Resea ch, 63, 2010, pp. 986-992. [12] S.M. Edwa ds, H. Li, and J. Lee, J., “Fo ced exposu e and psychological eac ance: An eceden s and consequences o he pe cei ed in usi eness o pop-up ads”, Jou nal o Ad e ising, 31, 2002, pp. 83-95. [13] G. Häubl, and K.B. Mu ay, “Double agen s: Assessing he ole o elec onic p oduc ecommenda ion sys ems”, Sloan Managemen Re iew, 47, 2006, pp. 8-12. [14] J.R. Hause , G.L. U ban, G. Libe ali, and M. B aun, “Websi e mo phing”, Ma ke ing Science, 28, 2009, pp. 202-223. [15] M. Kap ein, and P. Pa inen, “Ad ancing e- comme ce pe sonaliza ion: P ocess amewo k and case s udy”, In e na ional Jou nal o Elec onic Comme ce, 19, 2015, pp. 7-33. [16] R. Ki e z, and I. Simonson, “The e ec s o incomple e in o ma ion on consume choice”, Jou nal o Ma ke ing Resea ch, 37, 2000, pp. 427-448. [17] E. Lee, and J.K. Pa k, “Online se ice pe sonaliza ion o appa el shopping”, Jou nal o Re ailing and Consume Se ices, 16, 2009, pp. 83-91. [18] H. Li, S.M. Edwa ds, and J. Lee, “Measu ing he in usi eness o ad e isemen s: Scale de elopmen and alida ion”, Jou nal o Ad e ising, 31, 2002, pp. 37-47. [19] S. McCoy, A. E e a d, P. Polak, and D.F. Galle a, “The e ec s o online ad e ising”, Communica ions o he ACM, 50, 2007, pp. 84-88. [20] R.T. Mo ia y, and Spekman, R.E., “An empi ical in es iga ion o he in o ma ion sou ces used du ing he indus ial buying p ocess”, Jou nal o Ma ke ing Resea ch, 21, 1984, pp. 137-147. [21] O.J. Pos ma, and M. B okke, “Pe sonalisa ion in p ac ice: The p o en e ec s o pe sonaliza ion”, The Jou nal o Da abase Ma ke ing, 9, 2002, pp. 137-142. [22] A.M. Rashid, I. Albe , D. Cosley, S.K. Lam, S.M. McNee, J.A. Kons an, and J. Riedl, J., “Ge ing o know you: Lea ning new use p e e ences in ecommende sys ems”, P oceedings o he 7 h In e na ional Con e ence on In elligen Use In e aces ACM, 2002 pp. 127. 3867 [23] J. Scha e , J. Ben, A. Kons an, and J. Riedl, “E-comme ce ecommenda ion applica ions”, Applica ions o Da a Mining o Elec onic Comme ce, 2001, pp. 115-153. [24] A. Sela, I. Simonson, and R. Ki e z, “Bea ing he ma ke : The allu e o unin ended alue”, Jou nal o Ma ke ing Resea ch, 50, 2013, pp. 691-705. [25] A. Sha ma, and T.F. S a o d, “The e ec o e ail a mosphe ics on cus ome s’ pe cep ions o salespeople and cus ome pe suasion: An empi ical in es iga ion”, Jou nal o Business Resea ch, 49, 2000, pp. 183-191. [26] P.S. Speck, and M.T. Ellio , “P edic o s o ad e ising a oidance in p in and b oadcas media”, Jou nal o Ad e ising, 26, 1997, pp. 61-76. [27] D.G. Taylo , D.F. Da is, and R. Jillapalli, “P i acy conce n and online pe sonaliza ion: The mode a ing e ec s o in o ma ion con ol and compensa ion”, Elec onic Comme ce Resea ch, 9, 2009, pp. 203-223. [28] J. Van Doo n, and J.C. Hoeks a, “Cus omiza ion o online ad e ising: The ole o in usi eness”, Ma ke ing Le e s, 24, 2013, pp. 339-351. [29] S. Wa al, R. Telang, T. Mukhopadhyay, and P. Boa w igh , P., “Wha 's in a “name”? Impac o use o cus ome in o ma ion in e- mail ad e isemen s”, In o ma ion Sys ems Resea ch, 23, 2011, pp. 679-697. [30] T.B. Whi e, D.L. Zahay, H. Tho bjø nsen, and S. Sha i , S., “Ge ing oo pe sonal: Reac ance o highly pe sonalized email solici a ions”, Ma ke ing Le e s, 19, 2008, pp. 39-50.