Article

Journal of Marketing
1-20
Path to Purpose? How Online Customer ª American Marketing Association 2020

Journeys Differ for Hedonic Versus Article reuse guidelines:

sagepub.com/journals-permissions
Utilitarian Purchases DOI: 10.1177/0022242920911628
journals.sagepub.com/home/jmx

Jingjing Li, Ahmed Abbasi, Amar Cheema, and Linda B. Abraham

Abstract
The authors examine consumers’ information channel usage during the customer journey by employing a hedonic and utilitarian
(H/U) perspective, an important categorization of consumption purpose. Taking a retailer-category viewpoint to measure the
H/U characteristics of 20 product categories at 40 different retailers, this study combines large-scale secondary clickstream and
primary survey data to offer actionable insights for retailers in a competitive landscape. The data reveal that, when making hedonic
purchases (e.g., toys), consumers employ social media and on-site product pages as early as two weeks before the final purchase.
By contrast, for utilitarian purchases (e.g., office supplies), consumers utilize third-party reviews up to two weeks before the final
purchase and make relatively greater usage of search engines, deals, and competitors’ product pages closer to the time of
purchase. Importantly, channel usage is different for sessions in which no purchase is made, indicating that consumers’ information
channel choices vary significantly with the H/U characteristics of purchases. The article closes with an extensive discussion of the
significant implications for managing customer touchpoints.

Keywords
customer journey, hedonic and utilitarian products, information sources, path to purchase, touchpoint management
Online supplement: https://doi.org/10.1177/0022242920911628

With the proliferation of electronic commerce, examining the Given the nuances of customer journeys on the internet,
role of various information channels during the customer jour- identifying the mechanisms behind information channel
ney is becoming increasingly important. A “customer journey” choices and usage remains challenging. Channel choices and
is the series of actions a customer takes to arrive at the moment usage are contingent on diverse retailer- and product-category
of purchase (Lemon and Verhoef 2016). Importantly, these characteristics as well as heterogeneous consumer preferences.
actions include an examination of various information sources For example, retailers differ considerably in their product and
and the evaluation of alternatives before the purchase decision. service offerings, brand equity, and target customer segments,
Online information channels, which conveniently provide a potentially influencing consumers’ information search beha-
variety of pertinent information (Shankar et al. 2011), signifi- viors. Moreover, consumers’ shopping characteristics (Kush-
cantly affect purchase decisions (Batra and Keller 2016; Li and waha and Shankar 2013), prior shopping experience (Frambach
Kannan 2014). As retailers’ spending on online marketing con- et al. 2007), trust (Bart et al. 2005), and demographics (Inman,
tinues to grow, understanding how to best allocate resources
across various touchpoints necessitates a “360-degree view” of
how customers interact with and leverage multiple information Jingjing Li is Assistant Professor of Commerce, McIntire School of Commerce,
channels throughout the customer journey (Kannan and Li University of Virginia, USA (email: jl9rf@virginia.edu). Ahmed Abbasi is
2017). Accordingly, several path-to-purchase information Professor of Commerce and Murray Research Professor, McIntire School of
channels have garnered considerable attention, including Commerce, University of Virginia, USA (email: ana6e@comm.virginia.edu).
search engines (Ghose and Yang 2009), social media (Rishika Amar Cheema is Professor of Marketing and William F. O’Dell Professor of
Commerce, McIntire School of Commerce, University of Virginia, USA (email:
et al. 2013), review sites (Chevalier and Mayzlin 2006), deal cheema@virginia.edu). Linda B. Abraham is Co-Founder of comScore Inc., Vice
sites (Kumar and Rajan 2012), and retailer product pages Chair of Upskill, and Managing Director of Crimson Capital, USA (email:
(Huang, Lurie, and Mitra 2009). linda@crimsoncap.us).
2 Journal of Marketing XX(X)

Shankar, and Ferraro 2004) may also lead to significantly dif- of 20 product categories sold by 40 retailers—a total of 115
ferent channel choices. Furthermore, channel choices are inter- retailer–product combinations. To understand the H/U effect
dependent and could vary at various stages of the customer on actual channel usage, we analyze a large volume of com-
journey (De Haan, Wiesel, and Pauwels 2016; Moe 2003; Score clickstream data that includes all internet activities from
Valentini, Montaguti, and Neslin 2011). Consequently, market- 4,356 consumers with 22,751 purchases that account for $1.2
ing managers continue to wrestle with how to best allocate million sales during a 24-month period. We use a hierarchical
resources for a variety of product offerings across an array of Bayesian approach to consider channel interdependency, retai-
online touchpoints at different stages of the customer journey ler, product, and individual heterogeneity in online information
(Anderl, Schumann, and Kunz 2016; Batra and Keller 2016). channel usage.
In addition to a utility-based perspective popular in the We find that consumers making hedonic purchases tend to
extant literature, recent studies have called for a social and utilize social media and product page views on the target retai-
psychological angle (e.g., Kushwaha and Shankar 2013) to ler’s website more extensively than people engaging in utilitar-
investigate information channel usage patterns and customer ian purchases. By contrast, consumers making utilitarian
journeys. Purpose has been identified as an important consid- purchases tend to use search engines, third-party reviews, deal
eration. A recent study published by Google’s Zero Moment of sites, and product page views on the competing retailer’s web-
Truth (Taniguchi 2019) finds consumer search behaviors to be site more frequently than those engaging in hedonic purchases.
driven by six needs: the need for surprise, help, reassurance, Furthermore, we explore the dynamics of channel usage
education, thrill, or the need to be impressed. These needs and between hedonic and utilitarian purchases throughout the cus-
purposes are shaped by not only the product category but also tomer journey. We find the H/U effect on the usage of social
by where consumers are in their journey, namely their “path to media and third-party review sites to be stronger earlier in the
purpose.” customer journey. Conversely, the effect of H/U differences on
In this study, we use a hedonic–utilitarian (H/U) perspec- search engines and deal site usage is stronger closer to the point
tive—a purpose-oriented categorization of consumption exten- of purchase. While the H/U effect on product page views is
sively studied in the marketing literature (Holbrook and significant throughout the customer journey, the magnitude
Hirschman 1982)—to explore information channel usage pat- decreases toward the end of the customer journey.
terns across customer journeys. The H/U characteristics of pur- Finally, for unconverted hedonic purchases, consumers visit
chases reflect affective and instrumental motives, which could social media sites less often, visit deal sites more, and are more
provide a richer picture of consumers’ perceptions toward pur- likely to benchmark with competing retailers’ product pages
chases as well as consumers’ information search behaviors. For compared with converted sessions, suggesting a potential guilt-
example, emotions such as fun and guilt have emerged as justification effect. For unconverted utilitarian purchases, the
important considerations for hedonic purchases, with implica- four channels that facilitate information search are less utilized,
tions for preferences of certain information channels (Moe indicating an insufficient information search for purchase
2003; Schulze, Schöler, and Skiera 2014). Furthermore, many decisions.
customer-centric measurement scales have been proposed to Our research aims to make at least four important academic
quantify the characteristics of product categories or brands contributions. First, we extend the customer journey literature
based on the H/U dimensions (Babin, Darden, and Griffin by complementing the utility-centric perspective with a social/
1994; Batra and Ahtola 1991; Voss, Spangenberg, and Groh- psychological angle. We analyzed the H/U effect on six pre-
mann 2003), which has the potential for large-scale analysis of purchase information channels for 20 product categories across
individual customer journeys across a myriad of retailers and 40 retailers. An examination of this interplay is conceptually
product categories. and theoretically significant because it provides a new angle to
Therefore, we contribute to the literature by examining how understand the role of affective mechanisms such as amuse-
the usage of a rich set of digital information channels—search ment seeking, guilt justification, and brand affect during the
engines, social media, third-party reviews, deal sites, and prod- shopping process. Second, we uncover the dynamics of H/U
uct pages of target retailers and competing retailers—during effects throughout the customer journey, which allows more
the customer journey differs by the retailer-product-level H/U actionable insights for marketing managers and adds to the
characteristics. The specific research questions we answer are nascent research on the temporal effect of the customer journey
the following: (Batra and Keller 2016; Lemon and Verhoef 2016). Third, we
highlight the importance of considering H/U characteristics at a
1. Do consumers use digital information channels differ- more granular level. Unlike the existing H/U literature, which
ently for H/U purchases? mostly focuses on product-level differences, our survey shows
2. How does this usage vary over the customer journey? a considerable variation of H/U characteristics for similar prod-
3. Does this usage vary between converted and uncon- uct categories across retailers, calling for a retailer-category
verted sessions? vantage point for future H/U research. Finally, we contribute
to the literature regarding the use of Big Data for deriving
We examine these research questions using both primary marketing insights in complex digital environments (Bradlow
and secondary data. We first survey the H/U characteristics et al. 2017; Kitchens et al. 2018; Sudhir 2016; Wedel and
Li et al. 3

Kannan 2016). We demonstrate the benefits of conducting point of purchases. First, the theoretical underpinnings for
innovative Big Data marketing research by combining a variety H/U draw from cognitive/social psychology—particularly in
of research methods and data sources. Our research utilizes consideration of both affective and cognitive attitudes (Batra
survey analysis, clustering, text mining, and Bayesian model- and Ahtola 1991; Sirgy 1982). These affective–cognitive trade-
ing and seamlessly combines survey-based primary data and offs have the potential to complement the utility-centric
large-scale secondary clickstream. As a result, we provide a information-seeking view adopted by the S/E perspective (Nel-
more comprehensive view of customer journey across channels son 1970). For instance, emotions such as pleasure and guilt
and stages. have emerged as important considerations for certain forms of
Our work offers several actionable implications for mar- consumption, with implications for path-to-purchase channels
keting managers’ digital spend allocation and online market- such as social media (Schulze, Schöler, and Skiera 2014).
ing strategies. We suggest that marketing managers collect Furthermore, when processing information about the product
consumers’ H/U perceptions of their product offerings rela- (e.g., the product name), consumers process hedonic products
tive to their competitors. Leveraging our empirical model, more holistically than utilitarian products (Melnyk, Klein, and
marketing managers can use the obtained H/U characteristics Volckner 2012). Second, in contrast to the S/E’s narrow focus
of their products to understand the shopping purposes of their on the product categories, the H/U perspective enables
customers, most valuable information channels, and the most customer-centric thinking by quantifying the H/U characteris-
common sequences of touchpoint prospects at different tics of product categories or brands from the customer’s per-
stages of the customer journey. Accordingly, they can design spective. One of the most popular scales (from Voss,
their marketing strategies on the basis of not only a utility- Spangenberg, and Grohmann 2003) allows measurement of the
centric view but also the social/psychological needs of their customer’s perceived H/U characteristics of purchases at both
customers (Batra and Keller 2016), thereby enhancing the the product-category and retailer levels, making it highly con-
consumer experience on the path to purchase (Lemon and ducive to a large-scale examination of purchases from different
Verhoef 2016).
customers on various retailers’ product categories.
Third, the hedonic and utilitarian dimensions are indepen-
dent (Voss, Spangenberg, and Grohmann 2003). As Batra and
Conceptual Development Ahtola (1991, p.161) observe, hedonic and utilitarian
Hedonic and Utilitarian Characteristics of Purchases “motivations for consumption need not be (and usually are not)
mutually exclusive: a toothpaste may both prevent cavities and
The importance and impact of consumer goals on the purchase
provide pleasure from its taste.” Thus, the bidimensional anal-
process have been emphasized extensively in the prior litera-
ysis allows for granular assessment of the role of purpose in the
ture. Different purpose-oriented categorizations of consump-
customer journey. For these three reasons, we use H/U as our
tion have been employed, with search–experience (S/E) and
primary perspective to examine path-to-purchase channels.
H/U perhaps being the two most prevalent. In line with differ-
ences in the cognitive processes related to the acquisition of However, to be more holistic in our operationalization of pur-
alternative forms of information, the S/E perspective highlights pose, we also include S/E as a control variable in our model.
the different information-seeking behaviors associated with Despite the tremendous potential of the H/U perspective to
search goods and experience goods (Huang, Lurie, and Mitra enhance our understanding of path-to-purchase tendencies,
2009). By contrast, the H/U perspective emphasizes the bidi- prior studies have typically not considered the role of H/U
mensional consumer attitudes toward brands and consumption characteristics on consumers’ channel usage during the pur-
that stem from affective and instrumental motives (Holbrook chase funnel. In addition, previous studies of customer jour-
and Hirschman 1982). Hedonic consumption is based on the neys focus on a few channels or on a single retailer site, which
consumer’s experience of shopping, emotional attachment, calls for a more comprehensive view with multiple channels,
focusing on fun, playfulness, enjoyment, excitement, and the product categories, and retailers. Moreover, H/U characteris-
need for surprise (Arnold and Reynolds 2003; Babin, Darden, tics are not unique to the product (category) level but also
and Griffin 1994). By contrast, utilitarian consumption is often manifest at the retailer (brand) level (Voss, Spangenberg, and
more goal-directed and pertains to the need to complete spe- Grohmann 2003). Furthermore, there is potential for differ-
cific tasks efficiently and effectively (Childers et al. 2002; ences in consumers’ behavior between product-category-level
Mathwick, Malhotra, and Rigdon 2001). Recent studies and retailer-level H/U characteristics. For example, indepen-
demonstrate the importance of H/U characteristics for pur- dent of the product, consumers demonstrate greater affective
chases. For example, Kushwaha and Shankar (2013) show that involvement with hedonic retailers and relatively more cog-
consumers who make hedonic purchases are likely to utilize nitive involvement with utilitarian retailers (Zaichkowsky
multiple purchase options. Moreover, Park et al. (2018) find 1994). In addition, retailer-level characteristics are usually
that gamers’ social network ties on an online gaming platform associated with brand positioning (Park, Jaworski, and
significantly influence the spending for hedonic products. MacInnis 1986). These findings suggest the potential for a
The H/U perspective affords at least three opportunities to simultaneous retailer- and product-category-level effect that
enrich and enhance insights gained through the S/E vantage may influence H/U motivations.
4 Journal of Marketing XX(X)

A: Selected Categories at Walmart, Home Depot,

and Amazon B: Across Categories Within Amazon
30 30
walmart
home depot
amazon a_computing
h_garden
20 20 a_books
a_office a_office a_home
a_electronics
a_home a_electronics
a_telecom
w_office a_sports a_apparel a_toys a_sports a_apparel a_toys
a_garden
10 10 a_garden a_photography
w_garden w_home a_pet
a_automotive a_music
a_beauty w_electronics a_beauty
Utilitarian

Utilitarian
w_apparel
w_beauty
a_gifts a_cds_dvds
h_home w_toys
0 0
w_sports a_jewelry a_jewelry

h_electronics a_artsantiques
-10 -10
a_wines
w_jewelry

-20 -20

-30 -30
-30 -20 -10 0 10 20 30
0 -30 -20 -10 0 10 20 30
0
Hedonic Hedonic

Figure 1. H/U plots for Amazon, Home Depot, and Walmart for selected categories and for various product categories at Amazon.
Notes: x- and y-axes are calibrated as absolute deviations from the mean (0, 0).

To illustrate this effect, we conducted a survey involving Differences in Customer Journey for Hedonic and
3,250 Amazon Mechanical Turk (MTurk) participants to report Utilitarian Purchases
their H/U perceptions toward 115 common retailer categories.
Each participant randomly evaluated the H/U characteristics Drawing from research on hedonic and utilitarian purchases,
for six retailer-category combinations, resulting in approxi- utilitarian purchasing is a relatively more goal-directed cogni-
mately 100 responses for each retailer-category combination. tive process, while hedonic purchasing is a comparatively more
Figure 1 shows the H/U plots for some common product cate- goal-ambiguous, emotional experience. This cognitive and
gories at Walmart, Home Depot, and Amazon (Panel A), and affective dichotomy not only defines the goals of online shop-
within Amazon (Panel B). Looking at the Panel A, we see that ping but also influences channel preferences. Consumers eval-
the same product category is perceived differently across retai- uate the outcome of an exchange process with another entity
lers. For the electronics category, Home Depot is positioned (e.g., channel, retailer) by comparing the relevant perceived
low on utilitarian and in the middle for hedonic, while Amazon benefits against perceived costs (Bagozzi 1975). These benefits
and Walmart are high on both utilitarian and hedonic dimen- and costs include economic utility and social and psychological
sions. Similarly, Amazon’s jewelry and sports products are returns, such as enjoyment, trust, and respect. In the context of
considered more hedonic relative to Walmart’s. Panel B shows hedonic and utilitarian purchases, we expect that these inherent
differences in consumer H/U perceptions across many of Ama- cognitive and affective differences would result in varying
zon’s product categories. benefits and costs associated with different digital channels.
In summary, these charts highlight the notions that (1) the Moreover, channel usage would likely differ depending on
same product category can have varying H/U perceptions where consumers are on their journey, time-wise. Thus, we
across different retailers and (2) the same retailer can have also explore how the H/U effects change dynamically as the
different H/U characteristics for its product categories. Collec- consumer progresses through the purchase funnel. Finally,
tively, to account for these important variations, the plots prior studies (e.g., Kushwaha and Shankar 2013) have shown
underscore the value of examining H/U at the “retailer- that the choice of information channels could affect conversion
category” level. In addition, the plots reinforce the potential outcomes. Accordingly, we also examine the H/U effect on the
value of considering the hedonic and utilitarian dimensions path to nonpurchases. Next, we discuss how information chan-
separately to allow for more nuanced analysis between retailer nel usage may differ across hedonic and utilitarian purchases.
categories in the four quadrants, as well as between the ones A review of relevant literature is included in Table 1.
along the same diagonals. Next, we discuss how cross-channel Utilitarian purchases are rational and goal-driven, with the
customer journeys may vary across hedonic and utilitarian objective of making the best purchasing decision (Novak, Hoff-
retailer-category combinations. man, and Duhachek 2003). Therefore, they often require deeper
Li et al. 5

Table 1. Summary of Prior Research on Hedonic/Utilitarian Purchases and Online Information Channel Usage.

ProdPage_ ProdPage_ Study Multiple Multiple

Study Search Social Review Deal Target Competitor Design Data Products Sites

Ghose and Yang (2009) Uþ Empirical Secondary Yes No

Chiang and Dholakia (2003) Uþ Experiment Primary Yes No
Kim and LaRose (2004) Uþ Survey Primary No No
Lin and Lu (2015) Hþ Survey Primary No No
Schulze, Schöler, and Skiera U Empirical Secondary Yes Yes
(2014)
Park et al. (2018) Hþ Empirical Secondary Yes Yes
Sen and Lerman (2007) H Experiment Primary Yes No
Kushwaha and Shankar (2013) Hþ Hþ, U Empirical Primary þ Yes Yes
secondary
Khan and Dhar (2010) Hþ Experiment Primary Yes No
Wakefield and Inman (2003) H Experiment Primary Yes No
O’Curry and Strahilevitz (2001) Hþ Experiment Primary Yes No
Okada (2005) Hþ Experiment Primary Yes No
Moe (2003) Uþ, H Empirical Secondary Yes No
Moe and Fader (2001) H Empirical Secondary No No
Novak, Hoffman, and Hþ Experiment Primary Yes No
Duhachek (2003)
Sloot, Verhoef, and Franses H Interview Primary Yes Yes
(2005)
Noble, Griffith, and Uþ Survey Primary Yes No
Weinberger (2005)
Chaudhuri and Holbrook Hþ Survey Primary Yes No
(2001)
Van Trijp, Hoyer, and Inman Hþ Experiment Primary Yes No
(1996)
Heitz-Spahn (2013) Uþ Survey Primary Yes Yes
Mallapragada, Chandukala, and H Survey þ Primary þ Yes Yes
Liu (2016) empirical secondary
Hughes, Swaminathan, and Hþ Experiment Primary þ Yes Yes
Brooks (2019) secondary
Current study Uþ Hþ Uþ Uþ Hþ Uþ Survey þ Primary þ Yes Yes
empirical secondary
Notes: H ¼ hedonic; U ¼ utilitarian; PPT ¼ ProdPage_Target; PPO ¼ ProdPage_Competitor; þ ¼ positive effect;  ¼ negative effect.

information processing across concrete, predefined purchase product offerings from different retailers (Kim and LaRose
attributes in an efficient manner (Mathwick, Malhotra, and 2004). Second, third-party review sites provide quantitative
Rigdon 2001; Park et al. 2018). Moreover, utilitarian purchases and qualitative information about product attributes for com-
are often deliberate and planned, with well-defined dominant parison, making them more conducive for utilitarian purchases
attributes that are easy to compare. Accordingly, this ease of that usually have well-defined and searchable attributes. Simi-
comparison reduces brand differentiation and increases price sen- larly, deal sites allow consumers to search for the best deals
sitivity (Noble, Griffith, and Weinberger 2005). Consequently, efficiently and conveniently, which could be useful for consu-
consumers purchasing utilitarian products tend to prefer informa- mers to optimize their spending. Finally, because brand differ-
tion channels that allow for convenient and efficient searches and entiation in utilitarian purchases is less extensive, consumers
comparisons for product attributes and prices across various alter- are more likely to browse product pages across multiple retai-
natives so as to optimize purchasing decisions. lers to optimize their time, place, and possession needs (Noble,
According to prior literature, certain channels could be more Griffith, and Weinberger 2005). As a result, they could adopt a
effective for utilitarian purchases. First, search engines pro- “cross-channel free-riding” behavior where one retailer’s chan-
mote efficiency-oriented shopping by allowing customers to nel is used to prepare a purchase that is eventually completed at
easily and quickly find products through specifying attributes another retailer (Heitz-Spahn 2013). In this regard, their prod-
of interest via search queries (Chiang and Dholakia 2003; uct page browsing on competing retailers’ websites could be
Ghose and Yang 2009). The list-wise, clear, and condensed more extensive.
format of the search results and the large-scale indexed content Consumers making hedonic purchases seek surprise, adven-
enable consumers to quickly navigate alternatives and compare ture, fun, and variety during their shopping process (Arnold and
6 Journal of Marketing XX(X)

Reynolds 2003; Novak, Hoffman, and Duhachek 2003). These extend page views across multiple retailers (Kushwaha and
goals imply a unique set of perceived benefits that consumers Shankar 2013; Novak, Hoffman, and Duhachek 2003; Van
may consider when seeking and attaining information pertain- Trijp, Hoyer, and Inman 1996). Consequently, due to the mixed
ing to hedonic purchases. Due to the affective nature of hedonic findings in the prior literature, there remains a need to formally
purchases, consumers are more likely to rely on simple cues examine the effect of H/U characteristics on information chan-
and heuristics rather than deeper information processing to nel usage.
reach their purchase decision (Park et al. 2018). Instead of
trying to find the best alternatives, consumers could have a
strong “affective attachment” to brands (Chaudhuri and Hol- Empirical Analyses
brook 2001) and may process information more holistically
To provide a comprehensive picture of how consumers’ utili-
(Melnyk, Klein, and Volckner 2012). Consequently, consumers
zation of different online information channels throughout a
making hedonic purchases could spend less time on searching
customer journey varies with the H/U purchases, we operatio-
and comparing. However, prior research has also found that
nalized the H/U characteristics at the retailer-category level
consumers buying hedonic products may engage in guilt-
through a survey. We then combined primary (survey) and
reducing justification behaviors (Kivetz and Simonson 2002;
secondary (clickstream) data to demonstrate the H/U effect
O’Curry and Strahilevitz 2001; Okada 2005) by spending more
on the usage of different information channels prior to pur-
time in the search process. Consumers could also engage in a
chases. The data analysis process and the conceptual model are
variety-seeking behavior (Kushwaha and Shankar 2013;
depicted in Figure 2.
Novak, Hoffman, and Duhachek 2003) due to considerable
product differentiation in hedonic purchases (Van Trijp, Hoyer,
and Inman 1996). Therefore, the complex nature and multiple Retailer and Product Category Selection
mechanisms of hedonic purchases could have different impli- Our retailer-category analysis covers 20 product categories
cations for information channel search under various contexts. from 40 online retailers. To select appropriate retailers, we
Regarding the information channels for hedonic purchases, started with 500 top internet retailers from 2014 sales rankings
social media has emerged as an influential channel, with 70%– (https://www.internetretailer.com/top500) and narrowed that
80% of study respondents reporting that their purchases are list down to 336 that have easily discernible product page URL
affected by the social media posts of companies and friends patterns. Combined with a comScore data set from 2013 to
(eMarketer 2017; Hewett et al. 2016; Kumar et al. 2016). Pre- 2014, we found that these retailers’ number of transactions
vious studies show that consumers find fun- and entertainment- shows a Pareto-like distribution, with the top 40 retailers
oriented social media to be a more suitable information source accounting for over 91% of all transactions. We also examined
for hedonic purchases (Hughes, Swaminathan, and Brooks the main products sold by these 40 retailers and found that they
2019; Liu et al. 2019; Park et al. 2018; Schulze, Schöler, and cover a wide range of hedonic and utilitarian product cate-
Skiera 2014). However, for third-party reviews, prior research gories. Thus, these retailers were included in our study (for
has not been definitive regarding their implications for hedonic details, see Web Appendix W1).
purchases. On the one hand, the abstract attributes of hedonic For product categories, we initially used the 22 categories
products are less conducive to comparisons through review proposed by Kushwaha and Shankar (2013), which adequately
aspects and dimensions. On the other hand, the lack of concrete captured the major product categories on the H/U spectrum.
attributes also results in uncertainty for hedonic purchase Drawing from the product purchases on these retailers on a
(Kushwaha and Shankar 2013), which might drive greater comScore data set, we found 115 unique retailer-category com-
usage of qualitative comments. binations (e.g., Amazon apparel). We then mapped all the
Similarly, the prior literature on deal websites has been extracted categories to the initial 22 and removed 2 that were
ambivalent regarding their implications for hedonic purchases. absent, resulting in a final set of 20 product categories.
Deals have been found to be more effective for hedonic pur-
chases (Khan and Dhar 2010), supporting the notion that users
favor guilt-alleviation mechanisms to justify hedonic consump-
Main Study: Clickstream Analysis
tion (Okada 2005). However, deals could also be less helpful Data. We collected approximately 1 terrabyte of the U.S. com-
because consumers are less price-sensitive due to the difficulty Score web clickstream data between January 2013 and Decem-
of comparing hedonic products (Wakefield and Inman 2003). ber 2014. The data recorded all online clicking behaviors in
Finally, the findings for product page views for hedonic pur- the form of URLs and timestamps from approximately
chases are also mixed. The experiential nature of hedonic pur- 100,000 randomly selected households each month. The
chases is more closely aligned with hedonic browsing behavior, clicked URLs and timestamps were grouped into clickstream
which is characterized by a leisurely examination of fewer sessions. The end of a clickstream session is determined when
product pages and often results in impulse purchases (Malla- clicking behaviors are inactive for a certain period of time.
pragada, Chandukala, and Liu 2016; Moe and Fader 2001; Park For the clickstream sessions involving a purchase, extra infor-
et al. 2012). Hedonic consumption is also associated with mation (e.g., purchased product categories) was provided by
greater variety-seeking behaviors, which could potentially comScore. The unit of analysis of our study is a purchase
Li et al. 7

Prestudy: H/U Survey based on

Main study:
Voss et al. (2003)
Clickstream analysis
H/U perceptions for
H/U effect on actual
20 product categories from
channel usage
40 retailers Identification
Strategy

Online Information Conceptual

Search During
Model
Customer Journey
Utilitarian Purchases
Goal-driven, cognitive
Concrete product attributes Search engine
More brand switching
Deliberate, deep, and attribute-
specific information processing
Social media
Hedonic Purchases
Fun, surprise, variety-seeking,
and guilt alleviation
Intangible product attributes
Reviews
Less brand switching
Ambiguous, holistic, simple-
cue, and heuristics-driven
information processing
Deals

Product pages
on target retailer
Control Variables
Demographics; retailer-, product-,
and purchase-specific Product pages
characteristics on competing
retailers

Figure 2. Identification strategy and conceptual model.

session (converted or unconverted), which captures a consu- are sometimes coded with product IDs. Accordingly, we used
mer’s decision-making process toward an actual or intended their product application programming interface to map prod-
purchase. A converted session denotes an online purchasing uct IDs to our focal product categories. The average purchase
cycle for a consumer, starting with an information search cycle for each product category is presented in Table W2.1 of
across various information channels and ending with a pur- Web Appendix W2. We found that while some product cate-
chase. Due to the complexity of information search, this pur- gories exhibit shorter or longer cycles (e.g., office, music),
chase cycle can last for several days. most of the cycles last around 14 days. Consequently, we used
Determining the length of the purchasing cycle is often 8–14, 2–7, and 0–1 day windows to capture consumers’ infor-
challenging. Existing literature has mixed findings regarding mation channel usage during the early, middle, and late stages,
cycle lengths; depending on the research context and product respectively, of the path to purchases.
categories, it could range from several days to one month (De Each consumer could have multiple purchases during the
Los Santos, Hortaçsu, and Wildenbeest 2012; Johnson et al. two-year period, which creates an opportunity for us to account
2004). We derived the purchase cycle length for each of the 20 for variations at both the session and consumer level. To
product categories from the comScore data. The intuition for cleanly attribute channel usage to a unique purchase cycle,
our method is that consumers may start a purchase cycle by we removed the purchases that overlapped within a 14-day
browsing product pages related to an intended product category window (20.78% of the sessions). Furthermore, we found that
on any channel. Thus, our algorithm tracked their first encoun- approximately 23.3% of those purchases involve multiple
ter with these pages. Web Appendix W2 documents the process product categories, which could complicate our analysis
in detail. The key to this method is to ensure that product page because these multicategory purchases could have more than
URLs could be accurately mapped to our 20 product categories. one H/U characteristic. Therefore, we also removed these pur-
Many retailer websites embed product names or categories in chases from our study.
the product page URLs. Therefore, we developed a text-mining To identify whether the H/U effect on information channel
algorithm (described in Web Appendix W3) to extract product usage is only restricted to converted sessions, we also included
categories from these URLs. This method is conducive for all unconverted sessions of the same consumers identified previ-
retailers except Amazon and Walmart, whose product URLs ously. Note that we only examine online unconverted sessions
8 Journal of Marketing XX(X)

for these consumers because our data set is limited to online W5 presents the set of social media, third-party reviews, and
web clickstream—it is possible that a consumer did not pur- deal sites included.
chase online but purchased through other channels (e.g., off-
line). Specifically, an online unconverted session denotes a Control variables. We included several control variables to
website visitation with an intended purchase (i.e., focused address the selection bias that commonly occurs in secondary
product browsing) but exit before completion (e.g., cart aban- data analysis. Drawing on this principle and previous studies
donment or leaving the website before adding products to the (Huang, Lurie, and Mitra 2009; Kushwaha and Shankar 2013),
cart). While the comScore data nicely flags clickstream ses- we identified five types of control variables:
sions without purchases, we don’t know whether these sessions
have product purchasing intentions. Thus, the H/U character- 1. Retailer-specific controls: It is important to control for
istics of this session could not be determined directly from the retailer heterogeneity because the key independent vari-
comScore data. Fortunately, we can use the text mining method able H/U scores are at the retailer-category level, and
described above to infer the intended products from browsed retailers’ diverse marketing strategies could affect
URLs. Specifically, we identified the product categories from channel usage (Valentini, Montaguti, and Neslin
all the browsed URLs during the day of a clickstream session. 2011). Thus, we included five types of retailer-level
The category that receives the highest presence is considered as controls. First, we used the top 500 Internet Retailer
the intended product category. We found this method to be sales ranking in 2014 to approximate each retailer’s
reasonable because people rarely browse product pages unless popularity rank. Second, we incorporated the visit vol-
they have a purchase intention—only 4.6% of the unconverted ume across the comScore clickstream to account for
clickstream sessions have product page views during the ses- retailer popularity rank specific to the comScore panel.
sion day. Therefore, we include only this subset of sessions Third, we used the number of page views per user
with purchase intentions in our unconverted data set. Further- obtained from Alexa (www.alexa.com) to control the
more, for each consumer, we removed all the unconverted level of consumer engagement with different retailers.
sessions that tap into the converted sessions (approximately Fourth, we collected the number of likes for each retai-
28% removed) or overlap with each other on a 14-day window ler’s Facebook page to control for their social media
(approximately 43% removed). Similarly, we derived 8–14, 2– presence. Because all these controls are highly posi-
7, and 0–1 day windows to examine the early, middle, and late tively skewed, we log-transformed them before inclu-
stages, respectively, of the path to nonpurchases. sion. Finally, we used a 20-dimensional product
category vector to represent the product assortment of
each retailer. Specifically, each dimension corresponds
Key independent variable. We conducted a survey on MTurk to
to a product category, and the dimensional value
derive our key independent variables related to the H/U aspects
denotes the proportion of this category purchased dur-
of a product category purchased on a specific retailer. Details
ing a two-year period. The higher the proportion, the
of the survey are provided in Web Appendix W4. Following
greater the likelihood that it is a primary category for
Kushwaha and Shankar (2013), we calculated a mean compo-
this retailer.
site hedonic (utilitarian) score by averaging the scores of the
2. Product category control: We included an S/E dummy
five hedonic (utilitarian) scale items. Because H/U have a low
variable to control for category-level characteristics
correlation of .16, we used separate H/U scores for each
other than H/U. The S/E assignment for each category
retailer-category combination. To account for the heterogene-
is based on the prior literature (e.g., Huang, Lurie, and
ity of H/U perceptions across consumers, we imputed the H/U
Mitra 2009). For example, home and garden products
scores for the clickstream data using the insights from the
are search goods (S/E ¼ 1), and beauty and automotive
MTurk survey. Consequently, consumers with varying demo-
products are experience goods (S/E ¼ 0). By including
graphic characteristics in the clickstream will have different H/
this control, we also intend to empirically illustrate how
U scores despite the same retailer-category purchases. Finally,
the two divergent perspectives complement one
both scores are mean-centered before being included in the
another, as alluded to in the conceptual development
study.
section.
3. Prior purchase experience: According to Valentini,
Dependent variables. Information channel usage is defined as the Montaguti, and Neslin (2011), channel usage may be
number of visited channel URLs corresponding to a search determined by state dependence: how many purchases a
engine, social media, third-party reviews, deals, product page consumer has previously made. Therefore, we included
views on target retailers, and product page views on competing the number of prior purchases, as well as the number of
retailers. Following prior work, the visited URLs were mapped purchases specific to the intended product category
to six channels using URL token matching (Moe 2003). Web prior to the current session, to control for a potential
Appendix W1 shows the product categories for identifying systematic shift in channel usage over time.
competing retailers (i.e., retailers offering the same product 4. Price: Prior research shows that H/U characteristics
categories are considered as competitors), and Web Appendix might be correlated with the dichotomy of luxuries and
Li et al. 9

necessities (Khan, Dhar, and Wertenbroch, 2005). Table 2. Operationalization of Variables in the Clickstream Data.
Thus, we included price to control for channel usage
Variable Operationalization
driven by other category- and retailer-specific factors.
Note that this variable is only available for converted Dependent Variables
sessions. Search Search engine visits (e.g., Google, Bing,
5. Demographics: Prior studies (e.g., Inman, Shankar, and Yahoo)
Ferraro 2004) have shown that demographics such as Social Social media site visits (36 websites
including Facebook, Twitter, Pinterest,
age, family size, and education play an important role in etc.)
determining channel usage. Therefore, we included user Deal Deal site visits (31 websites including
demographic information accompanying the comScore Slickdeals, eBates, Coupons, etc.)
clickstream as consumer-level controls. Specifically, Review Third-party review site visits (32 websites
we incorporated household size, age, gender, education, including Consumer Reports, Epinions,
income level, and the presence of children. Yelp, etc.)
ProdPage_Target Product page views on the target retailers
To account for unobserved consumer heterogeneity in informa- ProdPage_Competitor Product page views on the competing
tion channel usage (in addition to demographics), we adopted a retailers (retailers sharing the same
product categories)
hierarchical Bayesian approach to estimate the parameters of
Independent Variables
interest. Given the complexity of the proposed model, it is Hedonic Hedonic composite score at the retailer-
challenging to estimate the entire data set—the computational category level (centered on the grand
time increases significantly with the number of consumers. We mean)
noticed that 55.70% of the consumers made only two purchases Utilitarian Utilitarian composite score at the
in a two-year period but only occupied 19.60% of the total retailer-category level (centered on the
transactions. We removed these consumers without losing gen- grand mean)
Control Variables
eralizability, resulting in 22,751 converted sessions and 30,550
Rank Log-transformed average sales rank for
unconverted sessions from 4,356 consumers in a two-year each retailer in 2014
period, generating approximately $1.2 million in total sales.1 VisitCS Log-transformed visit volume for each
Note that only 3,854 of these consumers have qualified uncon- retailer from comScore
verted sessions. A detailed description of all variables included PageViews/User Log-transformed page viewers/user for
in our model appears in Table 2. Summary statistics for the each retailer from Alexa
14-day sample appear in Web Appendix W6. These statistics Like Log-transformed number of likes at each
retailer’s Facebook page
show that most channel usage and control variables are differ-
ProductOffering A 20-dimensional vector representing
ent in hedonic and utilitarian conditions. proportional product offerings of each
retailer
Model formulation. We developed a multivariate multilevel S/E Dummy variable representing S/E for each
model to explain how the channel usage patterns differ from category (search ¼ 1)
H/U characteristics and other factors. Consider a consumer i PurExp Prior purchases before the current
who has made a purchase at retailer j of category k on session
occasion t. This consumer can utilize M channels to gather CatExp Prior purchases related to the focal
product category before the current
necessary information throughout the customer journey. The session
channel utilization is denoted by the number of URL visits to Price Price of the purchased product (not
each channel Yijktm. Because the distribution of URL visits is available for unconverted sessions)
heavily positively skewed, we log-transformed the visits. We Age Age of a customer
add one to observations where the number of visits is zero Gender Dummy variables representing the
(Criscuolo et al. 2019). Thus, our model becomes a semilog customer’s gender (female ¼ 1)
model and the 100  slope parameters measure the percentage Income Seven-level ordinal representing income
level of the household
change of the information channel usage given a one-unit abso- HHSize Five-level ordinal variable representing
lute change of the explanatory variables. Given the time allo- the household size
cated to shopping, consumers make trade-offs between the Education Five-level ordinal variable representing
usage of different channels. To catch the potential interdepen- the education level of the customer
dencies among the six channels in our study, we let the channel Child Dummy variable for the presence of
utilization follow a multivariate normal distribution. Following children in the household (has child ¼ 1)
previous studies (e.g., Li and Kannan 2014), we also consider

1 consumer heterogeneity in channel utilization. Thus, we

The original data set before any matching, pruning, and sampling contains
13,805 consumers making 50,479 purchases on 40 websites, with develop a multilevel setting to allow every consumer to have
approximately $2.7 million sales. a unique channel usage intercept.
10 Journal of Marketing XX(X)

Level 1: we omit Cluster 1 as the base cluster and introduce two dummy
variables to the Level 1 model to account for retailer hetero-
lnðY ijktm Þ ¼ a im þ b m1 Hedonic ijk þ b m2 Utilitarian ijk geneity. Web Appendix W7 describes the clustering details, the
þ c m3 PurExp it þ c m4 CatExp ikt specific clustering assignment for each retailer, and a cluster
þ c m5 Price jk þ c m6 S= E k þ c m7 Retailer j centroid table that depicts characteristics of the three clusters.
þ e ijktm Model estimation. We estimated six models corresponding to 8–
ð1Þ 14, 2–7, and 0–1 day windows for converted and unconverted
 T  X  sessions. We conducted a Kolmogorov–Smirnov test on the
e ijkt ¼ e ijkt1 ; e ijkt2 ; . . . ; e ijktM * MVN 0; e
parameter samples obtained from converged Markov chain
Monte Carlo iterations to assess the significance of slope dif-
Level 2: ference for hedonic and utilitarian scores between converted
a im ¼ a m0 þ d m1 Demo i þ u im ð2Þ and unconverted sessions. We conducted this estimation of the
  multivariate multilevel model using a Gibbs sampler pro-
u im * N 0; s 2 : grammed in JAGS (Plummer 2003), with uninformative priors
for all parameters. To promote the efficiency of estimation and
In the Level 1 Model, aim is a random intercept that allows the ease of interpretation, we median split all the ordinal vari-
for variation in baseline channel usage across consumers. ables, including income, household size, and education. A
Hedonicijk and Utilitarianijk are grand-mean-centered hedonic robustness check on the 0–1 data found that median splitting
and utilitarian scores for a product category k purchased at a these variables does not change the sign and significance of the
retailer j, which can vary depending on consumer i’s charac- H/U effect. The final estimates are posterior means based on
teristics. We controlled for product prices Pricejk, prior pur- 40,000 Markov chain Monte Carlo iterations with a thinning
chase experiences PurExp it and CatExp ikt , and product factor of 4, after 40,000 burn-ins. To assess the convergence of
category heterogeneity S/Ek. The model for unconverted ses- the model estimates, we use three diagnostic methods, includ-
sions is estimated separately and does not have the control ing the Geweke (1992) diagnostics, the Gelman and Rubin
Pricejk. The variables Retailerj help control for the retailer (1992) diagnostics, and the effective sample size (Kass et al.
heterogeneity, such as retailers’ sales ranking (Rank), visit vol- 1998). The Geweke statistics for all the parameters are less than
ume (PurchaseCS), Alexa traffic (PageViews/User), number of 1.96, confirming that all the parameters have reached the sta-
likes on retailer’s social media page (Like), and proportional tionary posterior distributions. We run two additional chains
offerings of 20 product categories for each retailer (ProductOf- with different sets of initial values with the same number of
fering). Because a consumer in our data, on average, has five burn-ins. The potential scale reduction factors are approxi-
sessions, but retailer controls consist of 24 variables, we took mately 1.001 (<1.2) for all parameters, supporting the conver-
an alternative route to identify the Level 1 model. Specifically, gence of all three chains. The effective sample size is above
we performed K-means clustering to construct three retailer 500 for all parameters, suggesting that previous samples are not
groups and include two cluster dummies into our model. Clus- highly autocorrelated with the samples from the posterior
tering details are discussed in the “Retailer Clustering” subsec- distribution.
tion. Therefore, cm7 is a vector of parameters corresponding to
retailer cluster variables. Finally, because Pln(Yijktm) follows a Model comparison. We compare the proposed model with a
multivariate normal P distributionPMVN(m, e), the error term model that does not allow for variation in baseline channel
follows MVN(0, e), where e is a variance–covariance usage across consumers (i.e., a single-level model without uim
matrix that allows the error terms to be correlated across chan- term)
P and a univariate model (i.e., no off-diagonal elements for
nels. Due to a lack Pof prior knowledge about the correlation e), using the deviance information criterion (DIC; Spiegel-
pattern, we allow e to be unstructured to allow flexible var- halter et al. 2002) on the 0–1 window data. Our model
iance–covariance matrix. (DICproposed ¼ 236,788) is substantively better than the two
For the Level 2 model, to account for heterogeneity across benchmarking models (DIC fixed ¼ 261,370; DIC univar ¼
consumers, we assume aim * N(am, s2), where am and s2 485,641), suggesting substantial consumer heterogeneity and
measure the mean effect and dispersion of aim across consu- channel interdependencies. We also estimated a multilevel
mers, respectively. am can be further decomposed into an inter- multivariate Poisson log-normal model (El-Basyouny, Barua,
cept am0 and the effect of demographic controls Demoi, such as and Islam 2014) on the 0–1 window data and found consistent
Age, Gender, Income, Education, household size (HHSize), results.
and whether a child is present in the household (Child).

Retailer clustering. Incorporating all 24 retailer-level controls,

Results and Discussion
we performed K-means clustering to categorize the 40 retailers Tables 3 and 4 summarize the posterior means of all the para-
into a more manageable set of clusters. Using the appropriate meters for the converted and unconverted sessions with 8–14,
evaluation criteria, three clusters emerged. In the main model, 2–7, and 0–1 day windows (early, middle, and late stages,
Li et al. 11

Table 3. Retailer-Category H/U Effects on Information Channel Usage for Converted Sessions.

Variables Search Social Review Deal ProdPageTarget ProdPageCompetitor

8–14 days (early) Intercept 3.0489*** 2.7999*** .1960*** .2195*** .4172*** .2025***
Hedonic .0014 .0043*** .0006* .0005 .0017** .0011***
Utilitarian .0012 .0001 .0011** .0004 .0004 .0014***
Cluster2 .0532 .0559 .0096 .0257* .1923*** .0714***
Cluster3 .0603 .0442 .0056 .0429** .1004*** .1301***
S/E .0420 .0408 .0067 .0026 .0444** .0174*
PurExp .0027 .0085*** .0018*** .0020*** .0018 .0008
CatExp .0030 .0142*** .0020 .0012 .0113*** .0032***
Price .0002** .0002 4.4E-05 2.93E-05 .0001 7.11E-05
Age .0004 .0001 .0007** .0002 .0004 1.22E-05
Gender .0623 .1350*** .0101 .0073 .0042 .0098
Education .0561 .0337 .0020 .003 .0303 .0163
Income .0372 .0529 .0112 .0186 .0283 .0110
Size .0451 .0141 .007 .0026 .0378* .0085
Child .0899* .0653 .0228** .0029 .0319 .004
RMSE 2.0850 2.3596 .5317 .5894 1.0115 .4976
2–7 days (middle) Intercept 3.0581*** 2.7666*** .1654*** .2200*** .5738*** .2068***
Hedonic .0011 .0032** .0007** .0004 .0017** .0010***
Utilitarian .0016 .0003 .0007 .0009** .0006 .0006
Cluster2 .0552 .0694 .0013 .0399*** .0951*** .0838***
Cluster3 .0608 .0198 .0069 .023 .1393*** .1292***
S/E .0602* .0603* .0012 .0095 .0090 .0196**
PurExp .0003 .0089*** .0016*** .0012* .0026** .0013**
CatExp .0016 .0101** .0032*** .002 .0090*** .0040***
Price .0001 .0003** 2.57E-05 1.05E-05 .0002*** 2.56E-05
Age .0005 .0002 .0003 7.42E-06 .0008 3.01E-05
Gender .0453 .0993** .0069 .0012 .0218 .0089
Education .0896* .0074 .0061 .0168 .0090 .0117
Income .0564 .0790 .0011 .0043 .0004 .0085
Size .0621 .0738 .0198* .0058 .0445** .0135
Child .0850* .0982* .0282*** .0082 .0420* .0007
RMSE 2.0126 2.2993 .5053 .5615 1.0472 .4735
0–1 days (late) Intercept 2.0804*** 1.7121*** .0342*** .1910*** 1.5028*** .1270***
Hedonic .0001 .0039*** .0001 .0005** .0013* .0006***
Utilitarian .0023* .0012 .0004* .0004 .0022** .0009***
Cluster2 .0070 .0019 .0144** .1131*** .3883*** .0675***
Cluster3 .0200 .0192 .0069 .0512*** .5002*** .1089***
S/E .0362 .0462* .0028 .0019 .0201 .0178***
PurExp .0017 .0072*** .0010*** .0008* .0088*** .0007*
CatExp .0019 .0037 .0023*** .0014 .0022 .0029***
Price .0001 .0001 1.11E-05 8.45E-06 .0003*** 4.65E-05**
Age .0006 .0010 4.95E-05 .0002 .0007 1.86E-05
Gender .0417 .0904** .0015 .0045 .0257 .0034
Education .1034*** .0271 .0034 .0225*** .0025 .0102
Income .0342 .0363 .0035 .0122 .0077 .0033
Size .0651* .0569 .0054 .0055 .0276 .0094
Child .0610* .0628 .0059 .0087 .0469** .0010
RMSE 1.5849 1.7677 .2392 .3649 1.0463 .3274
*90% of the HPD interval does not contain 0.
**95% of the HPD interval does not contain 0.
***99% of the HPD interval does not contain 0.
Notes: Hedonic ¼ mean-centered hedonic score; Utilitarian ¼ mean-centered utilitarian score; RMSE ¼ root mean squared error.

hereinafter). We used highest posterior density (HPD) intervals characteristics, and demographics. In the following subsec-
to evaluate the significance of the model parameters. Consis- tions, we discuss the changes in the baseline channel usage
tent with the prior literature (e.g., Huang, Lurie, and Mitra across the customer journey. Subsequently, we discuss the
2009), we found that usage of at least some information channel-specific H/U effects using a typical hedonic product
channels differs across S/E, purchase sequence, retailer (toys) and a representative utilitarian product (office supplies).
12 Journal of Marketing XX(X)

Table 4. Retailer-Category H/U Effects on Information Channel Usage for Unconverted Sessions.

Variables Search Social Review Deal ProdPageTarget ProdPageCompetitor

8–14 days (early) Intercept 3.3920*** 3.0655*** .1679*** .2556*** .2575*** .1494***
Hedonic .0011 .001 .0004 .0002 .0022*** .0005*
Utilitarian .0012 .001 .0005 .0007 .0012 .0021***
Cluster2 .2878*** .3196*** .0768*** .0025 .5789*** .0204
Cluster3 .1228** .1828*** .0139 .0063 .0194 .0626***
S/E .0147 .0011 .0063 .0018 .0234 .0134*
PurExp .0192*** .0036 .0001 .0048*** .0323*** .0055***
CatExp .0337*** .0485*** .0045 .0063** .0375*** .0055**
Age .0019 .0002 .0004 .0002 .0003 .0005
Gender .0911** .1747*** .004 .0145 .0042 .0093
Education .0481 .0202 .008 .0028 .0110 .0190
Income .0195 .0292 .0192 .0185 .0197 .0044
Size .0420 .1431** .0057 .0217 .0172 .0091
Child .0833* .1543** .0044 .0673*** .0357 .015
RMSE 1.7080 2.1904 .6161 .6609 1.1497 .5308
2–7 days (middle) Intercept 2.9275*** 2.6771*** .1241*** .1414*** .1451*** .0748***
Hedonic .001 .0005 .0003 .0002 .0017*** .0001
Utilitarian .0023 .0032** .0002 .0013*** .001 .0017***
Cluster2 .2486*** .2742*** .0297** .0046 .4328*** .0365***
Cluster3 .1070*** .1118*** .0064 .0025 .0879*** .0540***
S/E .0114 .0131 .0106* .0097 .0049 .0152***
PurExp .0100*** .0105*** .0004 .0030** .0157*** .0011
CatExp .0280*** .0268*** .002 .0028 .0327*** .0028
Age .0004 .0002 .0003 .0001 .0003 .0002
Gender .1520*** .0551** .0316 .0045 .0196 .0015
Education .1955*** .0241 .0377 .0216 .0385 .0119
Income .013 .0476 .0079 .0108 .0257 .0057
Size .0042 .0246 .0019 .0005 .0162 .0107
Child .0022 .0103 .0194 .0161 .0216 .0013
RMSE 1.6488 2.1038 .4697 .5086 .9949 .4044
0–1 days (late) Intercept 2.3640*** 1.9460*** .0623*** .1132*** 1.6675*** .1569***
Hedonic .0005 .0025*** .0002 .0002 .0005 .0010***
Utilitarian .0008 .0010 .0009*** .0003 .0025*** .0001
Cluster2 .2212*** .2593*** .0217*** .0170* .0716*** .0635***
Cluster3 .1163** .0934 .0216* .0026 .1985*** .1073***
S/E .0016 .0279 .0058 .0022 .0462*** .0220***
PurExp .0029 .0133*** .0002 .0022*** .0131*** .0033***
CatExp .0125** .0118* .002 .0049*** .0159*** .0086***
Age .0007 .003 3.74E-05 .0003 .0007 .0001
Gender .0450 .1198*** .0061 .0071 .0057 .0046
Education .0085 .0356 .0008 .0205 .0374 .0014
Income .0023 .0328 .0054 .0018 .0077 .0085
Size .0307 .0708 .0010 .0039 .0064 .0029
Child .0656* .1101** .0015 .0192** .0043 .0028
RMSE 1.3285 1.7685 .3033 .3521 .7641 .3713
*90% of the HPD interval does not contain 0.
**95% of the HPD interval does not contain 0.
***99% of the HPD interval does not contain 0.
Notes: Hedonic ¼ mean-centered hedonic score; utilitarian ¼ mean-centered utilitarian score; RMSE ¼ root mean squared error.

Dynamic Channel Usage Across the Customer Journey Neslin 2011) showing that consumers utilize channels at differ-
ing intensities throughout the customer journey. Specifically,
We derived the average daily number of visits in each of the
all six channels’ usage increases toward the final purchase,
purchase windows (i.e., the intercepts from early, middle,
validating that they are important information search channels
and late-stage models) for the six digital channels, as presented
for online shopping. From the middle to the late stage of the
in Figure 3. The daily channel usage (intercept) varies signif-
journey, the increase in the rate of product page views on the
icantly across different time windows, confirming prior studies
target retailer’s website (ProdPage_Target) is greater than that
(Johnson et al. 2004; Moe 2006; Valentini, Montaguti, and
of product page views on competing websites
Li et al. 13

.39% increase in social media usage. In general, we find that a

10.00
1% increase in the Hedonic score involves a greater usage of
social media (early: .43%, middle: .32%, late: .39%) and more
product page views on the target retailers’ site (early: .17%,
middle: .17%). Furthermore, a 1% increase in the Utilitarian
score is associated with heavy usage of search engines (late:
1.00
.23%), third-party reviews (early: .11%, late: .04%), and deal
Early Middle Late sites (middle: .09%).
In contrast to the converted sessions, the H/U effects are
different for the unconverted sessions depicted in Table 4. For
example, the Hedonic score’s effect on social media is attenu-
ated (unconv_late: .25% vs. conv_late: .43%; p < .05) based on
.10 a Kolmogorov–Smirnov test, and the signs for deal site usage
Search Social (unconv_mid: .13% vs. conv_mid: .09%) and product page
Review Deal views on the target retailers (unconv_early: .22% vs. con-
ProdPage_Target ProdPage_Competitor v_early: .17%; unconv_mid: .17% vs. conv_mid: .17%) are
flipped, suggesting that the H/U effects could be potentially
Figure 3. Average daily channel usage (intercept) for three stages of related to conversion. To compare the differential H/U effects
the customer journey. across different stages of customer journeys for converted and
Notes: We calculate the average daily channel usage by exponentiating the unconverted sessions, we selected a typical hedonic prod-
intercepts of the early- (8–14 days), middle- (2–7 days), and late- (0–1 day)
uct—toys (Hedonic ¼ 10.82; Utilitarian ¼ 5.15)—and a
stage models divided by the number of days in each time window. The y-axis is
in logarithmic scale. The x-axis depicts the early, middle, and late stages of the typical utilitarian product—office supplies (Hedonic ¼
customer journey. 15.92; Utilitarian ¼ 6.55)—and visualize their percentage
change in channel usage relative to a product with the average
(ProdPage_Competitor), suggesting a consumer lock-in, con- H/U scores (H ¼ 59.86; U ¼ 66.35)2 in Figure 4, Panels A–D.
sistent with prior findings that consumers gradually narrow Drawing on this selection, we next discuss the important
their consideration set throughout the customer journey (John- channels for hedonic and utilitarian purchases at different
son et al. 2004; Moe 2006). Finally, the covariance among stages of the customer journey. Table 5 presents a summary
channel usage is all significant, verifying considerable interde- of these results.
pendency among channel usage. Specifically, for converted
sessions, search engine and social media usage are highly cor-
related with each other (e.g., correlation ¼ .54 for the late Important Channels for Hedonic Purchases
stage), representing high co-usage of these two information
channels. In addition, toward the end of the purchase, product For converted sessions, toys (hedonic) purchases utilize a
page views on target websites are the most correlated with rest greater level of social media than office supplies (utilitarian)
of other channels (average correlation ¼ .21) but product page purchases, with an average of 10% more channel utilization
views on competing websites are the least correlated (average throughout the purchase cycle (early toys: 4.65% vs. early
correlation ¼ .09), indicating a funneling effect of the infor- office: 6.85%; middle toys: 3.46% vs. middle office:
mation search on target websites. 5.09%; late toys: 4.22% vs. late office: 6.21%). Furthermore,
the hedonic effect for social media uses increases slightly
toward the end of the customer journey (middle toys: 3.46%
H/U Effect on Channel Usage vs. late toys: 4.22%), indicating that people may utilize the
social media channel more when they have a relatively clear
For the converted sessions described in Table 3, the utilization purchase intention. This finding nicely complements the social
of six information channels varies significantly with the H/U media literature (Kim and Ko 2012; Malhotra, Malhotra, and
characteristics of purchases, and across different stages. See 2012; Naylor, Lamberton, and West 2012) by showing
Because the H/U score follows a 0–100 scale, one unit change that, in addition to soliciting impulse buying, social media
of the H/U score on a 100-point scale would equal to a 1% might have become an information channel that consumers use
change of the H/U score. In our semilog model setting, given
to proactively search for information (e.g., finding product
that b is the parameter for the H/U score, a 1% change of H/U
pictures on Instagram). However, the positive hedonic effect
score, conditional on a focal consumer’s other purchase char-
on social media use is significantly smaller for unconverted
acteristics, will correspond to a 100  b percentage change of
sessions than for converted sessions (p < .05), suggesting that
information channel usage. For instance, the significant para-
meter .0039 (p < .01) of the Hedonic score on social media
during the late (0–1 day) window indicates that a 1% increase 2
The Hedonic and Utilitarian scores are mean-centered scores; the H and U
in the Hedonic score in the late stage of the journey leads to a scores are original scores.
14 Journal of Marketing XX(X)

A: Toys (Converted) B: Office (Converted)

5%
5%

3%
3%

1%
1%

−1% Late Middle Early

-1% Early Middle Late

−3%
-3%

−5%
-5%

−7%
-7%
Search Social
Search Social
Review Deal
Review Deal
ProdPage_Target ProdPage_Competitor
ProdPage_Target ProdPage_Competitor

C: Toys (Unconverted) D: Office (Unconverted)

6% 6%

4% 4%

2% 2%

0% 0%
Early Middle Lat e Late Middle Early
−2% −2%

−4% −4%

−6% −6%

−8% −8%

Search Social Search Social

Review Deal Review Deal
ProdPage_Target ProdPage_Competitor ProdPage_Target ProdPage_Competitor

Figure 4. Channel usage percentage differences for hedonic and utilitarian purchases.
Notes: Hedonic and Utilitarian are mean-centered H/U scores; H and U are the original H/U scores. Toys represent a typical hedonic product (Hedonic ¼ 10.82;
Utilitarian ¼ 5.15); Office represents a typical utilitarian product (Hedonic ¼ 15.92; Utilitarian ¼ 6.55). For all panels, the y-axis reflects the percentage change
of channel usage for toys and office relative to a product with mean original H/U scores (H ¼ 59.86; U ¼ 66.35). The x-axis depicts the early (8–14 days), middle
(2–7 days), and late (0–1 day) stages of a 14-day customer journey.

Table 5. Summary of Results: Information Channels Utilized at Different Stages of the Customer Journey Vary by Product Characteristics and
Purchase Conversion.

Hedonic Product Utilitarian Product

Stage of
Customer Journey Purchases Nonpurchases Purchases Nonpurchases

Early (8–14 days) Social Media Reviews

ProdPage_Target
Middle (2–7 days) Social Media Social Media Reviews
ProdPage_Target Deals Deals
ProdPage_Other
Late (0–1 days) Social Media Social Media Search Engine Reviews
Deals Deals
ProdPage_Other
Li et al. 15

a greater level of social media usage might be more useful for investigate this guilt-justification mechanism in the customer
realizing hedonic purchases. journey.
These findings reinforce the importance of emotive and
social aspects during the hedonic shopping process highlighted Important Channels for Utilitarian Purchases
by prior research in the H/U domain (Arnold and Reynolds
2003; Novak, Hoffman, and Duhachek 2003) and support the As illustrated in Figure 4, Panel B, consumers making utilitar-
notions that social media is more effective for viral marketing ian purchases such as office supplies utilize more third-party
of hedonic products (Berger and Schwartz 2011) and that reviews (early office: 1.68% vs. early toys: 1.21%; middle
online social connections are more influential for hedonic office: 1.11% vs. middle toys: .76%) at the beginning and the
spending (Park et al. 2018). Furthermore, given that social middle of the customer journey. They also visit product pages
media is utilized by hedonic purchases for both converted and on competing retailers more often (middle office: 1.59% vs.
unconverted sessions (with different effect sizes), social media middle toys: 1.08%) in the middle of the journey. In addition,
marketing might be more effective to reach potential consu- consumers demonstrate greater usage of search engines3 (late
mers and improve conversion. office: 1.51% vs. late toys: 1.18%) and deal sites (middle
It is important to note that our results regarding social media office: .59% vs. middle toys: .46%; late office: .80% vs. late
use are correlational in nature, and causal inferences should be toys: .54%) toward the end of the customer journey.4
made with caution. Because of privacy considerations, we can- These findings are consistent with the cognitive mechan-
not see what people are browsing on social media. While dif- isms discussed in the H/U literature (Novak, Hoffman, and
ferences in social media usage across products and over time Duhachek 2003). Utilitarian purchases are often rational and
give us some confidence in making prescriptive recommenda- goal-driven, with the objective of optimizing the purchase deci-
tions, and these results are reinforced by other studies (Colicev sion. With more tangible and well-defined utilitarian attributes,
et al. 2018; Colicev, Kumar, and O’Connor 2019; Hughes, information channels that facilitate flexible and direct search
Swaminathan, and Brooks 2019), the aforementioned caveat and allow for convenient comparisons among alternatives, such
remains. as search engines and third-party review sites, might be partic-
In addition, hedonic purchases involve more product page ularly useful. In addition, the less extensive differentiation
views on the target retailers up to two weeks before the con- associated with utilitarian products makes consumers easier
version, with as much as a 4.55% difference between toys and to benchmark across retailers (Noble, Griffith, and Weinberger
office supplies purchases (early toys: 1.84% vs. early office: 2005). Thus, utilitarian purchases involve more product page
2.71%), presenting a funneling effect toward to the final views on competing retailers than hedonic purchases.
purchases. However, this effect is reversed for unconverted For the unconverted sessions illustrated in Figure 4, Panel
sessions, (early toys: 2.38% vs. early office: 3.50%), suggest- D, the patterns observed for converted sessions are attenuated
ing that consumers seeking hedonic products might browse at the early and middle stages of the customer journey. In fact,
more product pages on competing retailers and make purchases we find that the overall late-stage channel usage patterns for the
there, leading to nonconversion on the focal retailer site. This unconverted sessions are more like the early-stage channel
finding reveals that sufficient on-site product page views up to usage patterns of converted sessions. Thus, analyzing channel
two weeks before conversion are crucial for realizing hedonic usage in sessions that have not led to purchase is important and
purchases, in support of the notion that consumers with hedonic could be viewed as information search in the early stage of a
purchases exhibit a greater level of “affective attachment” shopping funnel. In summary, we found that search engines,
(Chaudhuri and Holbrook 2001) toward retailer brands and are reviews, deal sites, and competing retailers’ product pages are
less likely to engage in a brand-switching behavior (Kim and important for utilitarian purchases, and we speculate that the
Ko 2012). nonconversion related to utilitarian purchases might be due to
Finally, while deal sites are utilized less for hedonic pur- insufficient information search and alternative comparisons.
chases than for utilitarian purchases toward the end of the
customer journey for converted sessions (middle toys: .46% General Discussion
vs. middle office: .59%; late toys: .54% vs. late office: .80%),
this effect is reversed for unconverted sessions (middle toys: Theoretical Contributions
.67% vs. middle office: .85%). Prior research has suggested This study makes several important contributions to the estab-
that consumers buying hedonic products can engage in guilt- lished literature on hedonic and utilitarian consumption, the
justification behavior (e.g., Khan and Dhar 2010; Okada 2005)
by looking for deals. Were consumers doing so during the
3
middle stage of the customer journey in the unconverted ses- Search engine usage for utilitarian purchases is marginally significant,
sions? Because our data set does not contain information about potentially indicating that with increasing experiential features and
the types of deals consumers viewed, we are unable to inves- ever-improving search engine marketing (SEM) strategies, the importance of
search engines for hedonic purchases is rising, thus shrinking the H/U effect.
tigate the exact association between the deal visits and non- 4
Similar to social media, privacy considerations prevent us from observing
conversion (e.g., is this nonconversion due to unsatisfying what people are searching for on search engines. As a result, these attributions
discount level?). Future research with more granular data could of search effects are correlational in nature.
16 Journal of Marketing XX(X)

emerging research on customer journey, and the nascent liter- Finally, we contribute to the growing literature of Big Data
ature on Big Data marketing. First, we extend the prior cus- marketing (Bradlow et al. 2017; Kitchens et al. 2018; Sudhir
tomer journey research on online information channel usage 2016; Wedel and Kannan 2016) by demonstrating the great
(e.g., Kushwaha and Shankar 2013; Li and Kannan 2014; potential unlocked by “Big Data” through a multidata, multi-
Neslin and Shankar 2009) by introducing a social/psychologi- method approach. By combining primary and secondary data,
cal angle. In addition to the utility-centric perspective of pre- we illustrate how primary data remain an important comple-
purchase information channels, we find that affective ment to large-scale clickstream data by providing critical per-
mechanisms such as pleasure seeking and affective attachment ceptual enrichment. To integrate these data sources and harness
documented in the H/U literature (Arnold and Reynolds 2003; the rich insights embedded in terabytes of data, we employ
Chaudhuri and Holbrook 2001) have significant implications survey analysis, text mining, machine learning, and Bayesian
for information channel usage (Lamberton and Stephen 2016). modeling. Our multivariate multilevel model carefully consid-
For example, we find that hedonic purchases (e.g., toy prod- ers the channel interdependency as well as the customer, retai-
ucts) utilize social media up to 10% more than their utilitarian ler, and product heterogeneity through a hierarchical Bayesian
counterparts (e.g., office supplies) throughout the purchase approach, providing a viable framework for future research on
cycle and are less likely to switch retailer brands by browsing customer journey. As a result, the study covers six channels and
on competing retailers’ websites. Thus, we employ new angles 20 product categories sold on 40 top internet retailers with 115
to study information channel choices related to online pur- retailer-category combinations, with over $1 million in sales in
chases and extend the H/U literature by testing its predictions a two-year period. To the best of our knowledge, this is the first
in the customer journey context (Batra and Keller 2016). study that incorporates such a comprehensive set of channels,
Second, our study provides a more nuanced view of the product categories, and retailers in the customer journey liter-
dynamic channel usage patterns during the paths to purchase. ature. Future Big Data marketing research could use our
Our results highlight the importance of considering the tem- approach as a framework to integrate multiple data sources
poral effect in the customer journey literature. By examining encompassing both primary and large-scale secondary data and
the early, middle, and late stages of the customer journey, we derive “big” insights accordingly.
can derive more actionable insights. For example, we find that
product page views on the transacting retailers are different
between hedonic and utilitarian purchases up to two weeks Managerial Implications
before the conversion, and that deal sites are visited by con- Our results have several actionable implications for marketing
sumers with utilitarian purchases one week before the final managers. By identifying varying H/U effects on six informa-
purchase. These nuanced findings indicate that when deploying tion channels across customer journey for a total of 115
marketing-mix, advertising, and promotion strategies, manag- retailer-category combinations, our model offers a more prin-
ers might want to leave a longer window for these strategies to cipled, theoretically driven inductive approach for tailoring
be effective. In summary, we contribute to the customer jour- marketing strategies to consumers’ shopping needs throughout
ney literature (e.g., Johnson et al. 2004; Li and Kannan 2014) their journey. In the following, we discuss general marketing
by demonstrating the necessity of considering the temporal strategies as well as Black Friday (and Cyber Monday) mar-
dimensions in studying the purchase funnel and the customer keting ideas for retailers selling hedonic and utilitarian prod-
journey (Lemon and Verhoef 2016). ucts. Multicategory retailers can customize their marketing
Third, we provide an actionable framework for incorporat- strategies based on their product types accordingly.
ing H/U scales in the online purchasing context. Unlike previ- First, for retailers selling hedonic products such as toys, we
ous studies that have operationalized the H/U characteristics at provide two actionable insights: (1) embrace social media and
the product-category level, we provide a more nuanced retailer- (2) monitor on-site product page views. Our study shows that
category vantage point. From our survey, we found that similar social media is being used extensively throughout the customer
product categories sold at different retailers (e.g., electronics at journey and is increasingly becoming a channel for proactive
Home Depot vs. Amazon) receive different H/U scores due to information search (eMarketer 2017). Therefore, marketing
the characteristics of the retailer brands. Managers can leverage managers should consistently invest in social media marketing
our scales to understand their H/U positions in relation to their to entice more consumers to visit their websites. In addition, we
competitors (see Figure 1 as an example for Home Depot vs. find that there is a potential guilt-justification need for consu-
Amazon). Subsequently, these retailer-category-level H/U mers who failed to complete hedonic purchases. Because social
scores could be plugged into our multivariate multilevel model media is extensively used at the beginning of the journey,
to identify effective touchpoints among different product cate- retailers could deploy social coupons with features that serve
gories (as shown in our toys vs. office example in Figure 4) or both the experiential and justification needs of hedonic pur-
across different retailers. Therefore, we contribute to the H/U chases (Kumar and Rajan 2012). Furthermore, we find that
literature by highlighting the necessity of considering retailer on-site product pages are leveraged extensively at the begin-
brand differences in the product category’s H/U perceptions ning of the journey and start to reduce one week before the
and bringing the H/U scales to the context of touchpoint man- purchase. Given the affective nature of hedonic purchases,
agement (Kannan and Li 2017). retailers should constantly improve the experiential features
Li et al. 17

of the product pages on their sites to convert more hedonic thereby lacking explicit URL patterns to help discern when a
purchases. Moreover, retailers can monitor their page views consumer has read a review. These limitations make it difficult
and reach out to heavy browsers with promotions with a longer to make unique causal attributions from social media, search
redemption time (e.g., two weeks). engine, and review site browsing to product purchases.
Second, for retailers selling utilitarian products such as Second, although we use three time-windows to capture the
office supplies, we offer two prescriptions: (1) benchmark price dynamic channel usage effect at the early, middle, and late
and product and (2) prioritize search engine marketing (SEM). stages of the customer journey (8–14, 2–7, and 0–1 days,
Our study shows that consumers tend to optimize their utilitar- respectively), we do not consider the order of channel usage
ian purchase by visiting third-party review sites, exploring deal within each time window. The temporal proximity of channel
sites, and browsing product pages on competing retailers’ sites. usage might affect the final purchase. While acknowledging
Therefore, retailers should employ price and product bench- this limitation, our goal was to highlight the interplay between
mark analysis to understand whether their price is above or retailer-category-level H/U scores and online channel usage
below the market price and what potential customers see and leading up to consumer purchases. We believe the work on
experience when searching for similar products. Given the rise multichannel attribution models (e.g., Li and Kannan 2014)
of competitive intelligence, managers could invest more in represents an important related body of literature that could
automated benchmarking tools to monitor, listen, and analyze be incorporated into future studies.
the key competitive metrics (e.g., price, live deals, Yelp Similarly, we did not analyze search depth, which is often
reviews) in real time. In addition, we find that consumers mak- represented as total time spent on each URL. Search depth
ing utilitarian purchases tend to use search engines more could be important for the H/U effect, as suggested by prior
toward the end of the journey. Because search engine optimi- studies (Kushwaha and Shankar 2013; Okada 2005). Moreover,
zation is more powerful in driving organic traffic at the top of we did not employ category-specific cycle length because we
the funnel, and SEM is more effective in driving conversions at could not differentiate product-based searches from baseline
the bottom of the funnel, retailers should prioritize SEM over channel usage for certain channels. Future studies with detailed
search engine optimization. In addition, they should choose search log and clickstream data could solve this problem.
paid keywords that are more related to product features and Finally, due to data limitations, we did not analyze other
benefits, provided that utilitarian purchases usually involve important information channels. These include email, referrals,
more product comparisons. television (see, e.g., De Haan, Weisel and Pauwels 2016; De
Finally, our dynamic view of channel usage across customer Vries, Gensler, and Leeflang 2017; Srinivasan, Rutz, and Pau-
journey could offer specific guidelines for a Black Friday (and wels 2016) and visits to brick-and-mortar stores (showroom-
Cyber Monday) marketing strategy. Retailers selling hedonic ing). Given the increasing importance of social media
products could market their promotional content on social (eMarketer 2017; Hewett et al. 2016), further research could
media and send reminder emails inviting on-site traffic two investigate differential effects of various types of social media
weeks before Black Friday, when their customers start to during the customer journey: examples include firm-generated
engage in social media and on-site product pages. Retailers content (Colicev et al. 2018), social influencers (Hughes, Swa-
selling utilitarian products could extend their sales because minathan, and Brooks 2019), and the consumers’ individual
consumers start to visit deal sites one week before they make expression on social media (Hollebeek and Macky 2019; Lam-
purchases. In addition, they could optimize their SEM strategy berton and Stephen 2016). In a similar vein, because our data
during Black Friday or Cyber Monday to enhance the conver- set is restricted to desktop clickstream data, we did not observe
sion rate. purchases that occurred offline or through mobile sites. With
the surge in mobile usage, cross-device channel usage could
constitute an increasingly important future direction of
Limitations and Future Directions research (e.g., De Haan et al. 2018).
Admittedly, this study has several limitations that future work
could address. First, we examined observed channel usage
behaviors mostly at the URL level. We do not have data on
Conclusion
the specific types of information searches that consumers per- Our results show that consumers’ utilization of various path-to-
formed—for example, seeing what a friend “liked” on a social purchase channels differs across the retailer-category hedonic
networking site, as well as the content about the products avail- and utilitarian characteristics of purchased products. Specifi-
able at these channels. Although such data are difficult to col- cally, consumers making hedonic purchases seek fun, enjoy-
lect and raise significant privacy concerns, analyzing them ment, and pleasure in their shopping process; prefer social
could offer additional insights regarding the H/U effect on media; and are more likely to browse product pages on the
specific types of touchpoints as well as how this effect interacts target retailers’ website. By contrast, consumers making utili-
with product availability. In a similar vein, although we track tarian purchases prefer channels that facilitate convenient and
usage of review sites, we do not have measures of consumers’ efficient search across alternatives. Therefore, they prefer
use of product reviews within a retailer’s site because on-site leveraging search engines, reading more reviews on the third-
product reviews are often embedded within the product pages, party review sites, comparing prices on deal sites, and browsing
18 Journal of Marketing XX(X)

more product pages on competing retailers’ websites than Babin, Barry J., William R. Darden, and Mitch Griffin (1994), “Work
hedonic purchasers. and/or Fun: Measuring Hedonic and Utilitarian Shopping Value,”
Because channel usage changes dynamically throughout the Journal of Consumer Research, 20 (4), 644–56.
customer journey, the H/U effect also varies. Specifically, for Bagozzi, Richard P. (1975), “Marketing as Exchange,” Journal of
hedonic purchases, social media is used as early as two weeks Marketing, 39 (4), 32–39.
before the final purchase. Conversely, for utilitarian purchases, Bart, Yakov, Venkatesh Shankar, Fareena Sultan, and Glen L. Urban
third-party review sites are engaged two weeks before the final (2005), “Are the Drivers and Role of Online Trust the Same for All
purchase, and this effect is attenuated toward the purchase day. Web Sites and Consumers? A Large-Scale Exploratory Empirical
Search engines and deal sites are utilized to a greater extent Study,” Journal of Marketing, 69 (4), 133–52.
closer to the day of a utilitarian purchase. We also find that the Batra, Rajeev and Olli T. Ahtola (1991), “Measuring the Hedonic and
H/U effect on product page views decreases over time, suggest- Utilitarian Sources of Consumer Attitudes,” Marketing Letters, 2
ing a consideration set narrowing process. (2), 159–70.
The analysis of unconverted sessions demonstrates a differ- Batra, Rajeev and Kevin Lane Keller (2016), “Integrating Marketing
ent H/U effect. For hedonic purchases, social media is only Communications: New Findings, New Lessons, and New Ideas,”
used by hedonic purchases closer to the end of the journey. Journal of Marketing, 80 (6), 122–45.
Deal site visits and product page views on competing retailers’ Berger, Jonah and Eric M. Schwartz (2011), “What Drives Immediate
sites increase, indicating a possible guilt-justification demand and Ongoing Word of Mouth?” Journal of Marketing Research, 48
commonly shown in hedonic consumption. Conversely, chan- (5), 869–80.
nels used by consumers making utilitarian purchases are not Bradlow, Eric T., Manish Gangwar, Praveen Kopalle, and Sudhir
employed at the same level in the unconverted sessions, indi- Voleti (2017), “The Role of Big Data and Predictive Analytics in
cating that the nonconversion might be due to insufficient Retailing,” Journal of Retailing, 93 (1), 79–95.
information search. Chaudhuri, Arjun and Morris B. Holbrook (2001), “The Chain of
As digital marketing and monitoring spending continue to Effects from Brand Trust and Brand Affect to Brand Performance:
grow, our findings provide important implications for market- The Role of Brand Loyalty,” Journal of Marketing, 65 (2), 81–93.
ing managers who want to better allocate resources across Chevalier, Judith A. and Dina Mayzlin (2006), “The Effect of Word of
digital channels. We also believe this study constitutes an Mouth on Sales: Online Book Reviews,” Journal of Marketing
important step toward examining the interplay between hedo-
Research, 43 (3), 345–54.
nic and utilitarian characteristics of online purchases and their
Chiang, Kuan-Pin and Ruby R. Dholakia (2003) “Factors Driving
implications for digital path-to-purchase channels—a direction
Consumer Intention to Shop Online: an Empirical Investigation,”
on which we hope future research can continue to build.
Journal of Consumer Psychology 13 (1), 177–83.
Childers, Terry L., Christopher L. Carr, Joann Peck, and Stephen
Acknowledgments
Carson (2002), “Hedonic and Utilitarian Motivations for Online
The authors thank Raj Venkatesan, Zijun Ke, and Chenhui Liu for Retail Shopping Behavior,” Journal of Retailing, 77 (4), 511–35.
helpful comments. Research support from the Center for Business Colicev, Anatoli, Ashish Kumar, and Peter O’Connor (2019),
Analytics at the McIntire School of Commerce is gratefully
“Modeling the Relationship Between Firm and User Generated
acknowledged.
Content and the Stages of the Marketing Funnel,” International
Journal of Research in Marketing, 36 (1), 100–16.
Associate Editor
Colicev, Anatoli, Ashwin Malshe, Koen Pauwels, and Peter O’Connor
P.K. Kannan
(2018), “Improving Consumer Mindset Metrics and Shareholder
Value Through Social Media: The Different Roles of Owned and
Declaration of Conflicting Interests Earned Media,” Journal of Marketing, 82 (1), 37–56.
The author(s) declared no potential conflicts of interest with respect to Criscuolo, Chiara, Ralf Martin, Henry G. Overman, and John Van
the research, authorship, and/or publication of this article. Reenen (2019), “Some Causal Effects of an Industrial Policy,”
American Economic Review, 109 (1), 48–85.
Funding De Haan, Evert, P.K. Kannan, Peter C. Verhoef, and Thorsten Wiesel
The author(s) received no financial support for the research, author- (2018), “Device Switching in Online Purchasing: Examining the
ship, and/or publication of this article. Strategic Contingencies,” Journal of Marketing, 82 (5), 1–19.
De Haan, Evert, Thorsten Wiesel, and Koen Pauwels (2016), “The
References Effectiveness of Different Forms of Online Advertising for Pur-
Anderl, Eva, Jan Hendrik Schumann, and Werner Kunz (2016), chase Conversion in a Multiple-Channel Attribution Framework,”
“Helping Firms Reduce Complexity in Multichannel Online Data: International Journal of Research in Marketing, 33 (3), 491–507.
A New Taxonomy-Based Approach for Customer Journeys,” Jour- De Los Santos, Babur, Ali Hortaçsu, and Matthijs R. Wildenbeest
nal of Retailing, 92 (2), 185–203. (2012), “Testing Models of Consumer Search Using Data on Web
Arnold, Mark J. and Kristy E. Reynolds (2003), “Hedonic Shopping Browsing and Purchasing Behavior,” American Economic Review,
Motivations,” Journal of Retailing, 79 (2), 77–95. 102 (6), 2955–80.
Li et al. 19

De Vries, Lisette, Sonja Gensler, and Peter S.H. Leeflang (2017), Kannan, P.K. and Hongshuang A. Li (2017), “Digital Marketing: A
“Effects of Traditional Advertising and Social Messages on Framework, Review and Research Agenda,” International Journal
Brand-Building Metrics and Customer Acquisition,” Journal of of Research in Marketing, 34 (1), 22–45.
Marketing, 81 (5), 1–15. Kass, Robert E., Bradley P. Carlin, Andrew Gelman, and Radford M.
El-Basyouny, Karim, Sudip Barua, and Md Tazul Islam (2014), Neal (1998), “Markov Chain Monte Carlo in Practice: A Round-
“Investigation of Time and Weather Effects on Crash Types Using table Discussion,” American Statistician, 52 (2), 93–100.
Full Bayesian Multivariate Poisson Lognormal Models,” Accident Khan, Uzma and Ravi Dhar (2010), “Price-Framing Effects on the
Analysis and Prevention, 73, 91–99. Purchase of Hedonic and Utilitarian Bundles,” Journal of Market-
eMarketer (2017), “Social Commerce 2018: Its Influence in the Path ing Research, 47 (6), 1090–99.
to Purchase” (December 18), https://www.emarketer.com/Report/ Khan, Uzma, Ravi Dhar, and Klaus Wertenbroch (2005), “A Beha-
Social-Commerce-2018-Its-Influence-Path-Purchase/2002175. vioral Decision Theory Perspective on Hedonic and Utilitarian
Frambach, Ruud T., Henk C.A. Roest, and Trichy V. Krishnan Choice,” in Inside Consumption: Consumer Motives, Goals, and
(2007), “The Impact of Consumer Internet Experience on Chan- Desires, S. Ratneshwar and David Glen Mick, eds. New York:
nel Preference and Usage Intentions Across the Different Stages Routledge, 144–65.
of the Buying Process,” Journal of Interactive Marketing, 21 (2), Kim, Angella J. and Eunju Ko (2012), “Do Social Media Marketing
26–41. Activities Enhance Customer Equity? An Empirical Study of Luxury
Gelman, Andrew and Donald B. Rubin (1992), “Inference from Itera- Fashion Brand,” Journal of Business Research, 65 (10), 1480–86.
tive Simulation Using Multiple Sequences,” Statistical Science, 7 Kim, Junghyun and Robert LaRose (2004), “Interactive e-Commerce:
(4), 457–72. Promoting Consumer Efficiency or Impulsivity?” Journal of Com-
Geweke, John (1992), “Evaluating the Accuracy of Sampling-Based puter-Mediated Communication, 10 (1), https://academic.oup.
Approaches to the Calculation of Posterior Moments,” in Bayesian com/jcmc/article/10/1/JCMC10112/4614485.
Kitchens, Brent, David Dobolyi, Jingjing Li, and Ahmed Abbasi
Statistics. New York: Oxford University Press, 169–93.
(2018). “Advanced Customer Analytics: Strategic Value through
Ghose, Anindya and Sha Yang (2009), “An Empirical Analysis of
Integration of Relationship-Oriented Big Data,” Journal of Man-
Search Engine Advertising: Sponsored Search in Electronic Mar-
agement Information Systems, 35 (2), 540–74.
kets,” Management Science, 55 (10), 1605–22.
Kivetz, Ran and Itamar Simonson (2002), “Earning the Right to
Heitz-Spahn, Sandrine (2013), “Cross-Channel Free-Riding Con-
Indulge: Effort as a Determinant of Customer Preferences Toward
sumer Behavior in a Multichannel Environment: An Investigation
Frequency Program Rewards,” Journal of Marketing Research, 39
of Shopping Motives, Sociodemographics and Product
(2), 155–70.
Categories,” Journal of Retailing and Consumer Services, 20 (6),
Kumar, Ashish, Ram Bezawada, Rishika Rishika, Ramkumar
570–78.
Janakiraman, and P.K. Kannan (2016), “From Social to Sale: The
Hewett, Kelly, William Rand, Roland T. Rust, and Harald J. van
Effects of Firm Generated Content in Social Media on Customer
Heerde (2016), “Brand Buzz in the Echoverse,” Journal of Mar-
Behavior,” Journal of Marketing, 80 (1), 7–25.
keting, 80 (3), 1–24.
Kumar, V. and Bharath Rajan (2012), “Social Coupons as a Marketing
Holbrook, Morris B. and Elizabeth C. Hirschman (1982), “The
Strategy: A Multifaceted Perspective,” Journal of the Academy of
Experiential Aspects of Consumption: Consumer Fantasies, Feel- Marketing Science, 40 (1), 120–36.
ings, and Fun,” Journal of Consumer Research, 9 (2), 132–40. Kushwaha, Tarun and Venkatesh Shankar (2013), “Are Multichannel
Hollebeek, Linda D. and Keith Macky (2019), “Digital Content Mar- Customers Really More Valuable? The Moderating Role of Prod-
keting’s Role in Fostering Consumer Engagement, Trust, and uct Category Characteristics,” Journal of Marketing, 77 (4), 67–85.
Value: Framework, Fundamental Propositions, and Implications,” Lamberton, Cait and Andrew T. Stephen (2016), “A Thematic Explo-
Journal of Interactive Marketing, 45, 27–41. ration of Digital, Social Media, and Mobile Marketing: Research
Huang, Peng, Nicholas H. Lurie, and Sabyasachi Mitra (2009), Evolution from 2000 to 2015 and an Agenda for Future Inquiry,”
“Searching for Experience on the Web: An Empirical Examination Journal of Marketing, 80 (6), 146–72.
of Consumer Behavior for Search and Experience Goods,” Journal Lemon, Katherine N. and Peter C. Verhoef (2016), “Understanding
of Marketing, 73 (2), 55–69. Customer Experience Throughout the Customer Journey,” Journal
Hughes, Christian, Vanitha Swaminathan, and Gillian Brooks (2019), of Marketing, 80 (6), 69–96.
“Driving Brand Engagement Through Online Social Influencers: Li, Hongshuang and P.K. Kannan (2014), “Attributing Conversions in
An Empirical Investigation of Sponsored Blogging Campaigns,” a Multichannel Online Marketing Environment: An Empirical
Journal of Marketing, 83 (5), 78–96. Model and a Field Experiment,” Journal of Marketing Research,
Inman, J. Jeffrey, Venkatesh Shankar, and Rosellina Ferraro (2004), 51 (1), 40–56.
“The Roles of Channel-Category Associations and Geodemo- Lin, Kuan-Yu and Hsi-Peng Lu (2015), “Predicting Mobile Social
graphics in Channel Patronage,” Journal of Marketing, 68 (2), Network Acceptance Based on Mobile Value and Social Influ-
51–71. ence,” Internet Research, 25 (1), 107–30.
Johnson, Eric J, Wendy W. Moe, Peter S. Fader, Steven Bellman, and Liu, Fei, Eric T.K. Lim, Hongxiu Li, Chee-Wee Tan, and Dianne Cyr
Gerald L. Lohse (2004), “On the Depth and Dynamics of Online (2019), “Disentangling Utilitarian and Hedonic Consumption
Search Behavior,” Management Science, 50 (3), 299–308. Behavior in Online Shopping: An Expectation Disconfirmation
20 Journal of Marketing XX(X)

Perspective,” Information & Management (published online Plummer, Martyn (2003), “JAGS: A Program for Analysis of Baye-
August 31), DOI: 10.1016/j.im.2019.103199. sian Graphical Models Using Gibbs Sampling,” Proceedings of the
Malhotra, Arvind, Claudia Kubowicz Malhotra, and Alan See (2012), 3rd International Workshop on Distributed Statistical Computing,
“How to Get Your Messages Retweeted,” MIT Sloan Management 125 (125), 1–10.
Review, 53 (2), 61. Rishika, Rishika, Ashish Kumar, Ramkumar Janakiraman, and Ram
Mallapragada, Girish, Sandeep R. Chandukala, and Qing Liu (2016), Bezawada (2013), “The Effect of Customers’ Social Media Partic-
“Exploring the Effects of ‘What’ (Product) and ‘Where’ (Website) ipation on Customer Visit Frequency and Profitability: An Empiri-
Characteristics on Online Shopping Behavior,” Journal of Market- cal Investigation,” Information Systems Research, 24 (1), 108–27.
ing, 80 (2), 21–38. Schulze, Christian, Lisa Schöler, and Bernd Skiera (2014), “Not All
Mathwick, Charla, Naresh Malhotra, and Edward Rigdon (2001), Fun and Games: Viral Marketing for Utilitarian Products,” Journal
“Experiential Value: Conceptualization, Measurement and Appli- of Marketing, 78 (1), 1–19.
cation in the Catalog and Internet Shopping Environment,” Journal Sen, Shahana and Dawn Lerman (2007), “Why Are You Telling Me
of Retailing, 77 (1), 39–56. This? An Examination into Negative Consumer Reviews on the
Melnyk, Valentyna, Kristina Klein, and Franziska Völckner (2012), Web,” Journal of Interactive Marketing, 21 (4), 76–94.
“The Double-Edged Sword of Foreign Brand Names for Companies Shankar, Venkatesh, J. Jeffrey Inman, Murali Mantrala, Eileen Kelley,
from Emerging Countries,” Journal of Marketing, 76 (6), 21–37. and Ross Rizley (2011), “Innovations in Shopper Marketing: Cur-
Moe, Wendy W. (2003), “Buying, Searching, or Browsing: Differen- rent Insights and Future Research Issues,” Journal of Retailing, 87
tiating Between Online Shoppers Using In-Store Navigational (Supplement 1), S29–S42.
Clickstream,” Journal of Consumer Psychology, 13 (1), 29–39. Sirgy, M. Joseph (1982), “Self-Concept in Consumer Behavior: A
Moe, Wendy W. (2006), “An Empirical Two-Stage Choice Model Critical Review,” Journal of Consumer Research, 9 (3), 287–300.
with Varying Decision Rules Applied to Internet Clickstream Sloot, Laurens M., Peter C. Verhoef, and Philip H. Franses (2005), “The
Impact of Brand Equity and the Hedonic Level of Products on Con-
Data,” Journal of Marketing Research, 43 (4), 680–92.
sumer Stock-Out Reactions,” Journal of Retailing, 81 (1), 15–34.
Moe, Wendy W. and Peter S. Fader (2001), “Modeling Hedonic Port-
Spiegelhalter, David J., Nicola G. Best, Bradley P. Carlin, and
folio Products: A Joint Segmentation Analysis of Music Compact
Angelika Van Der Linde (2002), “Bayesian Measures of Model
Disc Sales,” Journal of Marketing Research, 38 (3), 376–85.
Complexity and Fit,” Journal of the Royal Statistical Society:
Naylor, Rebecca Walker, Cait Poynor Lamberton, and Patricia M.
Series B (Statistical Methodology), 64 (4), 583–639.
West (2012), “Beyond the ‘Like’ Button: The Impact of Mere
Srinivasan, Shuba, Oliver J. Rutz, and Koen H. Pauwels (2016),
Virtual Presence on Brand Evaluations and Purchase Intentions
“Paths To and Off Purchase: Quantifying the Impact of Traditional
in Social Media Settings,” Journal of Marketing, 76 (6), 105–20.
Marketing and Online Consumer Activity,” Journal of the Acad-
Nelson, Phillip (1970), “Information and Consumer Behavior,” Jour-
emy of Marketing Science, 44 (4), 440–53.
nal of Political Economy, 78 (2), 311–29.
Sudhir, K. (2016), “Editorial—The Exploration-Exploitation Tradeoff
Noble, Stephanie M., David A. Griffith, and Marc G. Weinberger
and Efficiency in Knowledge Production,” Marketing Science, 35
(2005), “Consumer Derived Utilitarian Value and Channel Utiliza-
(1), 1–9.
tion in a Multi-Channel Retail Context,” Journal of Business
Taniguchi, Nina (2019), “Same Search Terms, Different Emotions:
Research, 58 (12), 1643–51. Anticipate Customer Needs Throughout the Journey,” Think with
Novak, Thomas P., Donna L. Hoffman, and Adam Duhachek (2003), Google (October), https://www.thinkwithgoogle.com/consumer-
“The Influence of Goal-Directed and Experiential Activities on Online insights/search-intent-and-customer-needs/.
Flow Experiences,” Journal of Consumer Psychology, 13 (1), 3–16. Valentini, Sara, Elisa Montaguti, and Scott A Neslin (2011),
O’Curry, Suzanne and Michal Strahilevitz (2001), “Probability and “Decision Process Evolution in Customer Channel Choice,” Jour-
Mode of Acquisition Effects on Choices Between Hedonic and nal of Marketing, 75 (6), 72–86.
Utilitarian Options,” Marketing Letters, 12 (1), 37–49. Van Trijp, Hans C.M., Wayne D. Hoyer, and J. Jeffrey Inman (1996),
Okada, Erica Mina (2005), “Justification Effects on Consumer Choice “Why Switch? Product Category–Level Explanations for True
of Hedonic and Utilitarian Goods,” Journal of Marketing Variety-Seeking Behavior,” Journal of Marketing Research, 33
Research, 42 (1), 43–53. (3), 281–92.
Park, C. Whan, Bernard J. Jaworski, and Deborah J. MacInnis (1986), Voss, Kevin E., Eric R. Spangenberg, and Bianca Grohmann (2003),
“Strategic Brand Concept-Image Management,” Journal of Mar- “Measuring the Hedonic and Utilitarian Dimensions of Consumer
keting, 50 (4), 135–45. Attitude,” Journal of Marketing Research, 40 (3), 310–20.
Park, Eun Joo, Eun Young Kim, Venessa Martin Funches, and Wakefield, Kirk L. and J. Jeffrey Inman (2003), “Situational Price
William Foxx (2012), “Apparel Product Attributes, Web Brows- Sensitivity: The Role of Consumption Occasion, Social Context
ing, and e-Impulse Buying on Shopping Websites,” Journal of and Income,” Journal of Retailing, 79 (4), 199–212.
Business Research, 65 (11), 1583–89. Wedel, Michel and P.K. Kannan (2016), “Marketing Analytics for
Park, Eunho, Rishika Rishika, Ramkumar Janakiraman, Mark B. Data-Rich Environments,” Journal of Marketing, 80 (6), 97–121.
Houston, and Byungjoon Yoo (2018), “Social Dollars in Online Zaichkowsky, Judith Lynne (1994), “The Personal Involvement
Communities: The Effect of Product, User, and Network Inventory: Reduction, Revision, and Application to Advertising,”
Characteristics,” Journal of Marketing, 82 (1), 93–114. Journal of Advertising, 23 (4), 59–70.