Article

Journal of Service Research

2021, Vol. 24(1) 30-41
Engaged to a Robot? The Role of AI ª The Author(s) 2020
Article reuse guidelines:
sagepub.com/journals-permissions
in Service DOI: 10.1177/1094670520902266
journals.sagepub.com/home/jsr

Ming-Hui Huang1 and Roland T. Rust2

Abstract
This article develops a strategic framework for using artificial intelligence (AI) to engage customers for different service benefits.
This framework lays out guidelines of how to use different AIs to engage customers based on considerations of nature of service
task, service offering, service strategy, and service process. AI develops from mechanical, to thinking, and to feeling. As AI
advances to a higher intelligence level, more human service employees and human intelligence (HI) at the intelligence levels
lower than that level should be used less. Thus, at the current level of AI development, mechanical service should be performed
mostly by mechanical AI, thinking service by both thinking AI and HI, and feeling service mostly by HI. Mechanical AI should be
used for standardization when service is routine and transactional, for cost leadership, and mostly at the service delivery stage.
Thinking AI should be used for personalization when service is data-rich and utilitarian, for quality leadership, and mostly at the
service creation stage. Feeling AI should be used for relationalization when service is relational and high touch, for relationship
leadership, and mostly at the service interaction stage. We illustrate various AI applications for the three major AI benefits,
providing managerial guidelines for service providers to leverage the advantages of AI as well as future research implications for
service researchers to investigate AI in service from modeling, consumer, and policy perspectives.

Keywords
artificial intelligence, human intelligence, replacement, augmentation, service strategy, robots, automation, mechanical AI, thinking
AI, feeling AI, service process, engagement, standardization, personalization, relationalization

The purpose of this article is to delineate a strategic framework The service can scale up by learning with continuing use, accu-
for using artificial intelligence (AI) to engage customers in ser- mulating new information from new questions and more members.
vice. This framework addresses the critical strategic decision It provides personalized advice based on individual members’
that service providers need to make: When to use which AI more characteristics and specific situations (thinking AI for
in service as opposed to human intelligence (HI). Let’s share a personalization).
business example that illustrates the essence of the framework: The service can go further to counsel members, as the transition
into civilian life often is highly uncertain (members often don’t
know where or how to start the daunting process), and involves
USAA, a financial services provider for the military community
both the member and family. In addition to evidence-based,
that is renowned for excellent service, member engagement, and
informed advice, counseling can give members peace of mind
brand loyalty, collaborates with IBM using its Watson Engagement
(feeling AI for relationalization).
Advisor to help serve military members transitioning from the
Eventually it helps create exceptional personalized experiences
military to civilian life.
for USAA members, engaging them deeper, and enhancing brand
The use of this application starts from IBM Watson establish-
loyalty (IBM 2014).
ing, analyzing, and understanding a database based on more than
3,000 USAA documents on topics exclusive to military transitions.
Military members then can visit the USAA website or use their
own mobile device to ask Watson questions specific to leaving the
military, such as job searching, home purchasing, and military 1
Department of Information Management, College of Management, National
benefits, using text or natural language. Watson then searches the Taiwan University, Taipei
2
database it has established for answers. This ask-and-answer pro- Robert H. Smith School of Business, University of Maryland, College Park,
cess iterates and accumulates knowledge over time that persona- MD, USA
lizes answers with greater precision and relevancy.
Corresponding Author:
The service can be as simple as self-service. USAA members Ming-Hui Huang, Department of Information Management, College of
can access anytime anywhere by visiting the USAA website or Management, National Taiwan University, 1, Sec. 4, Roosevelt Rd., Taipei
using their own mobile device for routine, non-thinking, frequently 10617, Taiwan.
asked questions (mechanical AI for standardization). Email: [email protected]
Huang and Rust 31

The above opening case illustrates how AI can be used to that can learn, connect, and adapt. AI can learn, but it is not
engage customers at various service levels and throughout dif- always designed to learn, depending upon the needs of the
ferent stages of the service process. At a basic level, a major application. AI may have varying degrees of learning ability,
service firm like The United Services Automobile Association adaptivity, and connectivity. For example, hotel housekeeping
(USAA) can leverage mechanical AI for robotic self-service service robots are appropriately designed as being capable of
for routine, nonthinking tasks. Then, it can leverage thinking performing mechanical tasks such as making the bed and
AI to make service recommendations to customers based on vacuuming the floor on a routine and consistent basis rather
their current portfolios and prior record of service adoption. than chatting with customers interactively. Such AI applica-
Finally, it can leverage a combination of thinking and feeling tions are designed with a low level of learning capability or
AI to counsel customers (specifically military members) as are only reprogrammed on an occasional basis. There are two
they transition into civilian life. Those AIs are in common use. defining characteristics of AI that result in AI being able to
They are not science fiction. adapt to changing needs (Huang and Rust 2018; Huang, Rust,
Our framework has its theoretical roots in the work of Huang and Maksimovic 2019):
and Rust on AI in service. Rust and Huang (2014) point out that
technology (e.g., machine learning, big data, cloud computing) Self-learning. AI can self-improve automatically by learning
is a primary force for the expansion of the service economy. from various inputs (e.g., big data and machine learning). That
(Huang 2016; Rafaeli et al. 2017) develops a three- is why AI can adapt. Learning implies AI can act and adapt
generational framework for technology innovation—automated based on what has been learned. The deeper the learning, the
technology, thinking technology, and feeling technology—that more AI demonstrates human thinking and feeling capabilities.
illustrates how different technologies can be used for different For example, AI assistant Alexa continuingly learns customer
goals in the frontline service context. Huang and Rust (2018) needs and requirements over time, and its algorithms can adapt
refine this framework into four AIs: mechanical, analytical, to customer personal needs to serve customers better (Dawar
intuitive, and empathetic, which predicts the timing of when 2018; Kaplan and Haenlein 2018).
AI would/will replace human service labor. Huang, Rust, and
Maksimovic (2019) simplify the four AIs into three to demon- Connectivity. AI is seldom standalone machines; even for
strate empirically that AI is driving the economy from mechan- Roomba that appears to stay at its home base quietly, the
ical, thinking, to feeling. vacuuming robot can map a customer’s house floor plan and
The view that there are multiple AIs serves as the important can be connected to Amazon’s Echo and Alexa to enable voice
conceptual basis of the framework proposed here. The frame- control by sharing the floor mapping information. It’s all con-
work directly answers the critical question mentioned above. nected to make the service even better and easier. The connec-
Regarding when to use more AI in service, service providers tion can be machine to machine, machine to customer, or
need to realize that there are multiple AIs—mechanical, think- machine to employee. When Roomba connects with Amazon,
ing, and feeling—that can be used in service to engage custom- Apple, or Google, it is a scenario of machine-to-machine con-
ers. The more complex, idiosyncratic, and emotional service is, nection. When Roomba takes your commands and cleans your
the higher the level of AI (in terms of how difficult for AI to bedroom, it is machine-to-customer connection. When
perform the service function) is required. When AI at a given Roomba sends your floor information to iRobot (its manufac-
intelligence level is used more, HI (human service employees) turer), it is machine-to-employee connection. The connectivity
is used less. Regarding which AI to use to engage customers, of AI is most manifested by the Internet of Things (IoT) such
service providers need to understand the strengths of each AI, that machines, humans, and objects are all connected together,
with mechanical AI being good for standardization, thinking AI and data flow and shared ubiquitously that facilitate learning.
being good for personalization, and feeling AI being good for For example, Hoffman and Novak (2018) state that in the IoT,
relationalization; different AIs can be applied based on con- both smart objects’ and consumers’ experiences are enabled
siderations of the nature of service. and constrained by the network. Ng and Wakenshaw (2017)
In the following sections, we first provide a working defini- state that IoT as a service system is socio-cyber-physical and is
tion of AI in service and delineate the key decisions for using AI liquifying information and data.
in service. We then provide managerial implications for service
providers for using various AIs to engage customers. Finally, we
provide modeling, consumer, and societal implications for future
AI in Service
research to further shape the use of AI in service. AI that can learn, connect, and adapt is increasingly utilized in
service, and today is a major source of service innovation and
revolution (Rust and Huang 2014). For example, service robots
AI in Service interact with customers with scale and consistent quality (Wirtz
et al. 2018), automate social presence in the frontline (van
Defining AI in Service Doorn et al. 2017), and have become part of routine service
AI is machines that exhibit aspects of HI. AI is distinct from experiences (Mende et al. 2019). Big data and machine learn-
general information technology in that it involves technologies ing applications personalize recommendations to customers
32 Journal of Service Research 24(1)

Feeling Intelligence
Learn and adapt from experience
Thinking Intelligence Relationalization
Learn and adapt from data
Mechanical Intelligence Personalization
Learn and adapt only minimally
Standardization

Figure 1. Three artificial intelligences and their benefits to service.

(Chung, Rust, and Wedel 2009; Chung, Wedel, and Rust 2016) standardization (Huang and Rust 2017, 2018). The more
and marketing analytics personalize marketing (Wedel and goods-like the service is, the more appropriate for it to be auto-
Kannan 2016). Anthropomorphized consumer robots make mated for scale and efficiency. This is similar to the McService
consumers feeling warm (Kim, Schmitt, and Thalmann strategy proposed in Huang and Rust (2017); when customers
2019), and natural language–based social robots engage cus- have homogeneous demand for the service and are low in poten-
tomers (Lee, Hosanagar, and Nair 2018). These common and tial customer lifetime value, technologies should be used to
emerging usages of AI for various functions in the service automate service for efficiency. Examples are fast-food ordering
process reveal that service providers need to carefully consider and delivery, self-service, budget service, and customer service
how to use AI to engage customers in a more systematic and for routine issues. In this use of mechanical AI, routine and
strategic way. repetitive human service is transformed into self-service or is
mass produced. For example, for routine purchases, AI assistants
AIs and Their Benefits to Service offer convenient, uninterrupted household operations (Dawar
2018). Another common mechanical AI application is service
Based on the multiple AIs view of Huang and Rust (2018) and robots, for example, hotel housekeeping service robots, which
Huang, Rust, and Maksimovic (2019), we propose that there replace human employees to perform this routine service.
are three AIs—mechanical, thinking, and feeling—that can be
used differentially in service for customer engagement. This
multiple AIs view maintains that there are multiple AIs, with Thinking AI for Service Personalization
each designed to perform different tasks, with mechanical AI
used for simple, standardized, repetitive, and routine tasks, Thinking AI learns and adapts from data. It can be analytical or
thinking AI for complex, systematic, rule-based, and well- intuitive. Analytical AI is designed with the goal of exploring
defined tasks, and feeling AI for social, emotional, communi- customer diversity to identify meaningful patterns (i.e., data
cative, and interactive tasks. The development of AI is mining, text mining). The more advanced intuitive AI is
cumulative. Once AI advances to a higher level, it also typi- designed for maximizing decision-making accuracy (i.e., sol-
cally possesses the lower level intelligence capabilities. ving problems, maximizing the accuracy of answering ques-
Mechanical AI is the lowest and easiest, meaning that current tions in data science language). Thus, thinking AI is ideal for
AI can handle service tasks that require such intelligence pro- service personalization for optimal service productivity (Rust
ficiently. Thinking AI, currently a mainstream research and and Huang 2012). As AI capabilities are cumulative, intuitive
application focus, can analyze big data and make some intui- AI also may process all of the capabilities of analytical AI.
tive decisions. Feeling AI is the most advanced, but its full Thinking AI is ideal for service personalization, especially
potential is not yet realized. when there are abundant customer data available and when the
We propose that each of the three AIs can deliver its unique problems are well-defined, for example, predicting which new
benefits to service. Specifically, mechanical AI is ideal for services will be appealing to which customers. In this situation,
service standardization, thinking AI for service personaliza- there are ample existing customer preference data that can be
tion, and feeling AI for service relationalization (i.e., persona- used to suggest different new services to different customers.
lized relationship). Figure 1 illustrates this differential benefits The analytical subtype of thinking AI is good for uncovering
view of the three AIs. or discovering meaningful patterns in data as the basis of per-
sonalization. Using analytical AI for service personalization is
similar to the customized transaction strategy for service in
Mechanical AI for Service Standardization Huang and Rust (2017); when demand is heterogeneous and
Mechanical AI generally learns and adapts only to a minimal the potential customer lifetime value is lower, analytical AI can
degree. It is designed with the goals of maximizing efficiency be used to capture the individual differences in service prefer-
and minimizing variability. Thus, it is ideal for service ence based on cross-sectional data (e.g., other like-minded
Huang and Rust 33

Table 1. Nature of Service to Consider Whether to Use AI or HI More and Which AI Intelligence to Use.

Levels of AI/HI
Nature of
Service Mechanical AI/HI Thinking AI/HI Feeling AI/HI

Service task Mechanical tasks should be performed Thinking tasks should be performed by Feeling tasks should be performed mostly
mostly by mechanical AI. Mechanical HI both thinking AI and HI. Thinking HI by HI. Feeling HI may be augmented by
is often replaced is augmented lower level AI
Service offering Transactional service Utilitarian service Hedonic service
Service strategy Cost leadership Quality leadership Relationship leadership
Service process Service delivery Service creation Service interaction
Note: AI ¼ artificial intelligence; HI ¼ human intelligence.

customers). For example, for complex shopping decisions, AI There are two polarized applications of feeling AI. At the
assistants can learn consumers’ criteria and optimize whatever low end, feeling AI applications, such as virtual agents and
trade-offs the consumers are willing to make (e.g., higher price chatbots, are widely used to deliver mechanical AI-like cus-
or greener; Dawar 2018). Amazon’s product recommendation tomer service. This is similar to the relational service strategy
system also falls into this category. in Huang and Rust (2017), in which demand is homogeneous
The intuitive subtype of thinking AI has the capability to and potential customer lifetime value is high. Emotional ana-
learn and adapt based on a deeper understanding of the context lytics that operate like thinking AI but with emotional data or
(i.e., deep learning), not just based on observable systematic extracting emotions from data are typical applications. Affec-
patterns. Deep learning is AI that tries to replicate the human tiva (Dwoskin and Rusli 2015) and Magids, Zorfas, and Lee-
brain (neurons, neural network, brain-like virtual neural net- mon’s (2015) emotional-connection analytics are examples of
work, etc. ; Singh 2017). IBM Watson is such an intuitive AI emotional analytics. The current dialog systems popular in the
application that has been adopted by various sectors. With its consumer market, like Alexa, Cortana, and Siri, are another
more humanlike thinking capability, intuitive AI is good for type of application that uses natural language processing to
more profound and adaptive personalization. For example, interact with customers but in a rather mechanical manner.
some studies have demonstrated that AI can be used to build At the high end, feeling AI has potential for customer care
adaptive personalization systems that can personalize increas- that requires empathy and understanding, not just serving as a
ingly effectively for an individual customer over time (Chung, customer contact interface. Automatic speech emotion recog-
Rust, and Wedel 2009; Chung, Wedel, and Rust 2016). Using nition is considered to be the next big thing of AI (i.e., the next-
intuitive AI for dynamic personalization is similar to the rela- gen AI) that can be applied widely to health, retrieval, robotics,
tional service strategy in Huang and Rust (2017), such that security. Such AI needs to be able to truly read human emotions
dynamic (longitudinal) small data for each specific customer and react to the emotions like a human conversational partner
over time are available for personalization. For example, AI would (Schuller 2018). Sophia and more sophisticated chatbots
assistants as consumers’ decision aids are less biased and less are examples of such applications. Feeling AI is still in its early
forgetful. They will retain every last bit of information and stages of development, and thus, feeling mostly remains the
analyze the data to provide objective recommendations to con-
territory of human service employees for the time being.
sumers (Dawar 2018).

Feeling AI for Service Relationalization Applying AIs to Engage Customers

Feeling AI learns and adapts from experience. Experience is We propose that in making the strategic decision of using AI to
defined as data that are contextual- and individual-specific. engage customers, service providers need to realize that each of
This level of AI may possess all the mechanical and thinking the AIs—mechanical, thinking, and feeling—can provide its
AI capabilities but applies these capabilities to experience- unique benefits to service for engaging customers. Summariz-
based data. Feeling AI is ideal for service relationalization, ing from the literature and based on our theory, we consider
defined as personalized relationship, and for customer satisfac- that there are four major factors that shape the role of AI in
tion and retention (Huang and Rust 2017, 2018). It is critical for service: the nature of service task, service offering, service
maintaining customer relationships, in which interaction, com- strategy, and service process. Table 1 illustrates some sugges-
munication, understanding, and experience are critical. All tions along the four factors, and Figure 1 illustrates some sug-
relationships are by nature personal, and feeling AI that can gestions for the combinational use of AIs. Below we elaborate
handle such data is ideal for this purpose. the four factors involved in the decision.
34 Journal of Service Research 24(1)

Nature of Service Task and the three AIs can be combined in various ways to cater to
the nature of the service offering. Service offerings that lean
Tasks are activities involved in a job (Chui, Manyika, and
toward the utilitarian end should use thinking AI more. Utili-
Miremadi 2015). For AI, they are service functions to be per-
tarian service mainly provides instrumental, functional, non-
formed. AI applications do not necessarily replace HI (human
sensory benefits to customers (Huang 2003, 2005). They may
employees); rather, they are designed to perform specific ser-
be considered as high-tech and are naturally suited for thinking
vice tasks. In the end, some tasks may be performed by
AI. For example, cloud data service (e.g., Amazon) and finan-
humans, and others by AI, meaning AI and HI work as a team
cial analysis (e.g., IBM Watson) can be done by thinking AI. In
(Wilson and Daugherty 2018). Nevertheless, the more service
contrast, service offerings that lean toward the hedonic end
tasks involved in a service job can be done by AI, the fewer
should use feeling AI more. Hedonic service mainly provides
human employees will be needed, and the remaining human
sensory benefits such as fun, playfulness, and pleasure to cus-
employees will focus more on the tasks not performed by AI
tomers (Huang 2003, 2005). They may be considered high-
(Huang, Rust, and Maksimovic 2019).
touch and can benefit from feeling AI. For example, computer
Tasks (or service functions) that require a lower intelligence
games can use emotion-detection AI to engage players in a
level should generally be replaced first. That is, mechanical/
flow state, and smart houses (i.e., houses that are equipped and
routine/repetitive tasks should be performed mostly by
connected with smart devices so that customers can remote
mechanical AI. Given the current AI level, many mechanical
control the temperature, surroundings, and issue commands
HI tasks should be mostly replaced by AI. For manufacturing,
to adjust the house for their needs) can use feeling AI applica-
the nature of this type of tasks is more straightforward. Such
tions to further transform the houses into “feeling homes” that
tasks are mostly mechanical, in a well-specified context, and
provide emotional comfort to owners.
lack variation, and thus can often be easily done by manufac-
The nature of service offering can also vary in degree of
turing robots. For service, there are variations in this type of
transactional-relational. Transactional service has little to
tasks: Some are as mechanical as in manufacturing, such as
gain from a customer relationship and will benefit more from
cash withdrawals done by ATM (Meuter et al. 2000), whereas
AI replacement (Huang and Rust 2017). For example, fast-
some are routine and simple but require context or involve
food restaurants can use more mechanical AI to serve cus-
variations, such as restaurant table waiting service that cannot
tomers without undermining their value proposition, whereas
be done easily by mechanical AI.
high-end French restaurants should use less. In contrast, rela-
Thinking tasks should be performed by both thinking AI and
tional service can benefit from having a solid relationship
HI. This is the type of task where augmentation (skilled service
with customers because a higher customer lifetime value can
employees augmented by thinking AI) is most likely to occur.
be expected; thus, service providers should strive to use feel-
For example, managers can use business analytics to support
ing AI more.
their decisions, and physicians can benefit from health-care
In brief, utilitarian service should use thinking AI more,
applications to aid their diagnosis. The more analytical the
whereas hedonic service should use feeling AI more. Transac-
service task is, the more likely the analytical subtype of think-
tional service should use mechanical AI more, whereas rela-
ing AI can address the task with limited human intervention,
tional service should use feeling AI more. Together, the
while the more intuitive the service task is, the less likely it is
continua of utilitarian-hedonic and of transactional-relational
that AI can successfully address the task without human inter-
suggest four possible AI/HI portfolios. Figure 2 illustrates the
vention. Future progress in the intuitive subtype of thinking AI
combinational use of AIs based on nature of service.
is likely to change this conclusion in the not-so-distant future.
At that point, as thinking AI applications advance to the intui-
Utilitarian transactional service. This type of service should use
tive level, most of the thinking tasks can be done by machines.
analytical AI more, which is thinking AI but mostly performs
Feeling tasks should today be performed mostly by HI.
mechanical analyses. Analytical AI performs logical, analyti-
Feeling HI may be augmented by lower levels of AI applica-
cal, and rule-based learning. The nature of learning is mechan-
tions (i.e., mechanical and/or thinking AI). For example, for
ical, but the capability reaches the thinking level due to
low-end applications, mechanical feeling AI, such as chatbots,
rule-based learning from big data to achieve data- and
has been widely applied to handle standardized and routine
analytics-based personalization. Analytical HI will be gradu-
customer service. For high-end applications, Siri, Alexa, and
ally replaced as analytical AI gets more dominant for data- and
other similar voice recognition personal assistants that possess
computing-based learning.
analytical capability are used to search for information for
consumers and respond to consumer queries, using a human
Utilitarian relational service. This type of service should use both
voice.
intuitive AI and HI. Intuitive AI is the more advanced subtype
of thinking AI that is closer to feeling AI. It is thinking AI that
is capable of bounded rationality and commonsense thinking.
Nature of Service Offering Current AI has not yet achieved the full capability of human
The nature of service offering can vary in the utilitarian- meaning–based thinking, and thus, both AI and HI should be
hedonic continuum and the transactional-relational continuum, involved in this type of service. For this use of AI, the customer
Huang and Rust 35

relationship is built based on the deep understanding of and intuitive AI that can come up with a recommended travel
customer preferences, while not necessarily involving an plan are appropriate.
emotional connection with customers.
Relationship leadership emphasizes customer intimacy for
customer satisfaction. For high lifetime value customers, ser-
Hedonic transactional service. This type of service should use vice providers might eventually leverage feeling AI to engage
mechanical feeling AI more and HI less (i.e., unskilled human customers on an ongoing basis. This strategy also involves
employees). Mechanical feeling AI is mechanical AI with personalization and further focuses on using emotions as a
some feeling capability. Such AI learns and adapts from limited differentiator. For example, airlines provide both high-quality
emotional data and can do some relationalization. A lot of ground services and aircraft facilities to achieve quality lead-
conversational bots providing customer service are at this level, ership that can achieve relationship leadership but achieve best
in which the nature of service requires communication and results if they additionally provide emotional care and high-
emotions, but mostly are repetitive. touch service by ground staff and flight attendants. The latter
high-touch service promises to benefit from feeling AI by ana-
Hedonic relational service. This type of service should use both lyzing, recognizing, and understanding customer emotions
feeling AI and HI (i.e., skilled human employees with high (Schuller 2018), and responding and serving them in an emo-
emotional intelligence). Such AI learns and adapts from emo- tionally appropriate manner on an individual customer basis.
tional data to build connections and relationships with custom-
ers. For example, the AI companion, Replika, provides
emotionally personalized everyday conversation with custom- Stage of Service Process
ers. Natural language processing dialog systems based on gen-
We refer to a service process as how a service is provided or
erative machine learning methods and embedded social
delivered to a customer. The service process view recognizes
caretaking are some applications of such AI (McDuff and Czer-
that the service outcome depends on the dyadic interaction
winski 2018). Given the current AI level, this type of service
process between customers and service providers (Ma and
should be offered mostly by HI.
Dube 2011). It means that value is created continuously by
engaging customers and service providers jointly and interac-
tively (Ramaswamy and Ozcan 2018). As a result, customers
Strategic Emphasis of Service Providers experience the quality that providers deliver and reach an out-
come evaluation about the quality (not necessarily the same as
Extending Treacy and Wiersma’s (1993) three market leader-
what was objectively delivered; Golder, Mitra, and Moorman
ship strategies—operational excellence, product leadership,
2012). Bitner, Ostrom, and Morgan (2008) provide a detailed
and customer intimacy—to service, we recommend that service
service blueprinting about how a service process can be broken
providers that emphasize cost leadership (i.e., operational
down into subprocesses for service innovation.
excellence) use mechanical AI more, quality leadership (i.e.,
We break down the service process into three stages—
service performance) should use thinking AI more, and rela-
delivery, creation, and interaction—and order them in this
tionship leadership (i.e., customer intimacy) should (eventu-
way based on the three AI levels, from mechanical, to think-
ally) use feeling AI more.
ing, and to feeling. The three stages are reciprocal; they do not
Cost leadership emphasizes operational excellence by auto-
necessarily follow a sequence, as in physical goods manufac-
mating service processes to reduce costs. The more service
turing from production to delivery. Instead, each process
processes can be standardized, the more process automation
stage is likely to give rise to the other process stages. Figure 3
can be achieved by mechanical AI. For example, McDonald’s
illustrates this reciprocal process. Table 2 summarizes the key
uses robots to deliver ordered foods to customers, firms use
managerial questions for service providers to address, the
virtual bots to deliver customer service, and Amazon attempts
dominant AI at each process stage, the major service tasks
to use drones to deliver products.
to perform, and the respective AI application examples. For
Quality leadership emphasizes achieving premium quality each contact point (or process stage), a mix of AI can be used.
for customer experience. Such a competitive strategy may not For example, the service provider can use mechanical AI to
always involve big data as there is a higher degree of customer complete the transaction and deliver the service (e.g., Amazon
heterogeneity in quality expectation that cannot be sacrificed Prime Air’s delivery drones), use thinking AI for market pro-
for scale economies. Thinking AI that maximizes diversity for specting (e.g., Gap uses AI for predicting fashion trend), and
service personalization can be used to address individual cus- use feeling AI to provide customer service (chatbots for cus-
tomers’ needs and requirements. For example, trip planning for tomer service). It is important to note that breaking down the
travelers may take each traveler’s unique preferences into con- service process into the three stages is for the purpose of
sideration, such as a trip that can relax his or her mind or a trip illustrating which AI suits best for each stage. Not all services
that can see most of a city. Such planning is more complicated have a clear-cut boundary between the stages, for example,
than just buying air tickets and booking hotels, and thus, ana- for haircutting service, delivery, creation, and interaction all
lytical AI that can analyze like-minded travelers’ preferences occur simultaneously. For insurance, service creation is
36 Journal of Service Research 24(1)

Mechanical
Transactional
thinking AI
Mechanical AI
(analytical)
Utilitarian
Thinking AI
Relational Thinking-feeling AI
Feeling AI (intuitive)
Nature of
service
Transactional Mechanical feeling
Mechanical AI AI
Hedonic
Feeling AI
Relational
Feeling AI
Feeling AI

Figure 2. Combinational use of artificial intelligences based on nature of service.

Table 2. AI-Enabled Service Process.

Service Process Key Question to Address Dominant AI Service Tasks AI Applications

Service delivery How to deliver service Mechanical  Shipping  Service process automation
efficiently? AI  Delivery  Robotic self-service
automates  Payment  AI assistant to a consumer’s daily life (not as the
firm’s customer service agent; Dawar 2018)
Service creation How to create valuable Thinking AI  Identify new market  Predictive analytics (Agrawal, Gans, and
service to which predicts  Develop new Goldfarb 2018)
customers? service  Computing creativity
 Personalize service  Data mining
Service interaction How and what to Feeling AI  Engage customers  Speech emotion recognition (e.g., Alexa,
communicate with engages  Personalize service Cortana, Siri; Schuller 2018)
customers? adaptively  Sentiment analysis (the recognition of emotion
 Interact with from text, Schuller 2018)
customers  Deep learning
 Customer service  Convolutional neural networks
 Customer care  End-to-end learning
 Chatbots
 Dynamic optimization
Note. AI ¼ artificial intelligence.

figuring out what kind of policy to offer, delivery arises when goal is to maximize service productivity. Use of AI may
there is a claim, and interaction is the relationship between the improve delivery efficiency for service providers and conve-
agent and the customer. In other words, there may be services nience for customers.
in which two or more of the three stages combine (e.g., This is the stage that mechanical AI dominates. Mechanical
coproduction). AI has the advantages of consistency, reliability, and effi-
Below we delineate how the three AIs can be matched to the ciency; thus, service providers can leverage these advantages
service subprocesses of delivery, creation, and interaction. We of mechanical AI to deliver service. Two major applications of
also provide strategic suggestions for using AIs to engage cus- mechanical AI for service delivery are to (1) automate the
tomers along the service process. service process and (2) automate the offering of service to
customers.
Mechanical AI for service delivery. At this process stage, the key Automating the service process, such as automated payment
question that service providers need to ask is as follows: How or automated delivery (e.g., Amazon’s drone delivery), is
to deliver service efficiently? The service delivery tasks, such intended to streamline the service process such that customers
as shipping, delivery, and payment, are more routine and repe- can experience a smooth and uninterrupted service process.
titive than tasks involved in the other two process stages. The This is often a backend automation in which customers do not
Huang and Rust 37

individual customers. Service creation involves two related

questions: What service to create, and which customers would
consider it valuable. Major service tasks involved include iden-
tifying new markets, developing new service, and personaliz-
Service interaction ing service.
Feeling AI The dominant AI at this stage is thinking AI. In terms of AI
applications, predictive analytics can be used to predict cus-
tomer preferences, computing creativity can be used to develop
new service, and data mining (or any other types of pattern
mining) can be used to identify like-minded customers for
Service creation creating personalized service. Deep learning could help in cus-
tomer segmentation and provide different promotion cam-
Thinking AI
paigns for different groups of customers by experience from
learning from lots of data. Furthermore, deep learning may
even help in prediction of the trends in the industry because
of its continual learning from internal and external data.
For developing new service, thinking AI can be used to
Service delivery extract value from systematic patterns from data. For example,
Mechanical AI Gap, the fashion retailer, removed the position of creative
director and replaced it with a big data-driven collective crea-
tive ecosystem to identify the fashion trend for the next season
(Israeli and Avery 2017). Netflix uses data to decide which new
series and movies to develop. Their CEO, Reed Hastings, said
Figure 3. The circular service process with dominant artificial such a decision can be considered as “informed intuition”
intelligences. because data science alone is insufficient to predict which
product will hit (Israeli and Avery 2017).
interact directly with the mechanical AI, but service providers For predicting customer preferences, Amazon’s anticipatory
deploy it on the backend to automate the process. Amazon’s shipping and Gap’s relying on data to predict fast fashion trends
one-click buying is a classic example that allows customers to are examples of predictive analytics. Such uses of analytical AI,
purchase with one click without having to go through the such as in-car intelligent systems with big data analytics, create
multiple steps of filling shopping cart, providing shipping service that is data-, information-, and knowledge-intensive.
information and credit card information, and so on. Being an For creating personalized service, thinking AI can be used to
e-commerce giant, Amazon invests heavily and experiments predict (e.g., fashion trend prediction) and create service (cus-
with various mechanical AI applications to deliver its offerings tomer/market prospecting) based on systematic patterns from
(goods and service) to customers, such as Amazon Prime Air’s data. For example, Harley-Davidson uses “Albert,” an AI-
delivery drones and Amazon Go. Many AI applications have driven marketing tool to adjust its marketing strategy instantly
also been used to streamline the payment process, such as to different customers, which increased New York sales leads
mobile money transfer, digital wallet, and virtual banking. AI by 2,930% in 3 months (Power 2017). Albert analyzes the
start-up Elementum’s intelligent automation deploys AI on top existing customer data to identify what high-value customers
of automation technologies to achieve more efficient processes are like and what kind of marketing campaigns would be most
and services. Elementum automates supplier management pro- successful with them. It then automatically changes campaigns
cesses that enable its clients to get early warnings of potential to make them more attractive to important potential customers.
problems and manage the problems before their competitors
even know the problems (Wallis and Santiago 2017). AIoT is Feeling AI for service interaction. At this stage, the key question to
another instance for mechanical AI, which is the application of address is how and what to communicate with customers. The
AI on top of IoT. dominant AI involved at this stage is feeling AI, which is used
Automating the offering of service to customers is deliver- to perform the service tasks of engaging customers, personaliz-
ing service as self-service. Robotic self-service, such as Pepper ing service adaptively, and providing customer service. AI
and Roomba, and AI assistants, such as Alexa, use intelligent applications need to be able to detect, understand, and respond
devices to deliver service automatically. This is a frontend to customer emotions adaptively to perform those tasks. This is
automation in which customers interact with the mechanical often the most important service process stage but currently has
AI directly. the lowest level of AI involvement due to the less mature AI
applications at this intelligence level.
Feeling AI delivers social-, emotional-, and relational-based
Thinking AI for service creation. At the service creation stage, the service to customers. It can be used to provide emotional sup-
key question to address is how to create valuable service for port and emotional satisfaction to customers (e.g., customer
38 Journal of Service Research 24(1)

interaction, customer service, and customer experience). Feel- offering is transactional, that is, with limited relational benefit;
ing AI learns and adapts empathetically based on experience when the service strategy emphasizes low cost as the major
and understanding and thus can provide emotional support and benefit to customers; and when at the stage of service delivery
emotional satisfaction to customers. to enhance the convenience benefit to customers. In all these
Feeling AI is important for service interaction, which can be situations, mechanical AI can be used to automate service pro-
divided into two subprocesses: the marketing communication cess from the backend (process efficiency) and automate ser-
stage and the after-sale customer service stage. For marketing vice offerings and delivery to customers from the frontend
communication, service providers need to emotionally connect (convenience benefit).
with customers to differentiate themselves from competitors
(Magids, Zorfas, and Leemon 2015). For customer service, Using thinking AI for personalization. Service providers should use
service providers often face emotionally charged customers thinking AI more when the nature of service task is mostly
(e.g., customer complaints) that require feeling AI to handle. data-based, analytical, predictive, thinking; when the nature
Wallis and Santiago (2017) use the example of Capital One of service offering is utilitarian, that is, customers obtain func-
Eno to illustrate that AI actually needs to be empathic in inter- tional, high-tech benefits from the service; when the service
acting with customers. Eno is a natural language chatbot. Cus- strategy emphasizes high quality as the major benefit to cus-
tomers can text Eno anytime to review their accounts, pay tomers; and when the service creation stage requires figuring
credit bill, and ask general questions. Capital One surprisingly out valuable new services for specific customers. Thinking AI
learned that their customers tend to build relationships with can be used to predict new markets, create new services, pros-
those chatbots even while knowing that they are talking to a pect new customers, and customize service.
bot. Nevertheless, most service providers currently rely on text-
based chatbots to provide customer service that are not very Using feeling AI for relationalization. Service providers should use
effective in handling customer complaints. feeling AI more when the nature of service task is mostly
By being able to act and react empathetically, according to experience-based, emotional, and requires interaction and com-
Wallis and Santiago (2017), AI can be used to improve inter- munication; when the nature of service offering is hedonic, that
actions and deepen trust. For improved interactions, AI can is, customers obtain sensory, fun, high-touch benefits from the
deliver superior experiences to customers based on hyperper- service, or relational, that is, higher customer lifetime value;
sonalization and the curation of real-time information. On top when the service strategy emphasizes customer relationship as
of overall satisfaction improvement, this can also generate the platform for continuing improving service for customers;
greater acquisition and retention rates among customers. For and when at the service interaction stage provides an opportu-
deepened trust, AI can more effectively prevent and detect nity to communicate and interact with customers about the
anomalies. It also provides the ability to significantly reduce value of the service. Feeling AI can be used to engage custom-
false positives. As Dawar (2018) points out, AI platforms will ers, personalize service adaptively over time, and provide cus-
even know whether consumers are likely to adapt their require- tomer care and customer service.
ments in different contexts—for example, if a person on a diet
will make an exception for dessert when celebrating.
Research Implications
The strategic use of the three AIs in service also shapes the
Discussion research about AI in service. We discuss the implications for
We outline a framework for using various AIs in service to modeling, consumer behavior, and societal research respec-
engage customers. The framework helps service providers to tively, along the three service benefits of AI:
decide when and how to use mechanical AI for service stan-
dardization, thinking AI for service personalization, and feel- Modeling implications. For service modelers, key challenges for
ing AI for service relationalization, taking into consideration the three benefits of AI include:
the nature of service task, service offering, service strategy, and
service process.  For mechanical AI-enabled standardization, developing
new algorithms and models that can accommodate ser-
vice variability without losing the efficiency benefit of
Managerial Implications standardization is a key challenge.
Taking the four factors involved in the strategic decision into  For thinking AI-enabled personalization, personaliza-
consideration, we provide the following suggestions for using tion can be achieved using both cross-sectional and
different AIs in service: longitudinal data. Cross-sectional personalization
leverages the benefits of big data, but the personalization
Using mechanical AI for standardization. Service providers should is achieved by inference from like-minded customers.
use mechanical AI more when the nature of service task is Longitudinal personalization can capture a specific cus-
mostly standard, routine, repetitive, and can be performed tomer’s preferences along his or her lifetime, but such
without considering context; when the nature of service data tend to be small and sparse. Therefore, developing
Huang and Rust 39

new algorithms and models that can personalize based own. Thus, how to have AI and HI work together is a
on a combination of both cross-sectional and longitudi- major challenge. There are also general concerns about
nal customer data is desirable. whether thinking AI will displace even thinking work-
 For feeling AI-enabled relationalization, algorithms and ers, not just unskilled mechanical workers, and policy
models that can handle emotional data (i.e., data that are research looking into this issue will have important
individual- and context-specific) are still not well- implications for the future economy.
developed, and thus, more studies are needed to bring it  Biological emotions are human characteristics that AI,
fully to fruition. Challenges include the difficulty of cap- even the current feeling AI, cannot mimic. This AI tech-
turing emotional data, due to context tending to be lost in nological bottleneck has been demonstrated to give rise
the data capturing process, and the difficulty of modeling to the Feeling Economy because employment and wages
such data due to emotional data being multimodal and will be more attributable to feeling tasks and jobs when
containing more nuance than rational cognitive data. thinking AI can do most of the thinking tasks and jobs
(Huang, Rust, and Maksimovic 2019). The possibility
that the continuing advancement of feeling AI will result
Consumer behavior implications. For consumer researchers, a bet- in true emotional machines generates the concern of
ter understanding about consumers’ heterogenous preferences singularity, that is, AI becomes completely dominant
to service automation, concerns about trading personal infor- in all levels of intelligences over humans (Kurzweil
mation for personalization, and reactions to having relation- 2005). The issue of how best to approach feeling AI thus
ships with AI are major issues. is likely become a major policy challenge at some point
in the future.
 Many services have been automated using various
embedded and embodied mechanical AI applications,
yet consumers react to service automation differently: Conclusion
some prefer AI service, some prefer HI service; thus, it is We provide a theoretically driven strategic framework for
essential to have a better understanding about how con- when to use more AI (as opposed to HI) and which AI should
sumers react to AI-enabled service standardization. be used in service. The proliferation of AI in service diffuses in
 With thinking AI able to do more refined personaliza- order from mechanical, to thinking, and to feeling. As AI
tion using more personal data, consumer concerns about advances to a higher level, more AI of that intelligence level
privacy may also be heightened (Rust, Kannan, and should be used and less HI at that intelligence level be used.
Peng 2002). Understanding consumers’ preferred degree Thus, at the current level of technological development,
of personalization, relative to the degree of privacy con- mechanical service should be performed mostly by mechanical
cern, is imperative. AI, thinking service should be performed by both thinking AI
 How consumers react to AI that is used to establish and and HI (thinking HI is augmented), and feeling service should
maintain a relationship with them is a hot issue. Given the be performed mostly by HI.
uniqueness of human emotions having a physiological basis, Regarding which AI to use, in general, mechanical AI can
a potential future avenue of research is to take a neuroscience standardize service, thinking AI can personalize service, and
approach to capturing and understanding the full spectrum of feeling AI can relationalize service. Thus, mechanical AI
consumer emotional reactions to feeling AI. should be used more when the nature of the service task (i.e.,
service function) is routine and repetitive, the service offering
Societal implications. For policy makers, mitigating the nega- is transactional (i.e., goods-like and having limited relational
tive impact of mechanical AI displacing unskilled human benefits), the service strategy is cost leadership, and the service
service employees, making thinking AI and HI work process is at the delivery stage. Thinking AI should be used
together, and how and to what extent to develop feeling more when the nature of service tasks is data-based, analytical,
AI are major issues. and predictive; the service offering is utilitarian (i.e., high-tech
service); the service strategy is quality leadership; and the ser-
 Using more mechanical AI for standardization is bound vice process is at the creation stage. Feeling AI should be used
to result in displacement of mechanical service employ- more when the nature of service tasks is mostly experience-
ees. We have witnessed such displacement when the based, emotional, and requires interaction and communication;
service economy began replacing the manufacturing the service offering is relational (i.e., high customer lifetime
economy in the turn of the 20th century. Where will value) and hedonic (i.e., high-touch service); the service strat-
those unskilled service employees go and how we can egy is relationship leadership; and the service process is at the
retrain and relocate them are pertinent issues that econ- interaction stage.
omists, policy makers, and service providers need to
consider. Declaration of Conflicting Interests
 Thinking AI supports skilled service employees and can The author(s) declared no potential conflicts of interest with respect to
sometimes even perform such service functions on its the research, authorship, and/or publication of this article.
40 Journal of Service Research 24(1)

Funding Huang, Ming-Hui (2003), “Designing Website Attributes to Induce Experi-

The author(s) disclosed receipt of the following financial support for ential Encounters,” Computers in Human Behavior, 19 (4), 425-442.
the research, authorship, and/or publication of this article: This IBM (2014), “USAA and IBM Join Forces to Service Military
research was supported by grants (MOST 104-2410-H-002-142- Members,” IBM News Room (July 23), (accessed June 15, 2019),
MY3, MOST 106-2410-H-002-056-MY3, and MOST 107-2410-H- [available at https://www-03.ibm.com/press/us/en/pressrelease/
002-115-MY3) from the Ministry of Science and Technology, 44431.wss].
Taiwan. Israeli, Ayelet and Jill Avery (2017), “Predicting Consumer Tastes
with Big Data at Gap,” Harvard Business School Publication
ORCID iD (July 10).
Kaplan, Andreas ad Michael Haenlein (2018), “Siri, Siri, in My Hand:
Ming-Hui Huang https://orcid.org/0000-0001-8954-3172
Who’s the Fairest in the Land? On the Interpretations, Illustrations,
and Implications of Artificial Intelligence,” Business Horizons, 62
References (1), 15-25.
Agrawal, Ajay, Joshua Gans, and Avi Goldfarb (2018), Prediction Kim, Seo Young, Bernd H. Schmitt, and Nadia M. Thalmann (2019),
Machines: The Simple Economics of Artificial Intelligence. “Eliza in the Uncanny Valley: Anthropomorphizing Consumer
Brighton, MA: Harvard Business Review Press. Robots Increases their Perceived Warmth but Decreases Liking,”
Bitner, Mary Jo, Amy L. Ostrom, and Felicia N. Morgan (2008), Marketing Letters, 30 (1), 1-12.
“Service Blueprinting: A Practical Technique for Service Kurzweil, Ray (2005), The Singularity Is Near: When Humans Trans-
Innovation,” California Management Review, 50 (3), 66-94. cend Biology. New York: Penguin.
Chui, Michael, James Manyika, and Mehdi Miremadi (2015), “Four Lee, Dokyun, Kartik Hosanagar, and Harikesh S. Nair (2018),
Fundamentals of Workplace Automation,” McKinsey Quarterly “Advertising Content and Consumer Engagement on Social
(November), (accessed March 5, 2017), [available at http://www. Media: Evidence from Facebook,” Management Science, 64
mckinsey.com/businessfunctions/digitalmckinsey/ourinsights/ (11), 5105-5131.
fourfundamentalsofworkplaceautomation]. Ma, Zhenfeng and Laurette Dube (2011), “Process and Outcome
Chung, Tuck Siong, Michel Wedel, and Roland T. Rust (2016), Interdependency in Frontline Service Encounters,” Journal of
“Adaptive Personalization Using Social Networks,” Journal of the Marketing, 75 (3), 83-98.
Academy of Marketing Science, 44 (1), 66-87. Magids, Scott, Alan Zorfas, and Daniel Leemon (2015), “The New
Chung, Tuck Siong, Roland T. Rust, and Michel Wedel (2009), “My Science of Customer Emotions,” Harvard Business Review, 93
Mobile Music: An Adaptive Personalization System for Digital (11), 66-76.
Audio Players,” Marketing Science, 28 (1), 52-68. McDuff, Daniel and Mary Czerwinski (2018), “Designing Emotion-
Dawar, Niraj (2018), “Marketing in the Age of Alexa,” Harvard Busi- ally Sentient Agents,” Communications of the ACM, 61 (12),
ness Review (May-June). 74-83.
Dwoskin, E. and E. M. Rusli (2015), “The Technology that Unmasks Mende, Martin, Maura L. Scott, Jenny van Doorn, Dhruv Grewal, and
Your Hidden Emotions,” The Wall Street Journal (January 28), Ilana Shanks (2019), “Service Robots Rising: How Humanoid
http://www.wsj.com/articles/startups-see-your-face-unmask-your- Robots Influence Service Experiences and Elicit Compensatory
emotions-1422472398 (accessed May 15, 2016). Consumer Responses,” Journal of Marketing Research, 56 (4),
Golder, Peter N., Debanjan Mitra, and ad Christine Moorman (2012), 535-556.
“What Is Quality? An Integrative Framework of Processes and Meuter, Matthew L., Amy L. Ostrom, Robert I. Roundtree, and Mary
States,” Journal of Marketing, 76 (4), 1-23. Jo Bitner (2000), “Self-Service Technologies: Understanding Cus-
Hoffman, Donna L. and Thomas P. Novak (2018), “Consumer and tomer Satisfaction with Technology-Based Service Encounters,”
Object Experience in the Internet of Things: An Assemblage The- Journal of Marketing, 64 (3), 50-64.
ory Approach,” Journal of Consumer Research, 44 (6), 1178-1204. Ng, I.C.L. and Wakenshaw S.Y.L. (2017), “The Internet-of-Things:
Huang, Ming-Hui, Roland T. Rust, and Vojislav Maksimovic (2019), Review and Research Directions,” International Journal of
“The Feeling Economy: Managing in the Next Generation of Arti- Research in Marketing, 34 (1), 3-21.
ficial Intelligence (AI),” California Management Review, 61 (4), Power, Brad (2017), “How Harley-Davidson Used Artificial Intelli-
43-65. gence to Increase New York Sales Leads by 2,930%,” Harvard
Huang, Ming-Hui and Roland T. Rust (2018), “Artificial Intelligence Business Review, May 30, (accessed January 28, 2020), [available
in Service,” Journal of Service Research, 21 (2), 155-172. at hbr.org/2017/05/how-harley-davidson-used-predictive-analy
Huang, Ming-Hui and Roland T. Rust (2017), “Technology-Driven tics-to-increase-new-york-sales-leads-by-2930].
Service Strategy,” Journal of the Academy of Marketing Science, Rafaeli, Anat, Altman Daniel, Gremler Dwayne D., Huang Ming-Hui,
45 (6), 906-924. Grewal Dhruv, Iyer Bala, Parasuraman A., and Ruyter Ko de (2017),
Huang, Ming-Hui (2016), “Technology on the Frontline: From Dumb “The Future of Frontline Research: Invited Commentaries,”
to Thinking to Feeling,” Organizational Frontlines Research Sym- Journal of Service Research, 20 (1), 91-99.
posium, Tallahassee, FL. Ramaswamy, Venkat and Kerimcan Ozcan (2018), “Offerings as
Huang, Ming-Hui (2005), “Web Performance Scale,” Information & Digitalized Interactive Platforms: A Conceptual Framework and
Management, 42 (6), 841-852. Implications,” Journal of Marketing, 82 (4), 19-31.
Huang and Rust 41

Rust, Roland T. and Ming-Hui Huang (2014), “The Service Revolu- Wallis, Jodie and Deborah Santiago (2017), “The Future of
tion and the Transformation of Marketing Science,” Marketing Growth: AI Comes of Age,” Rotman Management Magazine
Science, 33 (2), 206-221. (December 1), Harvard Business Review Store Product #:
Rust, Roland T. and Ming-Hui Huang (2012), “Optimizing Service ROT350-PDF-ENG.
Productivity,” Journal of Marketing, 76 (2), 47-66. Wedel, Michel and P. K. Kannan (2016), “Marketing Analytics for
Rust, Roland T., P. K. Kannan, and Na Peng (2002), “The Customer Data-Rich Environments,” Journal of Marketing, 80 (6), 97-121.
Economics of Internet Privacy,” Journal of the Academy of Mar- Wilson, James H. and Paul R. Daugherty (2018), “Collaborative Intel-
keting Science, 30 (4), 455-464. ligence: Humans and AI are Joining Forces,” Harvard Business
Schuller, Bjorn W. (2018), “Speech Emotion Recognition: Two Review (July-August), 115-123.
Decades in a Nutshell, Benchmarks, and Ongoing Trends,” Wirtz, Jochen, Paul G. Patterson, Werner H. Kunz, Thorsten Gruber,
Communications of the ACM, 61 (5), 90-99. Lu Vinh Nhat, Stefanie Paluch, and Antje Martins (2018), “Brave
Singh, Aditya (2017), “Deep Learning Will Radically Change the New World: Service Robots in the Frontline,” Journal of Service
Ways We Interact with Technology,” Harvard Business Review, Management, 29 (5), 907-931.
January 30, (accessed January 28, 2020), [available at hbr.org/
2017/01/deep-learning-will-radically-change-the-ways-we-inter
act-with-technology]. Author Biographies
Treacy, Michael and Fred Wiersema (1993), “Customer Intimacy Ming-Hui Huang is a Distinguished Professor in the Department of
and Other Value Disciplines,” Harvard Business Review (January- Information Management, College of Management, National Taiwan
February), 84-93. University.
van Doorn, Jenny, Martin Mende, S. Noble, John Hulland, Amy
Ostrom, Dhruv Grewal, and A. Petersen, (2017), “Domo Arigato Roland T. Rust is a Distinguished University Professor and David
Mr. Roboto: Emergence of Automated Social Presence in Organi- Bruce Smith Chair in Marketing, and executive director of the Center
zational Frontlines and Customers’ Service Experiences,” Journal for Excellence in Service at the Robert H. Smith School of Business,
of Service Research, 20 (1), 43-58. University of Maryland.