A Strategic Framework For Artificial Intelligence in Marketing

Journal of the Academy of Marketing Science (2021) 49:30–50
https://doi.org/10.1007/s11747-020-00749-9
CONCEPTUAL/THEORETICAL PAPER
A strategic framework for artificial intelligence in marketing

Ming-Hui Huang 1 & Roland T. Rust 2
Received: 3 March 2020 / Accepted: 29 September 2020 / Published online: 4 November 2020
# The Author(s) 2020
Abstract
The authors develop a three-stage framework for strategic marketing planning, incorporating multiple artificial intelligence (AI)
benefits: mechanical AI for automating repetitive marketing functions and activities, thinking AI for processing data to arrive at
decisions, and feeling AI for analyzing interactions and human emotions. This framework lays out the ways that AI can be used
for marketing research, strategy (segmentation, targeting, and positioning, STP), and actions. At the marketing research stage,
mechanical AI can be used for data collection, thinking AI for market analysis, and feeling AI for customer understanding. At the
marketing strategy (STP) stage, mechanical AI can be used for segmentation (segment recognition), thinking AI for targeting
(segment recommendation), and feeling AI for positioning (segment resonance). At the marketing action stage, mechanical AI
can be used for standardization, thinking AI for personalization, and feeling AI for relationalization. We apply this framework to
various areas of marketing, organized by marketing 4Ps/4Cs, to illustrate the strategic use of AI.
Keywords Artificial intelligence . Machine learning . Mechanical AI . Thinking AI . Feeling AI . Strategic AI marketing .
Marketing strategy . Standardization . Personalization . Relationalization . Segmentation . Targeting . Positioning . 4Ps . 4Cs
Artificial intelligence (AI) in marketing is currently gaining personal assistant to customers. Lexus uses IBM Watson to
importance, due to increasing computing power, lower com- write its TV commercial scripts, “Driven by Intuition.”
puting costs, the availability of big data, and the advance of Affectiva, based on affective analytics, recognizes consumers’
machine learning algorithms and models. We see wide appli- emotions while they are watching commercials. Replika, a
cations of AI in various areas of marketing. For example, machine learning-based chatbot, provides emotional comfort
Amazon.com’s Prime Air uses drones to automate shipping to consumers by mimicking their styles of communication. It
and delivery. Domino’s pizza is experimenting with has even been asserted that AI will change the future of mar-
autonomous cars and delivery robots to deliver pizza to the keting substantially (Davenport et al. 2020; Rust 2020).
customer’s door. RedBalloon uses Albert’s AI marketing However, academic marketing research to date provides in-
platform to discover and reach new customers. Macy’s On sufficient guidance about how best to leverage the benefits of
Call uses natural language processing to provide an in-store AI for marketing impact.
The academic literature on AI in marketing may be sorted
into four main types. These are (1) technical AI algorithms for
John Hulland and Mark Houston served as Editors for this article.
solving specific marketing problems (e.g., Chung et al. 2009;
* Ming-Hui Huang
Chung et al. 2016; Dzyabura and Hauser 2011, 2019), (2)
huangmh@ntu.edu.tw customers’ psychological reactions to AI (e.g., Luo et al.
2019; Mende et al. 2019), (3) effects of AI on jobs and society
Roland T. Rust (e.g., Autor and Dorn 2013; Frey and Osborne 2017; Huang
rrust@rhsmith.umd.edu and Rust 2018), and (4) managerial and strategic issues related
1
to AI (e.g., Fountaine et al. 2019; Huang and Rust 2020).
Department of Information Management, College of Management,
National Taiwan University, 1, Sec. 4, Roosevelt Rd, Taipei 10617,
The fourth literature stream, managerial issues related to
Taiwan AI, is currently dominated by consultants gravitating to the
2
Center for Excellence in Service at the Robert H. Smith School of
latest hot topic, and largely lacks a solid academic basis, albeit
Business, University of Maryland, 3451 Van Munching Hall, there are some recent studies trying to tackle strategic market-
College Park, MD 20742, USA ing issues. Examples include unstructured data for various
J. of the Acad. Mark. Sci. (2021) 49:30–50 31
areas of marketing (Balducci and Marinova 2018), analytics Multiple AI intelligences

for consumer value in healthcare (Agarwal et al. 2020), ma-
chine learning prediction for mobile marketing personaliza- Figure 1 illustrates that AI can play critical roles in all three
tion (Tong et al. 2020), in-store technology (e.g., robots, smart strategic marketing stages. It shows that there are multiple AI
displays, or augmented reality) for convenience or social pres- intelligences that a marketer can leverage: mechanical, think-
ence (Grewal et al. 2020), and AI for personalized customer ing, and feeling.
engagement (Kumar et al. 2019). We conceptualize AI as the use of computational machin-
To facilitate the strategic use of AI in marketing, we devel- ery to emulate capabilities inherent in humans, such as doing
op a three-stage framework, from marketing research, to mar- physical or mechanical tasks, thinking, and feeling; the mul-
keting strategy (segmentation, targeting, and positioning, tiple AI intelligence view considers that, rather than treating
STP), to marketing actions (4Ps/4Cs), for strategic marketing AI as a thinking machine, AI can be designed to have multiple
planning incorporating AI. This strategic AI framework is intelligences, as humans have, for different tasks. Ordered by
based on a more nuanced perspective of the technical devel- the difficulty with which AI can address them, there are me-
opment of AI, existing studies on AI and marketing, and cur- chanical, thinking, and feeling AI intelligences (Huang and
rent and future AI applications. It can be used for strategic Rust 2018; Huang et al. 2019).
marketing planning, for organizing the existing AI marketing Mechanical AI is designed for automating repetitive and
studies, and for identifying research gaps in AI marketing. routine tasks. For example, remote sensing, machine transla-
This paper contributes to the strategic application of AI in tion, classification algorithms, clustering algorithms, and di-
marketing by developing a framework that guides the strategic mensionality reduction are some current technologies that can
planning of AI in marketing in a systematic and actionable be considered mechanical AI.
manner. It is achieved by bringing together diverse AI litera- Thinking AI is designed for processing data to arrive at
tures on algorithms (e.g., Bauer and Jannach 2018; Davis and new conclusions or decisions. The data are typically unstruc-
Marcus 2015), psychology (e.g., Lee et al. 2018; Leung et al. tured. Thinking AI is good at recognizing patterns and regu-
2018), societal effects (e.g., Autor and Dorn 2013; Frey and larities in data, for example, text mining, speech recognition,
Osborne 2017), and managerial implications (e.g., Huang and facial recognition. Machine learning, neural networks,
et al. 2019) to explore what those literatures can tell us about and deep learning (neural networks with additional layers)
managing AI in marketing. Marketing is an applied field, and are some of the current methods by which thinking AI pro-
using the more foundational literatures to inform marketing cesses data. IBM Watson, expert systems, and recommender
practice is an important role for marketing academia. This systems are some current applications for decision making.
paper also contributes to strategic marketing research by pro- Feeling AI is designed for two-way interactions involving
viding a systematic and rigorous approach to identifying re- humans, and/or for analyzing human feelings and emotions.
search gaps that bridge strategic AI marketing practice and Some current technologies include sentiment analysis, natural
research. language processing (NLP), text-to-speech technology, recur-
rent neural networks (RNN), chatbots for mimicking human
speech, embodied and embedded virtual agents for human
Conceptual foundation interactions, and robots with customized hardware for sensing
affective signals (McDuff and Czerwinski 2018).
The marketing research–strategy–action cycle It is important to note two qualifications of this multiple AI
intelligences view. First, although we set up three intelli-
We propose a three-stage strategic planning framework based gences, the assignment of some applications to a particular
on the marketing research–marketing strategy–marketing ac- intelligence is mainly based on the purpose they are used
tion cycle. Similar cycles have been proposed, such as for. Sometimes those applications may have some elements
Deming’s (1986) plan-do-check-act cycle, but that cycle omits of more than one intelligence; thus, suggesting that the three
the role of strategy. Our cycle views strategic planning as a intelligences are fuzzy sets (Varki et al. 2000). For example,
circular process, starting from conducting marketing research facial recognition that is trying to identify someone is thinking
to understand the market, the firm, the competitors, and the AI (e.g., customs uses it to identify potential terrorists), while
customers; to developing strategies for segmentation, facial recognition that is trying to figure out somebody’s emo-
targeting, and positioning; and to designing specific marketing tional state from their facial expression is feeling AI (e.g.,
actions to execute the strategy. This cycle does not stop at advertisers use it to identify audience responses to ads).
marketing actions. The execution of marketing actions will Second, we do not have true feeling AI yet1; thus, the current
feed back to marketing research as market data, which consti-
tutes a continuous cycle for marketing research–strategy–ac- 1
According to Huang and Rust (2018), true feeling AI needs to be able to
tion, as illustrated in Fig. 1. recognize, emulate, and respond appropriately to human emotions.
32 J. of the Acad. Mark. Sci. (2021) 49:30–50
Marketing Action Marketing Research

• Standardization (mechanical AI) • Data collection (mechanical AI)
• Personalization (thinking AI) • Market analysis (thinking AI)
• Relationalization (feeling AI) • Customer understanding (feeling AI)
Marketing Strategy
• Segmentation (mechanical AI)
• Targeting (thinking AI)
• Positioning (feeling AI)
Fig. 1 AI and strategic marketing decisions
practice is to use thinking AI to analyze emotional data (e.g., Thinking AI provides personalization benefits, due to its
affective analytics) and two-way interactions (e.g., chatbots ability to recognize patterns from data (e.g., text mining,
and social bots). Emotional data are distinct from cognitive speech recognition, facial recognition). Any marketing func-
data, in that they are contextual, individual-specific, and typ- tions and activities that can benefit from personalized out-
ically multimodal (speech, gestures, and language). Such data comes should consider thinking AI. The most common appli-
are about the individual in context, meaning that feeling AI cations in marketing are various personalized recommenda-
needs to incorporate contextual and individual-specific data tion systems (Chung et al. 2009; Chung et al. 2016), such as
into modeling the emotional state of an individual. Netflix movie recommendations and Amazon cross-selling
recommendations.
Feeling AI provides relationalization benefits (i.e., person-
Multiple benefits of AI alizes relationships), due to its capability to recognize and
respond to emotions. Any marketing functions or activities
Each of the AI intelligences can deliver its unique benefit: that require interaction and communication, with the goal of
mechanical AI is best for standardization, thinking AI is good relational benefits (e.g., when customer lifetime value is high)
for personalization, and feeling AI is ideal for relationalization should consider feeling AI—one example being customer ser-
(Huang and Rust 2020). vice. A broad range of marketing functions involves feelings,
Mechanical AI provides standardization benefits due to its for example, customer satisfaction, customer complaints, cus-
ability to be consistent. In marketing, various forms of me- tomer moods, and emotions in advertising, etc., and can make
chanical AI have been used to provide a standardization ben- use of feeling AI.
efit; for example, collaborative robots (cobots) help with pack-
aging (Colgate et al. 1996), drones distribute physical goods,
self-service robots deliver service, and service robots auto- The strategic AI framework
mate social presence in frontline (Mende et al. 2019; van
Doorn et al. 2017). All these applications aim to generate We propose a three-stage strategic framework for using AI in
standardized, consistent, and reliable outcomes. marketing that leverages the three AI intelligences and their
benefits, as shown in Fig. 1. At the marketing research stage, intelligence, including data about consumers, their activities,
AI is used for market intelligence, including mechanical AI and their environments, can be collected if they use connected
for data collection, thinking AI for market analysis, and feel- devices (Cooke and Zubcsek 2017), product usage and con-
ing AI for customer understanding. sumption experience can be visualized with Internet of Things
At the marketing strategy stage, AI is used for the strategic (IoT) (Ng and Wakenshaw 2017), various advanced technol-
decisions of segmentation, targeting, and positioning. ogies and analytics can capture unstructured marketing activ-
Specifically, mechanical AI is ideal for discovering novel cus- ity data (Balducci and Marinova 2018), in-car sensors can
tomer preference patterns in unstructured data, thinking AI is track driving behavior for determining insurance premiums
ideal for recommending the best segment(s) to target, and (Soleymanian et al. 2019), and retail technologies, such as
feeling AI is ideal for communicating with the targeted cus- heat maps, video surveillance, and beacons, can be used for
tomers about the product.2 profiling and recognizing retail shoppers (Kirkpatrick 2020).
At the marketing action stage, AI is used for the benefits of These studies show that, given the repetitive, routine, but
standardization, personalization, and relationalization, indi- high-volume nature of market data, mechanical AI can collect
vidually or synergistically. Marketers need to decide which data efficiently at scale.
AI intelligence(s) to use for which marketing actions. For The data collection capability of mechanical AI is not lim-
example, payment and delivery are functions that can benefit ited to observable behavioral data; it can also be used to facil-
from standardization by using mechanical AI, such as auto- itate survey or experimental data collection to capture con-
matic payment and delivery tracking. Digital marketing can sumer psychographics, opinions, and attitudes. For example,
benefit from personalization by using thinking AI, such as human administration and supervision of ongoing surveys are
various recommendation systems. Customer service and no longer required and can be automated. SurveyMonkey and
f r o n t l i n e c u s t o m e r i n t e r a ct i o n c a n b e n e f i t fr o m SurveyCake are two commercial survey platforms that auto-
relationalization by using feeling AI, such as social robots mate survey design and data collection.
greeting customers and conversational AI providing customer
service. The discussion of the strategic use of AI in marketing Thinking AI for market analysis Thinking AI can be used to
action is organized in terms of the marketing 4Ps/4Cs, to bal- identify competitors in a well-defined market or outside op-
ance both the marketer and customer sides. Table 1 defines tions in a new market, and to derive insights for a product’s
various strategic elements of this strategic AI framework, competitive advantages (i.e., the way the product can do better
Table 2 illustrates marketing actions by using mechanical AI than competitors to meet customer demands). For example,
for standardization, thinking AI for personalization, and feel- supervised machine learning can be used for a mature market
ing AI for relationalization, with various examples and current where the market structure is stable and known to marketers,
and future scenarios,3 and Table 3 summarizes the existing whereas unsupervised machine learning can be used for a new
literature for using AI in each of the strategic elements. We market or spotting outside options where the market structure
discuss this three-stage framework in the following sections. and trends are unstable and unknown to marketers.
In marketing practice, predictive analytics are commonly
used to predict volatile market trends and customers’ hetero-
Marketing research geneous preferences. For example, Gap, the fashion clothing
brand, uses it to predict fast fashion trends to better meet
At this strategic stage, mechanical AI can be used for data customer needs, and Amazon uses it to predict a customer’s
collection, thinking AI for market analysis, and feeling AI future orders (i.e., anticipatory fulfilling).
for customer understanding. Existing studies have demonstrated various potential uses
of thinking AI for market analysis. Automated text analysis
Mechanical AI for data collection Mechanical AI can automate can be applied for consumer research (Humphreys and Wang
data collection about the market, the environment, the firm, 2018), for marketing insights (i.e., prediction and understand-
the competitors, and the customers. In the digitally connected ing) (Berger et al. 2019), and for analyzing consumer consid-
world, market data can be easily tracked and monitored. Data eration heuristics (Dzyabura and Hauser 2011). Machine
sensing, tracking, and collection are routine, repetitive tasks learning algorithms and lexicon-based text classification can
that can be easily automated by mechanical AI. be used to analyze various social media datasets (Hartmann
Existing studies have shown various ways of using meet al. 2019). Also, big data marketing analytics is now a main-
chanical AI for data collection. For example, customer stream approach for generating marketing insights (Berger
2
et al. 2019; Chintagunta et al. 2016; Liu et al. 2016; Wedel
The term “product” is used to refer to both tangible goods and intangible
and Kannan 2016).
services.
3
The two strategic stages, marketing research and marketing strategy, are less Specific applications include mapping market structures
relevant for this marketing practice illustration. for large retail assortments using a neural network language
Table 1 A strategic framework for AI in marketing
AI intelligence Mechanical AI Thinking AI Feeling AI

Strategic decision
Marketing research Data collection Market analysis Customer understanding

Automate continuous market Use marketing analytics to identify Use emotional data and customer
and customer data sensing, competitors and competitive analytics to understand existing and
tracking, collecting, advantages potential customer needs and wants
and processing
Marketing strategy (STP) Segmentation Targeting Positioning
Use mechanical AI to identify novel Use thinking AI to recommend Use feeling AI to develop positioning
customer preference patterns the best target segments that resonates with customers
Marketing action (4Ps/4Cs) Standardization Personalization Relationalization
Product/Consumer Automate the process and output of Personalize products based on Understand and meet customer
meeting customer needs and wants customer preferences emotional needs and wants
Price/Cost Automate the process of price setting Personalize prices based on Negotiate price and justify the cost
and payment customer willingness to pay interactively
Place/Convenience Automate customer Personalize frontline interactions Personalize experience for
access to product customer engagement
Promotion/Communication Automate communication with Customize promotional content for Tailor communication based on
customers personal communication customer emotional preferences and
reactions
Note. Product includes tangible goods and intangible services
model, by analyzing the co-occurrences of products in shop- 2018) and by Harley-Davidson to identify high potential cus-
ping baskets (Gabel et al. 2019), detecting copycat mobile tomers based on the company’s customer relationship man-
apps using a machine learning copycat-detection method agement database, personalizing the marketing campaign ac-
(Wang et al. 2018), and aiding social media content engineer- cordingly (Power 2017).
ing by employing natural language processing algorithms that In academic research, existing studies have shown various
discover the associations between social media marketing approaches of using feeling AI to understand customers. For
content and user engagement (Lee et al. 2018). example, the sentiment expressed by consumers in social me-
dia (e.g., online reviews, tweets), including explicit and im-
Feeling AI for customer understanding Feeling AI can be used plicit language and discourse patterns, can be analyzed to
to understand existing and potential customer needs and understand consumer responses using their own language
wants, for example, who they are, what they want, and what (Hewett et al. 2016; Humphreys and Wang 2018; Ordenes
their current solutions are. The major distinction between mar- et al. 2017), the interaction between conversational AI and
ket analysis and customer understanding is that the latter often customers can be enhanced by applying analytical mapping
involves emotional data about customer sentiments, feelings, to script appropriate response sequences that make customers
preferences, and attitudes. Thus, feeling AI can do a better job feel that they have a “conversation” with AI (Avery and
of understanding customers than mechanical AI and thinking Steenburgh 2018), consumer consideration heuristics can be
AI, due to its capability of analyzing emotional data. understood by machine learning (Dzyabura and Hauser 2011),
For existing customers, marketers can use feeling AI to and customer needs can be identified from user-generated
answer the questions of whether they are happy with the prod- content using convolutional neural network machine learning
uct and why. Existing customers’ preferences are more stable, (Timoshenko and Hauser 2019).
and a company has past and current transaction data for a
deeper understanding. For example, Affectiva partnered with Marketing strategy (STP)
Ford to create AutoEmotive sentiment analysis, to try to figure
out drivers’ emotional states. At this strategic stage, marketers can use AI for the three key
For potential customers, marketers can use feeling AI to strategic decisions: segmentation, targeting, and positioning.
understand what they want and why they are happy with com- However, before proceeding to specific STP decisions,
petitors or outside options. Potential customers’ needs and marketers need to decide the overall strategic positioning to
wants are more difficult to predict, and their emotional data guide their STP decisions. Huang and Rust (2017) propose a
are less available. In marketing practice, Albert AI, technology-driven approach to positioning a firm’s strategy
Adgorithm’s AI-driven marketing platform, has been used along the dimensions of standardization–personalization and
by RedBalloon to discover and reach new customers (Sutton transaction–relationship. A firm can pursue a commodity
Table 2 Illustrative examples and scenarios of using the three AI intelligences in various areas of marketing organized by 4Ps/4Cs
4Ps/4Cs Area of marketing Mechanical AI for standardization Thinking AI for personalization Feeling AI for relationalization
Product/ Product/branding • Track and monitor product adoption • Gap uses predictive analytics • Train bots to have brand
Consumer and acceptance for fashion trends personality
• Tailor Brands automate logo design • Computing creativity for • Sky uses ML to recommend
service innovation programs according to the
• Brand tracking through text viewer’s mood
analysis • Real-time conversations with
• AI diet personalizes diet consumers
algorithm for each consumer • Replika, a ML chatbot, provides
emotional comfort to consumers
by mimicking their styles
of communication
Customer service• Text-based chatbots handle massive • NLP chatbots analyze • NLP chatbots analyze and
amount of routine cases simulta- heterogeneous customer respond to customer emotions
neously voices (e.g., customers with • Cogito’s emotional AI analyzes
accents or issues are conversation and provides
contextual-dependent) guidance to customer agents
Price/Cost Pricing • Automatic payment • Personalized pricing • Price negotiation
• Apple Pay, Google Pay, PayPal, • ML for price optimization by
Amazon Payments, and Square product, channel, and customer
Place/ Retailing/frontline • Self-check-in and check-out • Macy’s On Call (NLP personal • Customer greeting robots (e.g.,
Convenience • Giant grocery’s Marty robot shopping app) Pepper)
identifies hazards • Alibaba’s FashionAI smart interact with customers in the
• HaiDiLao’s robots deliver hotpot mirrors display complementary frontline
soup from kitchen to table side items
• Robotic kiosks answer routine
questions
Distribution/ • Cobots automate packaging • Amazon uses predictive analytics • Amazon Go convenience store
logistics/ • Amazon Prime Air and UPS’s drone for anticipatory shipping uses facial recognition to
delivery delivery • Domino’s pizza uses self-driving identify customers
• IoT automates consumption tracking cars for delivery
and order refilling
• Self-service
Promotion/ Advertising/digital • Automate targeting and retargeting • AI writer creates and personalizes • Affectiva tracks audience feeling
Communication marketing • Automate media scheduling and content and personalizes ad messages
buying • Lexus car uses IBM Watson to • Wylei uses predictive AI to create
• Automate consumer webpage create commercial script and deliver personalized content
browsing using hotlink assignment • Kantar’s ad analytics help for engaging customers
• Automate real-time posting, advertisers create content • Kia used ML to identify social
bidding, and updating • Albert AI personalizes campaign media influencers for its 2016
• Automate push content, ad, and for Harley-Davidson Super Bowl campaign
notification to consumers’ devices • Personalized campaign content • Emotion sensing from post
based on analytical customer message or conversational
profiling content
• Personalize search using social
influence
• Tracking and measuring
campaigns
Note. The two higher strategic levels, marketing research and marketing strategy, are not included, due to them being less observable from marketing
practice
strategy that uses automated/robotic technology for efficiency, personalization strategy, the firm may want to have a big,
a relational strategy that cultivates the existing customers’ relevant, existing and potential customer database and let un-
lifetime value, a static personalization strategy that uses supervised machine learning explore patterns of preference or
cross-sectional big data analytics (e.g., like-minded cus- purchase behavior as the basis of targeting and positioning. If
tomers) for personalization, or an adaptive personalization a firm pursues the adaptive personalization strategy, the firm
strategy that uses longitudinal customer data for dynamic per- may want to use supervised machine learning to continue
sonalization over time. This strategic positioning will guide analyzing existing customers’ satisfactions/dissatisfactions
firms’ STP decisions. For example, if a firm pursues the static over time (which may not be big). However, if firms embrace
Table 3 Prior and current AI research organized by the strategic framework

Strategic decision
Marketing Data collection Market analysis Customer understanding

research • IoT visualizes usage and • IoT reconfigures product and service • Deep learning and NLP analyze
experience data (Ng and that shifts boundaries of Things (Ng customer perceptions (Ramaswamy
Wakenshaw 2017) and Wakenshaw 2017) and DeClerck 2018)
• Connected devices collect • NLP and ML map market structures • Sentiment analysis for social media
customer intelligence (Cooke and for large retail assortments (Gabel content understands consumer
Zubcsek 2017) et al. 2019) responses using their own language
• Various online platforms make • Lexicon-based and ML algorithms (Hewett et al. 2016 ; Humphreys and
unstructured big data available for text mining social media data for Wang 2018 ; Ordenes et al. 2017).
cloud computing to predict sales marketing research (Hartmann et al. • ML for understanding consumer
and consumption (Liu et al. 2016) 2019) consideration heuristics (Dzyabura
• Unstructured data for marketing • Big data marketing analytics for and Hauser 2011)
insights (Balducci and Marinova marketing insights (Berger et al. • Convolutional neural network
2018) 2019 ; Chintagunta et al. 2016; Liu identifies customer needs from
• Sensors tracking driving behavior et al. 2016; Wedel and Kannan 2016) user-generated content (Timoshenko
provide insurers individual-level • Analytical and intuitive AI for service and Hauser 2019)
driving data (Soleymanian et al. analytics (Huang and Rust 2018) • Consumer and smart object in IoT as
2019) • AI for solving marketing problems relationship journey (Novak and
• Retail tracking technologies, such (Overgoor et al. 2019) Hoffman 2019)
as heat maps, video surveillance, • Consumer resistance to personal
and Beacons collect in-store medical AI (Longoni et al. 2019)
shopper data (Kirkpatrick 2020) • Consumers may not prefer
identity-related automated products
(Leung et al. 2018)
• Anthropomorphized consumer robots
increase perceived warmth but
decrease liking (Kim et al. 2019)
Marketing Segmentation Targeting Positioning
strategy • Text-mining and ML automatically • ML and causal forests for targeting • Data mining for distilling a
(STP) categorize loan borrowers (Netzer cancer outreach marketing (Chen customer-based perceptual map
et al. 2019) et al. 2020) (Daabes and Kharbat 2017)
• Conjoint analysis with sparse • ML for optimizing the targeting of • Tourism positioning slogans in top
learning for consumer preferences promotions for new customers destinations emphasize affective
in multiple segments (Chen, (Simester et al. 2020) component (Gali, Camprubi, and
Iyengar, and Iyengar 2017) • ML techniques for field experiments Donaire (2017)
• Automated text analysis and data to identify the best targets for
correspondence analysis for proactive churn programs (Ascarza
psychographic consumer 2018)
segmentation in the art market • Digital consumer profiling for
(Pitt et al. 2020) targeting (Neumann et al. 2019)
• Data mining for tourist segments • Combining statistical and data-mining
based on the meaning of techniques to target customers (Drew
destinations to consumers (Valls et al. 2001)
et al. 2018) • Variational Bayes ML segments
• Retail big data for micro segment customers based on retailer-brand
customers (Dekimpe 2020) and channel (Danaher et al. 2020)
Marketing Product/ Standardization Personalization Relationalization
action Consumer • Consumers may not prefer • Topic modeling and ML for service • Train conversational AI to have brand
(4Ps/4- automated products that are innovation and design (Antons and personality (Wilson and Daugherty
Cs) related to their identity (Leung Breidbach 2018) 2018)
et al. 2018) • Adaptive personalization systems • AI personalizes engagement marketing
(Chung et al. 2009; Chung et al. (Kumar et al. 2019)
2016; Dzyabura and Hauser 2019; • Service interactions can be
Liebman et al. 2019) personalized by “substituting
• Deep learning for personalized information for interaction” (Glushko
point-of-interest recommendation and Nomorosa 2013)
(Guo et al. 2018) • AI intelligences can be used to engage
• Video-based automated recommender customers in the service journey
systems (Lu, Xiao, and Ding 2016) (Huang and Rust 2020)
Table 3 (continued)

Strategic decision
• Track brand reputation through

sentiment analyses social media
content (Rust et al. 2020)
• Customers perceive AI chatbots to be
to be less empathetic (Luo et al. 2019)
Price/ • Consumers’ online WOM can be • Multiarmed bandit algorithms from • Interpersonal likeability would impact
Cost used to automate price changes statistical machine learning the price negotiation outcome in a
based on ML tools (Feng et al. dynamically adjusted online prices B2B relationship (Pulles and
2019) with incomplete price information Hartman 2017)
(Misra et al. 2019)
• Machine learning based on Bayesian
inference optimize online pricing
with sparse and noisy data (Bauer
and Jannach 2018)
• Consumers’ private information for
price personalization (Montes et al.
2019)
Place/ • Mechanical AI automates service • Video-based garment recommender • Service robots surface acting (Wirtz
Convenience process (Huang and Rust 2018) (Lu et al. 2016) et al. 2018)
• IoT optimizes retail processes • Service robots deliver customized • Consumers resistance to medical AI
(Grewal et al. 2018) experience (Wirtz et al. 2018) (Longoni et al. 2019)
• Service robots interact with scale • One-voice strategy for customer en-
and consistency (Wirtz et al. gagement across diverse interfaces
2018) enabled by machine-age technologies
• Service robots automate social in a customer journey (Singh et al.
presence in frontline (Mende 2020)
et al. 2019; van Doorn et al.
2017)
Promotion/ • Automate classification of social • Intelligent purchasing, dynamic • NLP analyzes social media ad content
Communication media content and sentiment pricing, targeting, and retargeting enhances consumer engagement (Lee
(Hartmann et al. 2019) (Kietzmann et al. 2018) et al. 2018)
• AI automates media buying • Content engineering by ML and NLP • AI chatbots for outbound sales calls
process (Chen et al. 2019) (Lee et al. 2018) (Luo et al. 2019)
• Combine ML and NLP to detect • NLP algorithms analyze Facebook
copycat apps (Wang et al. 2018) messages to examine the association
• AI transforms advertising creative between social media content and
processes (i.e., programmatic user engagement (Lee et al. 2018)
creative) (Chen et al. 2019) • Personalized ad backfires if privacy
• Personalized mobile marketing concern is heightened (Kim, Barasz,
strategies center on 4Ps (Tong et al. and John 2019)
2020)
a data-driven approach to STP, it can rely more heavily on AI Segmentation

to explore the STP possibilities.
In general, this stage of strategic decisions relies more on Segmentation is to slice a market into pieces, with customers
thinking AI, for its capability of processing data to arrive at in each piece having unique needs and wants, for example,
new conclusions or decisions. However, it is worth reiterating using gender to slice the shoe market into male and female
that the assignment of some applications to a particular intel- shoes segments; and using price and quality to slice the air
ligence is mainly based on the purpose an application is used travel market into budget and premium airlines segments.
for. For example, when thinking AI becomes completely rou- Mechanical AI, especially the various mining and grouping
tinized, as is often the case in segmentation applications, it techniques, has the strength of identifying novel patterns from
shares many of the characteristics of mechanical AI, because data.
it simply identifies patterns from data routinely and repetitive- AI segmentation is flexible, in that it can disaggregate the
ly, without involving much about the purpose of making new market into segments of one (i.e., each individual customer is
decisions (e.g., segmentation but not retargeting). a segment) and can aggregate scattered long tails into one
segment. Wang et al. (2017) demonstrate that transfer learning brand positioning or advertising positioning for its association
can be used to model the tail of the distribution, by learning with customer perceptions and communications to maintain a
from the head of the distribution and transferring the learning desirable perception. Daabes and Kharbat (2017) demonstrate
to the data-poor tail. This flexibility in aggregation and disag- how data mining techniques can be used to distill a customer-
gregation allows marketers to find the right size of segment. based perceptual map, as an alternative to marketer knowl-
Existing studies have shown how data mining can be used edge, from mining customers’ perceptions.
to uncover patterns that are difficult for human marketers to Compared with the mechanical-based segmentation and
see. For example, text-mining and machine learning can be the thinking-based targeting, positioning is more about speak-
used to automatically process and analyze loan requests to ing to customers’ hearts, typically as a positioning statement
slice borrowers into good customers (those will pay back the or slogan in promotional communication. Gali et al. (2017)
loan) and bad customers (those will not) (Netzer et al. 2019), find that tourism positioning slogans in top destinations tend
automated text analysis and correspondence analysis can be to emphasize the affective component.
used for psychographic consumer segmentation in the art mar- Some successful positioning statements help brands to oc-
ket (Pitt et al. 2020), data mining can be used to obtain tourist cupy a unique position in customers’ minds and thus succeed
segments based on the meaning of destinations to consumers, in the market for a long time. For example, Nike’s “Just do it,”
that is better than the classic clustering methods (Valls et al. Apple computer’s “Be different,” and McDonalds’ “I’m lov-
2018), and retail customers can be micro-segmented based on ing it” all communicate with customers by speaking to their
their preferences for personalized recommendation (Dekimpe hearts. Feeling AI, such as feeling analytics, is ideal for this
2020). strategic decision to help develop compelling slogans by un-
derstanding what resonates with target customers.
Targeting Academic research on this decision is sparse, indicating a
research gap for using feeling AI to create compelling
Targeting is choosing the right segment(s) on which to focus positioning.
the firm’s marketing actions. Slicing the market is more me-
chanical and can be done automatically by mechanical AI,
Marketing action
given the relevant data. However, choosing the right segment
requires domain knowledge, judgement, and intuition.
At this strategic stage, marketers can use mechanical AI for
Various technologies and analytics have been used for
standardization, thinking AI for personalization, and feeling
targeting, such as search engines using keywords searched
AI for relationalization (Huang and Rust 2020). Depending on
and browsing history to target search consumers, and social
which benefit is desirable, a marketer can use multiple AI
media platforms using interests, content, and connections to
intelligences individually or collectively. We illustrate the
target social media consumers (Liu 2020). The representative
use of AI intelligences in various areas of marketing with
AI for this decision is recommendation engines that can rec-
examples and current and future scenarios, and support the
ommend various potential targets for marketing managers’
illustration using the existing literature. The discussion is or-
final verdict, and predictive modeling that can be used to
ganized by marketing 4Ps (product, price, place, and promo-
choose which segment to target.
tion) (Kotler and Keller 2006) and the corresponding 4Cs
Existing studies show that various thinking AI can be used
(consumer, cost, convenience, and communication)
for this purpose. Examples include targeting customers using a
(Lauterborn 1990) to emphasize that the 4P actions need to
combination of statistical and data-mining techniques (Drew
be able to deliver consumer benefits.
et al. 2001), screening and targeting cancer outreach marketing
using machine learning and causal forests (Chen et al. 2020),
optimizing promotion targeting for new customers using vari- Product (consumer)
ous machine learning methods (Simester et al. 2020), identify-
ing the best targets for proactive churn programs from field Product (consmer) actions include goods and services as of-
experimental data using machine learning techniques (Ascarza ferings to meet the consumer’s needs and wants. Such actions
2018), and profiling digital consumers for targeting using on- typically include product design, packaging, branding, and
line browsing data (Neumann et al. 2019). returns, and the associated customer services in these activi-
ties. We illustrate this decision using product/branding and
Positioning customer service, with product/branding representing the
“product” side and customer service representing the “con-
Positioning bridges product attributes and customer benefits sumer” side. Product and branding are related in that branding
by finding a competitively advantageous position for the prod- is the identity (e.g., name, symbol, logo) of a product. We put
uct in customers’ minds. This term is often associated with them together to maintain the simplicity of the table.
Product/branding focuses on product creation (including Mechanical AI, such as text-based chatbots, is widely used
tangible goods R&D and production, and service innovation online to handle a massive amount of routine customer ser-
and process) and branding (i.e., the identity of a product). vice. Most customer questions can be answered by such bots.
Mechanical AI can be used for product/branding actions As long as such automation is not related to customers’ iden-
that can benefit from standardization. For example, brand logo tity (Leung et al. 2018), it is easy to implement, cost-efficient,
design can be automated by a decision-tree like machine and scales up easily.
learning using multiple-choice questions, allowing small bud- Thinking AI, such as natural language processing chatbots,
get marketers to have AI-assisted branding (Avery 2018). can handle more diversified customers and idiosyncratic is-
Product adoption and acceptance can be tracked and moni- sues (e.g., multicultural customers with accents and
tored automatically. While enjoying the standardization ben- contextual-dependent complaints). This is an AI version of
efit of mechanical AI, one existing study is cautious about the old telephone menu, except that customers are talking to
automating product decisions, when those products are related chatbots, rather than human customer service agents.
to consumers’ identity (Leung et al. 2018). Although a recent study shows that customers may not feel
Thinking AI can be used for product/branding actions that comfortable yet about talking to chatbots (Luo et al. 2019),
can benefit from personalization. For example, marketing ana- with the wider acceptance of AI and the further advance of
lytics can predict market trends for product design that cater chatbots, we can expect the acceptance to increase over time.
more precisely to target customers’ preferences, big data analyt- Feeling AI, such as Cogito’s emotional AI systems, can
ics can be used to inform product development to quickly adapt analyze the pace of speaking, energy and empathy, and com-
to consumer trends and changing preferences (Dekimpe 2020), mon errors of conversations, and gives in-call guidance to
topic modeling can advance service innovation and design customer service agents in call centers that make the conver-
(Antons and Breidbach 2018), adaptive systems can be used sations more natural and engaging.
to personalize service to each consumer’s preference (Chung
et al. 2009; Chung et al. 2016; Dzyabura and Hauser 2019;
Liebman et al. 2019), and deep learning can be used to person- Price (cost)
alize point-of-interest recommendations (Guo et al. 2018).
Feeling AI can be used for product/branding actions that Price (cost) action includes the tasks of payment, price setting,
can benefit from relationalization. For example, conversation- and price negotiation, which is the cost that the consumer pays
al AI can be trained to have brand personality to interact with for the product.
customers (Wilson and Daugherty 2018), machine learning The payment task is routine and can be handled best by
can recommend TV programs based on the viewer’s mood, mechanical AI. Apple Pay, Google Pay, PayPal, Amazon
brands can track their reputation through text and sentiment Payments, and Square are some popular automatic payment
analyses tweets, reviews, and social media posts (Rust et al. methods for online marketers.
2020), and chatbots can mimic customers’ communication The price setting task is calculation-intensive and analyti-
style to provide emotional support. Kumar et al. (2019) pro- cal, which is the strength of thinking AI. Misra et al. (2019)
vide a systematic exploration about the role of AI in person- demonstrate that multiarmed bandit algorithms from statistical
alized engagement marketing, an approach to create, machine learning can dynamically adjust online prices in real-
communicate, and deliver personalized offering to time, even if price information is incomplete. Bauer and
customers. Huang and Rust (2020) show that feeling AI can Jannach (2018) show that a machine-learning framework
be used to engage customers in service interaction. based on Bayesian inference can optimize online pricing even
From the consumer side, existing studies further show that when data are updated frequently, and are sparse and noisy.
customers have varying responses and attitudes toward using Prices can also be personalized by incorporating consumer
AI products. For example, consumers may be resistant to per- online WOM (Feng et al. 2019) and consumers’ private per-
sonal medical AI (Longoni et al. 2019), identity-based con- sonal information (Montes et al. 2019). Dekimpe (2020) pro-
sumption automation (Leung et al. 2018), and anthropomor- poses that retailers can use big data to optimize dynamic best-
phized consumer robots (Kim et al. 2019; Mende et al. 2019). response pricing algorithms that consider consumer choices,
These studies put a boundary condition for marketers when competitor actions, and supply parameters.
using AI in the product/branding actions to generate positive The price negotiation task is interactive; thus, feeling AI is
customer responses. in a better position to undertake this task. Pulles and Hartman
Customer service is emotionally charged, yet is also costly. (2017) hypothesize that interpersonal likeability would impact
A marketer can handle customer service using the three AI the price negotiation outcome in a B2B relationship, suggest-
intelligences to balance the cost/satisfaction tradeoff in serv- ing that interaction, communication, and sentiment may be
ing customers. critical for price negotiation.
Place (convenience) logistics, and delivery, to provide convenience benefits to

customers.
Place (convenience) is the way that the consumer can access Tasks in distribution are mostly mechanical, routine, and
the product. We discuss two broad categories of place action: repetitive; thus, the standardization benefit of mechanical AI
Retailing and frontline, virtually or physically, in which inter- fits well. We have seen cobots for packaging, drones for de-
actions play a key role; and distribution, logistics, and deliv- livery, IoT for consumption tracking and order refilling, and
ery, in which convenience is the key. self-service technologies for delivering service to customers
Retailing/frontline is the area of marketing that most em- directly. Such standardization provides a convenience benefit
ploys embodied AI (i.e., robots) to facilitate frontline to customers.
interactions. Moving up to the thinking AI level, we have seen that a
Mechanical AI can be used to automate backend marketing customer’s future orders and refills can be anticipated by pre-
processes and frontend interactions. In the backend, service dictive analytics, and ordered products can be delivered to
processes can be automated (Huang and Rust 2018) and retail customers using autonomous cars equipped with facial recog-
processes can be optimized using IoT (Grewal et al. 2018). In nition technology (e.g., JD.com and Domino’s use self-
the frontend, service robots can interact with scale and consis- driving cars for delivery).
tency (Wirtz et al. 2018), and can automate social presence in So far, feeling AI is not as widely used as the other two AI
the frontline (Mende et al. 2019; van Doorn et al. 2017). intelligences for this marketing action, due to distribution’s
Frontline service robots are common; for example, Giant gro- mechanical and thinking nature.
cery uses the robot Marty to identify hazards in store (e.g.,
detecting milk spilled on the floor) and HaiDiLao hotpot uses Promotion (communication)
robots to deliver soup base from kitchen to table side. Grocery
shopping is typically repeat purchase, which does not involve Promotion (communication) is the marketing communica-
too much interaction, communication, and emotion, and thus tions between the consumer and the marketer. It can include
using mechanical AI to automate the marketing function is personal selling, traditional mass media advertising, and more
desirable. commonly nowadays direct marketing, database marketing,
At the thinking level, due to the direct customer contact and digital marketing (social media marketing, mobile mar-
nature of retailing, AI is used to facilitate in-store shopping keting, search engine optimization, etc.). All these can benefit
for individual customers. Amazon Go, an experimental gro- from AI intelligences.
cery store, uses facial recognition technology to identify and Mechanical AI is ideal for automating various repeti-
remember each customer, Macy’s On Call, a mobile shopping tive, routine, and data intensive functions of promotion
personal aid, provides in-store information to help customers (Huang and Rust 2018). Most of these are about promo-
locate items they are looking for, and Alibaba’s FashionAI tional media planning and executions. Examples include
system uses smart mirrors in sales floor and changing rooms automating advertising media planning, scheduling, and
to display items that each customer selects and suggests com- buying; automating search campaigns execution, key-
plementary items. words researching, and bidding; automating social media
Feeling AI can be used to enhance interaction and engage- targeting, retargeting, and posting. Especially considering
ment. For example, service robots can easily do surface acting the real-time nature of digital marketing, such automation
(Wirtz et al. 2018), and “one-voice” AI can enhance customer greatly aids marketers’ efforts in the labor-intensive, high-
engagement by integrating various interfaces involved in a time-pressure process.
customer’s journey (Singh et al. 2020). At the feeling level, Thinking AI has great potential for promotion content cre-
various embodied robots are used to engage customers and ation and personalization. For example, AI content writers can
optimize their experience. For example, Pepper robots are facilitate the generation of ad or post content. A recent exam-
used by Marriott to greet and interact with customers. Hotels ple was a Lexus car commercial that used IBM Watson to
and travel typically involve more interactions and more emo- create the “Driven by Intuition” commercial script. Content
tions, and thus feeling AI naturally suits. Nevertheless, mar- can be personalized and optimized to different customer pro-
keters need to be cautious, in that anthropomorphized robots files at different locations and different times. Kantar
are found to increase perceived warmth but decrease liking Analytics uses content analytics to help advertisers create con-
(Kim et al. 2019); thus, in the case of embodied frontline tent that shortens the idea-to-value time and maximizes con-
robots, marketers need to take the appearance of robots into tent effectiveness (Gopinath 2019).
consideration. Feeling AI can be used to track real-time customer re-
Distribution/logistics/delivery is an area of marketing in sponse to promotional messages (like, dislike, disgusted, fun-
which many functions and processes can be highly automated; ny, etc.) and then adjust what to deliver and what to emphasize
including packaging, inventory, warehousing, supply chain, in both media and content. At the feeling level, more real-time
and accurate emotion sensing from posted messages can better Market analysis Current market analysis, although marching
engage customers and provide a better interaction experience toward machine learning-based analysis rapidly, still relies
(Hartmann et al. 2019; Lee et al. 2018). heavily on statistical analysis to analyze structured data for
marketing insights. It is also common for firms to purchase
third-party data and analysis, especially for external market
and competitor analysis. Such analysis tends to be standard-
Managerial implications
ized across firms (with limited degree of customization), and
thus the insights derived from it are less useful for deriving a
Our framework provides a roadmap for marketers to imple-
unique value proposition. Firms also monitor and analyze
ment various AI intelligences in marketing, strategically and
first-party data, especially for firm marketing resource analy-
systematically. Table 4 summarizes the managerial implica-
sis and for existing customer analysis for which internal data
tions of the framework by contrasting current marketing prac-
are available.
tice with the emerging AI-enabled marketing practice. The
By contrast, big data and machine learning-based analytics
two practices should be viewed as two ends of a continuum,
are the emerging approach for marketing insights. Online re-
with more AI intelligences expected to be used for more stra-
views, opinions, and behaviors all can be mined, and data can
tegic elements at the three stages over time. We discuss these
be in text, image, audio, or video. When the question at hand
implications below.
is clear (e.g., a mature brand), supervised machine learning
can be used to conduct theory-driven analysis, whereas when
Implications for marketing research the question at hand is unclear (e.g., a new brand), unsuper-
vised machine learning can be applied to obtain data-driven
At this stage, marketers need to decide (1) how to use AI to insights. Balducci and Marinova (2018) offer a detailed de-
identify competitors (including competitors in the same indus- scription of various methods of analyzing unstructured data in
try and outside options) and to understand competitive advan- marketing. More advanced approaches to marketing analysis
tages (i.e., the way the product can do better to meet cus- include using deep learning methods such as predictive ana-
tomers’ needs), and (2) how to use AI to discover and under- lytics, computational creativity, personalization algorithms,
stand existing and potential customers (i.e., who they are, and natural language processing systems, to come up with
what they want, and what their current solutions are) and to intuitive suggestions for marketing strategies.
understand their preferences and feelings. For firms that em-
brace a theory-driven approach to marketing strategies, data Customer understanding Current practice relies heavily on
and intelligences resulting from this stage play a critical role. focus groups to gain qualitative insights about customers.
Focus groups are time consuming and labor intensive, not to
Data collection Currently, surveys, experiments, interviews, mention not representative. Marketers also observe cus-
panels, and sales data are still the major approaches for mar- tomers’ behaviors and choices, and their reactions to promo-
keters to obtain data, even though the administration of these tions to understand their preferences and the underlying
methods can be partially automated or facilitated by technol- reasons.
ogy. Surveys and experimental methods tend to be more the- By contrast, data about customers’ feelings, moods, and
ory driven, while other methods tend to be more data driven. emotions can be obtained directly from customers’ interaction
Marketers often also rely on thirty-party syndicated data (e.g., with AI (e.g., conversational bots), rather than inferred from
YouGov), especially for external data that are difficult for the psychometrics, using conversational bots and analyzed using
firm to collect. These data are typically delayed, out of con- feeling analytics (e.g., posts on social media, voice recordings
text, and ad hoc, meaning that they are collected periodically, of customer interactions, and chat transcripts). Feeling analyt-
after the fact (after consumption has occurred), and not during ics can identify customer insights with scale and cost-effi-
data generation. ciently. Given that emotional data are personal and in context,
By contrast, emerging practices automate most of the data understanding customers in context provides richer insights
collection by connecting technology (e.g., IoT, social network- about who they are and what they like.
ing sites, mobile apps), sensor technology (e.g., remote sensing,
detection), and wearable technology (e.g., smart watch, Fitbit). Implications for marketing strategy (STP)
These mechanical AI approaches track and capture real-time
data when they are occurring. Thus, the data are in context, At this stage, marketers can leverage the three AI intelligences
about the customer, and during the consumption experience. for segmentation, targeting, and positioning, respectively. For
Such spontaneous data collection tends to be more data driven, firms that embrace a data-driven approach to marketing strat-
but if theories can be developed priori to guide and update the egy, this stage may play a bigger role than conclusions derived
continuous data collection, they can be theory driven as well. from marketing research.
Table 4 Managerial and research implications based on the three-tier strategic framework
Current practice Emerging practice Future research
Marketing research
Data collection • Scattered surveys, experiments, interviews, • Integrated multiple sources of data collected How would privacy and data security
panels, sales (e.g., scanner data), and automatically by mechanical AI (e.g., IoT, affect marketing’s use of AI for data
syndicated data (e.g., TNS) sensors, and wearables) collection?
• Semi-structured behavioral big data • Unstructured behavioral, opinion, and sentiment
• Data are delayed, out of context, and ad hoc data
• Data are real-time, during a customer’s
experience, and in context
Market analysis • Statistical analysis for structured data • Big data analytics for unstructured data How to balance data- and
• Theory-driven hypothesis testing • Data-driven exploration theory-driven market analysis?
Customer • Qualitative focus groups for customer insights • Feeling analytics for customer insights with scale What algorithms and models are
understand- • Observe customer behavior, choice, and • Understand customers in their own words and needed for feeling AI?
ing reactions to marketing actions expression
• Psychometrics for inferring customer thinking • Direct insights from customer interactions with
and feeling conversational AI
• Deep learning and NLP for contextual customer
thinking and feeling
Marketing strategy (STP)
Segmentation • Segment based on a few variables a priori • Discover novel segments based on unlimited How best to visualize segmentation?
• Aggregate customer equity model (e.g., number of variables
acquisition rate, churn rate) • Individual customer lifetime value model (i.e.,
segmentation of one)
Targeting • Target segments • Target individuals more What happens when the customer is
• Trade off target segment size and effectiveness • Flexible targeting and retargeting (aggregate and AI?
disaggregate)
Positioning • Human judgment and intuition • AI-based optimization of positioning How should marketers and AI
collaborate, to resonate with
customers?
Marketing action (4Ps/4Cs)
Product/ • Conjoint analysis decides attribute levels • Mechanical AI automates service and product How is AI best used in developing new
Consumer • Market testing process products to meet customer needs
• Aggregate sales as customer adoption • Thinking AI for product innovation and wants?
• Product life cycle • Feeling AI for service interaction
• Customer value
Price/ • Price menu on websites or apps • Mechanical AI automates price setting and How to manage AI-based price nego-
Cost • Segment-based price discrimination updating tiation?
• Humans handle price negotiations • Thinking AI personalizes pricing
• Feeling AI negotiates and bids prices
Place/ • Self-service/physical distribution • Mechanical AI automates distribution, logistics, How to manage customer
Convenience • Unskilled labor provides homogeneous and delivery disengagement due to place
shopping assistance • Thinking AI personalizes shopping automation?
• Frontline employees’ variable emotional labor • Feeling AI interacts with customers
Promotion/ • Skilled labor-intensive media planning • Mechanical AI automates promotional executions How to use AI to build strong
Communication • Relying on creative executives’ talent and media planning relational bonds?
• Delayed response to customer reactions to • Thinking AI for content creativity
promotions • Feeling AI for emotion sensing and reaction
Segmentation The current approach relies on the marketer’s marketers make the aggregate segments more personal and
intuition and domain knowledge to choose a limited number relatable.
of segmentation variables with which to slice the market, such By contrast, when data mining is used to segment the mar-
as demographics, psychographics, and behavioral variables. ket, it no longer requires marketers to decide segmentation
Such an approach sees customers as aggregate, not individual. variables on an a priori basis, because unsupervised machine
For example, some customer equity models focus on learning can discover the patterns itself. A virtually unlimited
segmenting customers based on their acquisition and churn number of variables can be used to slice the market in a novel
rates, and do not see them as individually unique (e.g., way that often goes beyond any pattern that human marketers
Blattberg and Deighton 1996; Gupta et al. 2004). Artificial can see. It is like the customer lifetime value model, in which
personas thus are often applied to these segments to help each customer is valuable in some way. The Target store
knowing a daughter is pregnant before her father knows, by 2020). This real-world example illustrates that positioning can
mining the daughter’s purchase patterns, is a classic example. be expected to be a human-AI collaboration for the immediate
future.
Targeting Currently targeting mostly uses marketers’ subjec-
tive judgment, based on the resources, the competitive advan- Implications for marketing action
tage of the firm, and the value of the segment to the firm. It is
typically at the segment level (not individual level), and often At this stage, marketers can leverage the three AI intelligences
trades off segment size for effectiveness. for the 4Ps to serve the 4Cs. The key questions to answers are
By contrast, after very refined segmentation, it is think- which AI to use and how to use it for marketing actions.
ing AI’s turn to recommend the best segment(s) to target.
It is very well likely to be a segment of one, since per- Product The current practice for product decisions is to use
sonalization is the strength of thinking AI. With the capa- conjoint analysis to decide what levels of product attributes to
bility to slice the market in unlimited ways and at the include in product development, use test markets to decide
individual customer level, targeting in emerging practice whether and to what degree the product will be accepted,
is more commonly at the individual customer level. For and use aggregate sales results after the launch of the product
example, online ads use cookies to target individual cus- as a proxy for customer feedback.
tomers by following them around, wherever they go on The emerging practice is to use mechanical AI to automate
the Internet. The new targeting also is flexible, because it production and service process (e.g., Huang and Rust 2020), use
can aggregate individual customers into a segment, if they thinking AI, such as cognitive technology, to facilitate product
have similar preferences (e.g., like-minded customers rec- research and development (which is currently more common in
ommendation, aggregating long-tail customers even when drug development), and use feeling AI, such social robots and
each individual customer may not be valuable), or it can conversational bots, to interact with customers, from which they
disaggregate a segment, if heterogeneity within the seg- obtain real-time, first-hand customer feedback about the prod-
ment becomes manifest. Targeting involves not just iden- uct. Such a process can become an adaptive loop that improves
tifying segments but also determining whether they the product continuously based on customer feedback.
should be pursued. Whether they should be pursued or
not is a matter of predicting the outcome if they pursue, Price The current practice is to list prices on retail stores, websites,
and prediction at the individual level is only scalable with or mobile apps, set the prices discriminately, based on segments,
the help of AI. and have salespeople handle price negotiations. Price menus
offline are difficult and labor intensive to change, and price menus
Positioning online, though easier to update, are also easier to be compared.
Price setting typically requires careful and extensive calculation,
Positioning is currently a human task, for it involves judgment, taking various factors into consideration. Price negotiation is more
intuition, and creativity that machines are not particularly good at of an art than a science, especially for big ticket items.
it yet (Davis and Marcus 2015; Schoenick et al. 2017). Kelly The emerging practice is to use mechanical AI to automate
(2019) argues that creativity is not just about novelty but is also price setting and changes, thinking AI for price personaliza-
about social acceptability. A novel idea has to be accepted by tion, and feeling AI for price negotiation. Price updating is a
community to be deemed as creative. Because creativity is so- simple routine task, price setting can be achieved by the pow-
cially embedded, a good positioning is in the eye of the targeted erful calculating machine, thinking AI, and can be personal-
customers. Although we have seen an increasing number of ex- ized taking individual customers’ preferences and sensitivity
amples of AI participating in the creative process, for example into consideration. Price negotiation can be done when feeling
using AI to compose its own music and to write short stories, AI detects customer reactions to the offered price in real-time.
there is still a long way to go for AI to be as creative as humans
while still maintaining strategic relevance. For example, the Place The current practice relies on self-service to automate
script of the 2018 Lexus car “Driven by Intuition” TV commer- routine delivery and labor-intensive physical distribution for
cial was created by AI by applying the machine learning ap- ordering processing, materials handling, and delivery; un-
proach. Lexus fed machines with award-winning luxury ads, skilled labor at the frontline to offer homogeneous assistance,
Lexus brand data, and emotion data, that were shown to connect and frontline employees for emotional labor.
with viewers, to tell the story about how Lexus generated the new Distribution, logistics, and delivery can mostly be automat-
ES executive saloon car. The commercial appears to have face ed with mechanical AI, and is a fast-growing emerging prac-
validity as a luxury car commercial. However, this commercial tice, for example, product tracking systems for firms to track
may not be very strategic because the ad has an unclear customer where the product is in the supply chain, and for customers to
segmentation and ambiguous positioning (Rust and Huang track when they can expect to receive the product. Thinking
AI, such as personal shopping assistants, has been used to capability. How should privacy and data security be han-
assist customers about where to find the product. Feeling AI, dled? The existing studies suggest that both data and
such as conversational bots, can be used to display emotions consumer characteristics need to be considered. It has
in service interaction without the need to actually experience been shown that some types of data are more sensitive,
emotions (Wirtz et al. 2018). such as personal data, financial data, health data, or
medical records. Agarwal et al. (2020) urge healthcare
Promotion The media planning part of promotion has a higher organizations to find ways of protecting patient privacy
degree of automation by mechanical AI, due to the repetitive as a competitive advantage, because healthcare consumers
nature of the task. The content creation part of promotion, though are especially weak in protecting their own data. It has also
having a lower degree of automation, is increasingly handled by been shown that some consumers are more sensitive to
thinking AI, such as AI writers, to generate content on its own, or data sharing. Thomaz et al. (2020) identify two types of
to stimulate human creativity. Customers’ reactions to promotion consumers who vary in willingness to share information.
are still mostly measured using traditional marketing research They advise that marketers need to understand which type
methods. The emerging practice is to use feeling AI, such as of consumers they are dealing with when providing per-
feeling analytics, to sense, react, and adjust promotions in real- sonalization. Some methods of protecting marketing data
time based on customers’ emotional reactions. have been proposed (Schneider et al. 2017, 2018). Future
research will need to explore more delicate approaches to
handle the ubiquitous data collected by AI.
Directions for future research & How to balance data- and theory-driven market analysis?
Traditional market research has the strength of hy-
The framework lays out a stage-by-stage circular process for pothesis testing, whereas the emerging machine
using different AI intelligences in marketing. It provides rich learning approach to analyzing unstructured big data
implications for future research based on the use of AI intel- has the merit of data exploration. How to balance
ligences. We discuss these implications, organized by the the two approaches of inquiry and benefit from
three AI intelligences. The last column of Table 4 illustrates them? The Silicon Valley viewpoint is that predic-
one example research question for each element. tion, driven by data, is everything. Balducci and
Marinova (2018) show a data-driven approach to
Marketing research marketing management and research and Ma and
Sun (2020) argue for machine learning methods for
When mechanical AI is used for data collection, it makes both marketing research, while Lehmann (2020) proposes
the competition and the customer more transparent, making a method to blend theory and data in the evolving
the governance of privacy issues more central to marketers. world of marketing research. These practices and
When thinking AI is used for market analysis, it turns theory- studies suggest that AI will expand the role of ex-
driven marketing research into data-driven, resulting in a de- ploration in the development of theory; thus, inves-
bate about whether a data or theory approach to marketing tigating how the two approaches can be balanced,
research should be embraced. When feeling AI is used for integrated, or blended would allow marketing to le-
customer understanding, it is as if AI can really understand verage the benefit of data-driven market analysis.
emotions, when we don’t have true emotional machines yet. & What algorithms and models are needed for feeling AI?
These issues give rise to various future research topics. Much of the progress in technical algorithms has been
made in the area of deep learning neural networks, applied
& How would privacy and data security affect marketing’s to thinking AI problems, such as personalized advertising.
use of AI for data collection? When mechanical AI is used The current neural network-based machine learning is
to collect and integrate multiple sources of real-time, in- mainly for prediction, rather than for understanding. We
context consumer data, the risk of privacy infringement need algorithms to understand emotions and react to emo-
and the consequences of a data breach are much higher. tions appropriately (Rust and Huang 2020). There are dif-
It is more difficult to keep such “all-in-one” data anony- ferent ways to understand emotions, such as understand-
mous, and consumers are more sensitive to data sharing ing emotions in written language (text-based sentiment
and data breaches. We have seen the public outcry when analysis), in oral conversation (natural language process-
Facebook permitted the unauthorized licensing of 30 mil- ing), or in facial expression (facial recognition). Although
lion user accounts to Cambridge Analytica, with its brand there is some work already in using thinking AI to analyze
reputation plummeting dramatically in two weeks (Rust emotional data (e.g., affective analytics), with respect to
et al. 2020). In the AI age, both competition and customers facial recognition and natural language processing,
become more transparent, due to AI’s data collection models in this area are still fairly rudimentary.
Marketing strategy Marketing action
For the three strategic elements of segmentation, targeting, For the four strategic elements of 4Ps/4Cs, as illustrated in
and positioning, mechanical AI can slice the market based Table 4, all three AI intelligences can be applied to each ele-
on virtually an unlimited number of variables, thinking AI ment. We illustrate one research question for each of the ele-
can recommend the target, even if it is a segment of AI cus- ments, highlighting the level of AI intelligence that is likely to
tomers, or a combination of AI and human customers, and help the most.
feeling AI can team up with marketers to come up with crea-
tive positioning that touches consumers’ hearts. We discuss & How is AI best used in developing new products to meet
these implications. customer needs and wants? Thinking AI can be innovative
and creative, but it may not be straightforward as to wheth-
& How best to visualize segmentation? The emerging AI ap- er the new products will meet customer needs and wants,
proach to segmentation can be based on an unlimited num- because many of those needs and wants are implicit, rather
ber of variables. This creates a challenge as to how best to than explicit. This is the challenge of converting product
visualize for marketers to make sense of these multi- attributes into consumer benefits. Rogers’ (1962) diffu-
dimensional segmentation results for targeting. Thinking sion of innovation theory suggests that both consumer
AI can easily decipher such complexity, but human mar- and product innovativeness matter for new product accep-
keters may not. Earlier studies have attempted to visually tance. AI, with its capability to find patterns and regular-
display three-dimensional segmentation and positioning ities in data that otherwise would be hard to see, can be
from a larger set of variables (e.g., Donthu and Rust used to uncover consumers’ implicit needs and wants, and
1989; Rust et al. 1992). There are some existing tools, but match those needs and wants to products at different
they are not designed for STP decisions. We need better stages of lifecycle. For example, for the earlier stages of
and friendlier human-computer interfaces to summarize product lifecycle, AI may play a bigger role, because con-
and display those extremely high dimensional results to sumers’ needs and wants tend to be implicit, whereas for
inform human marketers. the latter stages, human marketers may play a bigger role,
& What happens when the customer is AI? Alexa and Siri because those needs and wants become more explicit.
are not the only AI that can potentially buy things. & How to manage AI-based price negotiation? Negotiating
Increasingly, customers use AI as their agents for infor- price to achieve price personalization can be expected to
mation collection, price negotiation, or purchase. When it be more common and increasingly at the individual level
becomes more real that marketers will be catering to non- (e.g., price bidding). The price discrimination literature
human customers (Rust 1997, 2020) or a combination of (e.g., Montes et al. 2019) tends to view pricing as a pro-
humans and AI as customers, how AI customers would cess of marketers setting the price and consumers reacting
behave, and how marketers should serve them are both to the price. In AI-based price negotiation, the process is
pertinent issues. Are customer emotions still pertinent? dynamic and real-time. We have seen the popularity for
Will rational decision making become the only route to technology giants to use AI for price negotiation, such as
decisions? How will AI customers change the scope of Google’s search keywords bidding and Facebook’s ad
consumer research? bidding for reaching the desirable customers. What will
& How should marketers and AI collaborate, to resonate be the new mechanisms and methods for pricing when AI
with customers? Positioning requires creativity and empa- is used to negotiate prices more widely?
thy about the preferred way that a customer would like to & How to manage customer disengagement due to place
see a product. AI can optimize a positioning recommen- automation? The goal of the place decision is to provide
dation, but may not be mature enough to resonate with the customer a convenience benefit. When the entire place
customers, given that we do not have true feeling AI yet. process is automated, which is increasingly common these
Thus, for the time being, it is important to explore the best days, there is no human-to-human contact, and thus it is
approaches to marketer-AI collaboration for a positioning likely for customers to become less engaged with the
that resonates with customers. This is a question that in- brand. How to avoid customer disengagement while striv-
volves exploring the role of AI intelligences in creativity; ing to provide convenience? Are there ways to mitigate
for example, to what extent should marketers let machines the part that is lost?
be creative on their own (e.g., the Lexus commercial ex- & How to use AI to build strong relational bonds? When
ample), or use them as creative support? What will be two-way communications and interactions are made easier
consumers’ attitudes toward machine-generated creativi- and richer by AI, how are trust and commitment affected
ty? What will decide the acceptance or rejection of ma- by interacting with AI? What will be gained, and what will
chine creativity? be lost? Customers may have changed by interacting with
AI. For example, when AI does more thinking, will cus- withdrawal, and a refrigerator sensor sending inventory infor-
tomers think less when they may feel more? To what ex- mation to a vendor via IoT. Mechanical AI plays an important
tent does the advance of AI, from mechanical, to thinking, role in those routine interactions; however, it may come at the
to feeling, lead to customers feeling more, at the expense cost of customer intimacy (e.g., Treacy and Wiersma 1997).
of thinking, such as accepting fake news on social media To be able to figure out the customer side of strategy requires
at face value? customer data, collected and compiled by mechanical AI,
which is the only methodology that is powerful enough and
scalable enough to capture individual-level data. When cus-
tomers are removed from the interactions, it is more challeng-
Current limitations of AI ing for marketers to remain intimate with customers. Machine-
to-machine interactions and communications will be a new
Our strategic framework illustrates applications of multiple AI form of customer interactions that calls for more future studies
intelligences to various areas of marketing at different strate- to tell marketers how to approach them, and how to keep a
gic decision stages. These applications are not without limita- balance between operational efficiency and customer
tions. We discuss the major limitations of applying the three intimacy.
AI intelligences to marketing for marketers to use AI more
wisely. Limitations of thinking AI
Limitations of mechanical AI Current thinking AI, though powerful, may not be neutral and
transparent, which can result in biased recommendations or
Although current mechanical AI has the strong capability of entail unintended consequences.
collecting and integrating multiple sources of data autono-
mously, very often contexts of the data are lost, creating prob- Opaque AI How thinking AI comes up with a certain recom-
lems in modeling, especially for emotional data. The automat- mendation often is not transparent to human marketers. The
ed process of data collection also makes customer intimacy current dominant machine learning approach to AI designs
less achievable because it is machines talking to machines. machines to learn via a mapping mechanism (i.e., map input
pairs (X,Y) to output Y=F(X)), not via cognitive reasoning
Non-contextual data Many data collected by mechanical (Lewis and Denning 2018). This results in the output being
AI are non-contextual. This is especially the case for emo- unexplainable because it does not answer the “why” question.
tional data because such data are about the individual in Thus, there is a need for scholars to develop explainable AI so
context, meaning modeling the emotional state of a con- that thinking AI can be used for trustworthy and fair marketing
sumer requires contextual and individual-specific data. exchanges (Rai 2020). Opaque AI also results in liability is-
Contextual data are often lost during interaction. One sues. If AI output is not transparent, when AI goes wrong,
Dell AI expert said at a frontline service conference in marketers who use the AI are likely to be held accountable.
2019 that it is not that difficult to model emotions (mean- The accountability issue has emerged since the first fatal ac-
ing using the existing machine learning approach), but the cident of an autonomous car. Thus, marketers, as AI users,
difficulty lies in that emotional data are difficult to cap- when using thinking AI for strategic decisions, need to strive
ture, and thus are not analyzed. For example, in a custom- to use the most explainable AI, rather than just the most pow-
er service interaction, the content and sentiment of the erful AI, to keep the exchanges transparent to both the mar-
conversations are recorded, but not the context of the keter and the customer.
conversations. When an angry and frustrated customer
calls, his way of talking may be different, depending on AI biases Thinking AI is not neutral. If data input is erroneous
whether he is alone or with a group of friends, whether or biased, output is likely to be biased too. However, biased
the weather is gloomy or sunny, or whether the traffic is input is not the only way AI bias can occur. For example, it
jammed or smooth. Even if voice analytics can detect the has been shown that for loan decisions, discriminatory results
sentiment of his voice, it cannot provide guidance to the can occur even if there is no bigotry programmed into the
customer agent as to why the customer is angry, and what system, and the system only seeks to maximize profit
the best way to respond is (Rust and Huang 2020). (Ukanwa and Rust 2020). Researchers have also shown that
gender bias can occur without any conscious (or unconscious)
Machine to machine interactions Or communications are pre- attempt to produce a biased outcome—using only an unbiased
dicted to be the key emerging technology-enabled interactions algorithm (Lambrecht and Tucker 2019).
in digital environments (Yadav and Pavlou 2020). Examples Marketing researchers have addressed some issues, such as
include an ATM getting authorization from the bank for a cash managers and management education underpreparing the next
generation for feeling and emotional intelligence (Huang and revolutionary characteristics of modern thinking AI is its abil-
Rust 2018; Huang et al. 2019), and IoT may impose con- ity to personalize by analyzing big data in an automatic way.
straints and restrictions on consumer journeys (Hoffman and This creates a quantum leap in marketing’s ability to target
Novak 2018; Novak and Hoffman 2019). Thus, in using individual customers. Marketing primarily requires thinking
thinking AI for market analysis, for targeting, and for person- intelligence and feeling intelligence. Until now there has been
alized marketing actions, marketers need to be aware of the only limited ability of technology to help with those things.
potential AI biases and have better knowledge about how AI Now as thinking AI is advancing rapidly, it is gaining the
learns to avoid AI biases. ability to assume many of the thinking tasks in marketing.
Eventually will even assume many of the feeling tasks in
Limitations of feeling AI marketing, as AI develops further. Such efforts are already
underway by researchers.
Although using feeling AI for two-way interactions involving We see that marketers who cannot wait for technology to
humans and for analyzing human feelings and emotions is sufficiently advance use mechanical AI and thinking AI for
common in marketing due to the high-touch nature of many feeling tasks, due to true feeling AI not being ready yet. We
marketing functions (e.g., frontline interactions, customer ser- also see that AI intelligences may not be used in the most
vice, and emotional ad appeals), we don’t yet have true emo- effective way (e.g., collecting customer data indiscriminately
tional machines that can recognize, act, and react to human or accepting AI recommendation blindly). Thus, we develop
emotions appropriately. The substitutive use of mechanical this strategic framework to help marketers leverage the bene-
and thinking AI for feeling AI may generate some unintended fits of multiple AI intelligences for marketing impact. In this
consequences. framework, we lay out the ways in which various AI intelli-
gences can be used in marketing research, marketing strategy
Technology unreadiness The fact that marketers are using (STP), and marketing action (4Ps/4Cs). It shows the strategic
“lower” intelligence AI for feeling functions (i.e., using me- roles that AI can play in marketing, as well as points out the
chanical AI to capture emotional data and using thinking AI to limitations of current AI, to help marketers use AI wisely.
analyze emotional data) may inflate the perceived capability
of AI to assist marketers in understanding customer emotions. Funding This research was supported by grants (106-2410-H-002-056-
MY3 and 107–2410-H-002-115-MY3) from the Ministry of Science and
For example, marketers may overly rely on such feeing AI to
Technology, Taiwan.
interact with customers, resulting in customer disengagement.
Srinivasan et al. (2016) find that higher levels of customer Open Access This article is licensed under a Creative Commons
activity on social media lead to disengagement (i.e., Attribution 4.0 International License, which permits use, sharing, adap-
tation, distribution and reproduction in any medium or format, as long as
Facebook unlikes). Unlikes, as affective responses, may imply you give appropriate credit to the original author(s) and the source, pro-
that customer responses may be polarized more easily if tech- vide a link to the Creative Commons licence, and indicate if changes were
nologies (social media in their study) are not able to interact made. The images or other third party material in this article are included
with customers appropriately. in the article's Creative Commons licence, unless indicated otherwise in a
credit line to the material. If material is not included in the article's
Creative Commons licence and your intended use is not permitted by
Customer unreadiness Another consideration is that cus- statutory regulation or exceeds the permitted use, you will need to obtain
tomers may not be ready for interacting with feeling AI. Luo permission directly from the copyright holder. To view a copy of this
et al. (2019) find that many customers hang up on call-out licence, visit http://creativecommons.org/licenses/by/4.0/.
marketing chatbots once they realize they are talking to bots.
The Technology Readiness Index surveyed what people think
about AI in the workplace and only 10% consider feeling AI
to have the biggest impact in the past 5 years on their jobs,
indicating that customers are not aware that AI can “have” References
feelings and may constitute a threat to their jobs (Espino
Agarwal, R., Dugas, M., Gao, G., & Kannan, P. K. (2020). Emerging tech-
2019). nologies and analytics for a new era of value-centered marketing in
healthcare. Journal of the Academy of Marketing Science, 48(2), 9–23.
Ascarza, E. (2018). Retention futility: Targeting high-risk customers
might be ineffective. Journal of Marketing Research, 55(1), 80–98.
Contributions and conclusions Autor, D. H., & Dorn, D. (2013). The growth of low-skill service jobs and
the polarization of the US labor market. American Economic
Review, 103(5), 1553–1597.
The most disruptive aspect of AI is that it replaces and im- Avery, J. (2018). Tailor brands: Artificial intelligence-driven branding.
proves upon human thinking capability. One of the most Harvard Business School Case 519–017, (August).
Avery, J., & Steenburgh, T. (2018). HubSpot and motion AI: Chatbot- Frey, C. B., & Osborne, M. A. (2017). The future of employment: How
enabled CRM. Harvard Business School case 518-067, February. susceptible are jobs to computerisation? Technological Forecasting
Balducci, B., & Marinova, D. (2018). Unstructured data in marketing. and Social Change, 114(January), 254–280.
Journal of the Academy of Marketing Science, 46(4), 557–590. Gabel, S., Guhl, D., & Klapper, D. (2019). P2V-MAP: Mapping market
Bauer, J., & Jannach, D. (2018). Optimal pricing in e-commerce based on structures for large retail assortments. Journal of Marketing
sparse and noisy data. Decision Support Systems, 106(February), 53–63. Research, 56(4), 557–580.
Berger, J., Humphreys, A., Ludwig, S., Moe, W. W., Netzer, O., & Gali, N., Camprubi, R., & Donaire, J. A. (2017). Analyzing tourism
Schweidel, D. A. (2019). Uniting the tribes: Using text for market- slogans in top tourism destinations. Journal of Destination
ing insight. Journal of Marketing, 84(1), 1–25. Marketing & Management, 6(3), 243–251.
Blattberg, R. C., & Deighton, J. (1996). Manage marketing by the cus- Gopinath, D. (2019). Human + machine: How content analytics delivers
tomer equity test. Harvard Business Review, 74(4), 136–144. unsurpassed value to advertisers. MSI Lunch Lecture, (Sept 25).
Chen, Y., Lee, J. Y., Sridhar, S., Mittal, V., McCallister, K., & Singal, A. Grewal, D., Motyka, S., & Levy, M. (2018). The evolution and future of
G. (2020). Improving cancer outreach effectiveness through retailing and retailing education. Journal of Marketing Education,
targeting and economic assessments: Insights from a randomized 40(1), 85–93.
field experiment. Journal of Marketing, 84(3), 1–27. Grewal, D., Noble, S. M., Roggeveen, A. L., & Nordfalt, J. (2020). The
Chintagunta, P., Hanssens, D. M., & Hauser, J. R. (2016). Editorial— future of in-store technology. Journal of the Academy of Marketing
Marketing science and big data. Marketing Science, 35(3), 341–342. Science, 48(2), 96–113.
Chung, T. S., Wedel, M., & Rust, R. T. (2016). Adaptive personalization Guo, J., Zhang, W., Fan, W., & Li, W. (2018). Combining geographical
using social networks. Journal of the Academy of Marketing and social influences with deep learning for personalized point-of-
Science, 44(1), 66–87. interest recommendation. Journal of Management Information
Chung, T. S., Rust, R. T., & Wedel, M. (2009). My mobile music: An Systems, 35(4), 1121–1153.
adaptive personalization system for digital audio players. Marketing Gupta, S., Lehmann, D. R., & Stuart, J. A. (2004). Valuing customers.
Science, 28(1), 52–68. Journal of Marketing Research, 41(1), 7–18.
Colgate, E., Wannasuphoprasit, W., & Peshkin, M. (1996). Cobots:
Hartmann, J., Huppertz, J., Schamp, C. P., & Heitmann, M. (2019).
Robots for collaboration with human operators. In Proceedings of
Comparing automated text classification methods. International
the ASME Dynamic Systems and Control Division, New York, 58,
Journal of Research in Marketing, 36(1), 20–38.
433-439.
Hewett, K., Rand, W., Rust, R. T., & van Heerde, H. (2016). Brand buzz
Cooke, A. D. J., & Zubcsek, P. P. (2017). The connected consumer:
in the echoverse. Journal of Marketing, 80(3), 1–24.
Connected devices and the evolution of customer intelligence.
Journal of the Association for Consumer Research, 2(2), 164–178. Hoffman, D. L., & Novak, T. P. (2018). Consumer and object experience
Daabes, A. S. A., & Kharbat, F. F. (2017). Customer-based perceptual in the internet of things: An assemblage theory approach. Journal of
map as a marketing intelligence source. International Journal of Consumer Research, 44(6), 1178–1204.
Economics and Business Research, 13(4), 360–379. Huang, M. H., & Rust, R. T. (2017). Technology-driven service strategy.
Davenport, T., Guha, A., Grewal, D., & Bressgott, T. (2020). How arti- Journal of the Academy of Marketing Science, 45(6), 906–924.
ficial intelligence will change the future of marketing. Journal of the Huang, M. H., & Rust, R. T. (2018). Artificial intelligence in service.
Academy of Marketing Science, 48(2), 24–42. Journal of Service Research, 21(2), 155–172.
Davis, E., & Marcus, G. (2015). Commonsense reasoning and common- Huang, M. H., & Rust, R. T. (2020). Engaged to a robot? The role of AI in
sense knowledge in artificial intelligence. Communications of the service. Journal of Service Research, 109467052090226. https://
ACM, 58(9), 93–103. doi.org/10.1177/1094670520902266.
Dekimpe, M. (2020). Retailing and retailing research in the age of big Huang, M. H., Rust, R. T., & Maksimovic, V. (2019). The feeling econ-
data analytics. International Journal of Research in Marketing, 37, omy: Managing in the next generation of artificial intelligence (AI).
3–14. California Management Review, 61(4), 43–65.
Deming, W. E. (1986). Out of the Crisis. Cambridge: Massachusetts Humphreys, A., & Wang, R. (2018). Automated text analysis for con-
Institute of Technology, Center for Advanced Engineering Study. sumer research. Journal of Consumer Research, 44(6), 1274–1306.
Donthu, N., & Rust, R.T. (1989). Estimating geographic customer den- Kelly, S. D. (2019). What computers can’t create. MIT Technology
sities using kernel density estimation. Marketing Science, 8(2), 191– Review, 122(2), 68–75.
203. Kim, S. Y., Schmitt, B. H., & Thalmann, N. M. (2019). Eliza in the
Drew, J. H., Mani, D. R., Betz, A. L., & Datta, P. (2001). Targeting uncanny valley: Anthropomorphizing consumer robots increases
customers with statistical and data-mining techniques. Journal of their perceived warmth but decreases liking. Marketing Letters,
Service Research, 3(3), 205–219. 30(1), 1–12.
Dzyabura, D., & Hauser, J. R. (2011). Active machine learning for con- Kirkpatrick, K. (2020). Tracking shoppers. Communications of the ACM,
sideration heuristics. Marketing Science, 30(5), 757–944. 63(2), 19–21.
Dzyabura, D., & Hauser, J. R. (2019). Recommending products when Kotler, P., & Keller, K. L. (2006). Marketing Management. Pearson
consumers learn their preferences weights. Marketing Science, Prentice Hall: Upper Saddle River.
38(3), 365–541. Kumar, V., Rajan, B., Venkatesan, R., & Lecinski, J. (2019).
Espino, A. (2019). Artificial intelligence: A snapshot into the future. The Understanding the role of artificial intelligence in personalized en-
National Technology Readiness Survey, https://rockresearch.com/ gagement marketing. California Management Review, 61(4), 135–
artificial-intelligence-snapshot-future/. 155.
Feng, J., Li, X., & Zhang, X. (2019). Online product reviews-triggered Lambrecht, A., & Tucker, C. (2019). Algorithmic bias? An empirical
dynamic pricing: Theory and evidence. Information Systems study of apparent gender-based discrimination in the display of
Research, 30(4), 1107–1123. STEM career ads. Management Science, 65(7), 2966–2981.
Fountaine, T., McCarthy, B., & Saleh, T. (2019). Building the AI-
Lauterborn, B. (1990). New marketing litany: Four Ps passé: C-words
powered organization. Harvard Business Review, July–August,
take over. Advertising Age, 61(41), 26.
63–73.
Lee, D., Hosanagar, K., & Nair, H. S. (2018). Advertising content and Power, B. (2017). How Harley-Davidson used artificial intelligence to
consumer engagement on social media: Evidence from Facebook. increase New York sales leads by 2,930%. Harvard Business
Management Science, 64(11), 5105–5131. Review digital article, (may 30), https://hbr.org/2017/05/how-
Lehmann, D. R. (2020). The evolving world of research in marketing and harley-davidson-used-predictive-analytics-to-increase-new-york-
the blending of theory and data. International Journal of Research sales-leads-by-2930.
in Marketing, 37(1), 27–42. Pulles, N. J., & Hartman, P. (2017). Likeability and its effect on outcomes of
Leung, E., Paolacci, G., & Puntoni, S. (2018). Human versus machine: interpersonal interaction. Industrial Marketing Management, 66, 56–63.
Resisting automation in identity-based consumer behavior. Journal Rai, A. (2020). Explainable AI: from black box to glass box. Journal of
of Marketing Research, 55(6), 818–831. the Academy of Marketing Science, 48(1), 137–141.
Lewis, T. G., & Denning, P. J. (2018). Learning machine learning. Rogers, E. M. (1962). Diffusion of innovations (1st ed.). New York: Free
Communications of the ACM, 61(12), 24–27. Press of Glencoe.
Liebman, E., Saar-Tsechansky, M., & Stone, P. (2019). The right music Rust, R. T. (1997). The dawn of computer behavior: Interactive service
at the right time: Adaptive personalized playlists based on sequence marketers will find their customer is not human. Marketing
modeling. MIS Quarterly, 43(3), 765–786. Management, 6(fall), 31-34.
Liu, X. (2020). De-targeting to signal quality. International Journal of Rust, R. T. (2020). The future of marketing. International Journal of
Research in Marketing, 37(2), 386–404. Research in Marketing, 37(1), 15–26.
Liu, X., Singh, P. V., & Srinivasan, K. (2016). A structured analysis of Rust, R. T., & Huang, M. H. (2020). The feeling economy: How artificial
unstructured big data by leveraging cloud computing. Marketing intelligence is creating the era of empathy. Palgrave-Macmillan.
Science, 35(3), 363–388. Rust, R.T., Kamakura, W.A., & Alpert, M.I. (1992). Viewer preference
Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to segmentation and viewing choice models for network television.
medical artificial intelligence. Journal of Consumer Research, 46, Journal of Advertising, 21(1), 1–8.
629–650. Rust, R. T., Rand, W., Huang, M. H., Stephen, A. T., Brooks, G., &
Chabuk, T. (2020). Real-time brand reputation tracking using social
Luo, X., Tong, S., Fang, Z., & Qu, Z. (2019). Frontiers: Machines versus
media. Working paper.
humans: The impact of AI chatbot disclosure on customer pur-
Schneider, M. J., Jagpal, S., Gupta, S., Li, S., & Yu, Y. (2017). Protesting
chases. Marketing Science, 38(6), 937–947.
customer privacy when marketing with second-party data.
Ma, L., & Sun, B. (2020). Machine learning and AI in marketing –
International Journal of Research in Marketing, 34(3), 593–603.
Connecting computing power to human insights. International
Schneider, M. J., Jagpal, S., Gupta, S., Li, S., & Yu, Y. (2018). A flexible
Journal of Research in Marketing. https://doi.org/10.1016/j.
method for protecting marketing data: An application to point-of-
ijresmar.2020.04.005.
sale data. Marketing Science, 37(1), 153–171.
McDuff, D., & Czerwinski, M. (2018). Designing emotionally sentient
Schoenick, C., Clark, P., Tafjord, O., Turney, P., & Etzioni, O. (2017).
agents. Communications of the ACM, 61(12), 74–83.
Moving beyond the Turing test with the Allen AI science.
Mende, M., Scott, M. L., van Doorn, J., Grewal, D., & Shanks, I. (2019). Communications of the ACM, 60(9), 60–64.
Service robots rising: How humanoid robots influence service ex- Simester, D., Timoshenko, A., & Zoumpoulis, S. I. (2020). Targeting
periences and elicit compensatory consumer responses. Journal of prospective customers: Robustness of machine-learning methods
Marketing Research, 56(4), 535–556. to typical data challenges. Management Science, 66(6), 2495–2522.
Misra, K., Schwartz, E. M., & Abernethy, J. (2019). Dynamic online Singh, J., Nambisan, S., Bridge, R. G., & Brock, J. (2020). One-voice
pricing with incomplete information using multiarmed bandit exper- strategy for customer engagement. Journal of Service Research, 1–
iments. Marketing Science, 38(2), 226–252. 24. https://doi.org/10.1177/1094670520910267.
Montes, R., Sand-Zantman, W., & Valletti, T. (2019). The value of per- Soleymanian, M., Weinberg, C. B., & Zhu, T. (2019). Sensor data and
sonal information in online markets with endogenous privacy. behavioral tracking: Does usage-based auto insurance benefit
Management Science, 65(3), 1342–1362. drivers? Marketing Science, 38(1), 21–43.
Netzer, O., Lemaire, A., & Herzenstein, M. (2019). When words sweat: Srinivasan, S., Rutz, O. J., & Pauwels, K. (2016). Paths to and off pur-
Identifying signals for loan default in the text of loan applications. chase: Quantifying the impact of traditional marketing and online
Journal of Marketing Research, 56(6), 960–980. consumer activity. Journal of the Academy of Marketing Science,
Neumann, N., Tucker, C. E., & Whitfield, T. (2019). Frontiers: How 44(4), 440–453.
effective is third-party consumer profiling? Evidence from field Sutton, D. (2018). How AI helped one retailer reach new customers.
studies. Marketing Science, 38(6), 918–926. Harvard Business Review, (may 28), https://hbr.org/2018/05/how-
Ng, I. C. L., & Wakenshaw, S. Y. L. (2017). The internet-of-things: ai-helped-one-retailer-reach-new-customers.
Review and research directions. International Journal of Research Thomaz, F., Salge, C., Karahanna, E., & Hulland, J. (2020). Learning
in Marketing, 34(1), 3–21. from the dark web: Leveraging conversational agents in the era of
Novak, T. P., & Hoffman, D. L. (2019). Relationship journeys in the hyper-privacy to enhance marketing. Journal of the Academy of
internet of things: A new framework for understanding interactions Marketing Science, 48(2), 43–63.
between consumers and smart objects. Journal of the Academy of Timoshenko, A., & Hauser, J. R. (2019). Identifying customer needs from
Marketing Science, 47, 216–237. user-generated content. Marketing Science, 38(1), 1–20.
Ordenes, F. W., Ludwig, S., De Ruyter, K., & Grewal, D. (2017). Tong, S., Luo, X., & Xu, B. (2020). Personalized mobile marketing
Unveiling what is written in the stars: Analyzing explicit, implicit, strategies. Journal of the Academy of Marketing Science, 48(2),
and discourse patterns of sentiment in social media. Journal of 64–78.
Consumer Research, 43(6), 875–894. Treacy, M., & Wiersema, F. (1997). The Discipline of Market Leaders.
Pitt, C. S., Bal, A. S., & Plangger, K. (2020). New approaches to psycho- Reading, MA: Perseus.
graphic consumer segmentation: Exploring fine art collectors using Ukanwa, K., & Rust, R. T. (2020). Discrimination in service. Working
artificial intelligence, automated text analysis and correspondence paper.
analysis. European Journal of Marketing. https://doi.org/10.1108/ Valls, A., Gibert, K., Orellana, A., & Anton-Clave, S. (2018). Using
EJM-01-2019-0083. ontology-based clustering to understand the push and pull factors
for British tourists visiting a Mediterranean coastal destination. Wedel, M., & Kannan, P. K. (2016). Marketing analytics for data-rich
Information & Management, 55, 145–159. environments. Journal of Marketing, 80(6), 97–121.
van Doorn, J., Mende, M., Noble, S., Hulland, J., Ostrom, A., Grewal, D., Wilson, H. J., & Daugherty, P. R. (2018). Collaborative intelligence:
& Petersen, A. (2017). Domo arigato Mr. Roboto: Emergence of Humans and AI are joining forces. Harvard Business Review,
automated social presence in organizational frontlines and cus- July–August, 114–123.
tomers’ service experiences. Journal of Service Research, 20(1), Wirtz, J., Patterson, P. G., Kunz, W. H., Gruber, T., Lu, V. N., Paluch, S.,
43–58. & Martins, A. (2018). Brave new world: Service robots in the front-
Varki, S., Cooil, B., & Rust, R. T. (2000). Modeling fuzzy data in qual- line. Journal of Service Management, 29(5), 907–931.
itative marketing research. Journal of Marketing Research, 37(4), Yadav, M. S., & Pavlou, P. A. (2020). Technology-enabled interactions
480–489. in digital environments: A conceptual foundation for current and
Wang, Q., Li, B., & Singh, P. V. (2018). Copycats vs. original mobile future research. Journal of the Academy of Marketing Science,
apps: A machine learning copycat-detection method and empirical 48(2), 132–136.
analysis. Information Systems Research, 29(2), 273–291.
Wang, Y. X., Ramanan, D., & Hebert, M. (2017). Learning to model the
tail. 31st conference on neural information processing systems Publisher’s note Springer Nature remains neutral with regard to jurisdic-
(NIPS). tional claims in published maps and institutional affiliations.

A Strategic Framework For Artificial Intelligence in Marketing

Uploaded by

Copyright:

Available Formats

A Strategic Framework For Artificial Intelligence in Marketing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Strategic Framework For Artificial Intelligence in Marketing

Uploaded by

Copyright:

Available Formats

Journal of the Academy of Marketing Science (2021) 49:30–50

A strategic framework for artificial intelligence in marketing

areas of marketing (Balducci and Marinova 2018), analytics Multiple AI intelligences

Marketing Action Marketing Research

Fig. 1 AI and strategic marketing decisions

Table 1 A strategic framework for AI in marketing

AI intelligence Mechanical AI Thinking AI Feeling AI

Marketing research Data collection Market analysis Customer understanding

Note. Product includes tangible goods and intangible services

Table 3 Prior and current AI research organized by the strategic framework

AI intelligence Mechanical AI Thinking AI Feeling AI

Marketing Data collection Market analysis Customer understanding

AI intelligence Mechanical AI Thinking AI Feeling AI

• Track brand reputation through

a data-driven approach to STP, it can rely more heavily on AI Segmentation

Place (convenience) logistics, and delivery, to provide convenience benefits to

Current practice Emerging practice Future research

Marketing strategy Marketing action

You might also like