Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

How artificial intelligence will change the future of marketing

How artificial intelligence will change the future of marketing In the future, artificial intelligence (AI) is likely to substantially change both marketing strategies and customer behaviors. Building from not only extant research but also extensive interactions with practice, the authors propose a multidimensional framework for understanding the impact of AI involving intelligence levels, task types, and whether AI is embedded in a robot. Prior research typically addresses a subset of these dimensions; this paper integrates all three into a single framework. Next, the authors propose a research agenda that addresses not only how marketing strategies and customer behaviors will change in the future, but also highlights important policy questions relating to privacy, bias and ethics. Finally, the authors suggest AI will be more effective if it augments (rather than replaces) human managers. . . . . . Keywords Artificial intelligence Marketing strategy Robots Privacy Bias Ethics AI is going to make our lives better in the future. ride-sharing businesses must evolve to avoid being marginal- —Mark Zuckerberg, CEO, Facebook ized by AI-enabled transportation models; demand for auto- mobile insurance (from individual customers) and breathaly- zers (fewer people will drive, especially after drinking) will likely diminish, whereas demand for security systems that Introduction protect cars from being hacked will increase (Hayes 2015). Driverless vehicles could also impact the attractiveness of real In the future, artificial intelligence (AI) appears likely to in- estate, because (1) driverless cars can move at faster speeds, fluence marketing strategies, including business models, sales and so commute times will reduce, and (2) commute times processes, and customer service options, as well as customer will be more productive for passengers, who can safely work behaviors. These impending transformations might be best while being driven to their destination. As such, far flung understood using three illustrative cases from diverse indus- suburbs may become more attractive, vis-à-vis the case today. tries (see Table 1). First, in the transportation industry, driver- Second, AI will affect sales processes in various industries. less, AI-enabled cars may be just around the corner, promising Most salespeople still rely on a telephone call (or equivalent) to alter both business models and customer behavior. Taxi and as a critical part of the sales process. In the future, salespeople Thomas Davenport, Abhijit Guha, Dhruv Grewal and Timna Bressgott contributed to the writing of the paper. Mark Houston served as accepting Editor for this article. * Timna Bressgott Department of Technology, Operations, and Information t.bressgott@maastrichtuniversity.nl Management, Babson College, Babson Park, MA 02457, USA Department of Marketing, Darla Moore School of Business, Thomas Davenport University of South Carolina, Columbia, SC 29208, USA tdavenport@babson.edu Department of Marketing, Babson College, Babson Abhijit Guha Park, MA 02457, USA abhijit.guha@moore.sc.edu Department of Marketing and Supply Chain Management, Dhruv Grewal Maastricht University, Tongersestraat 53, 6211, LM dgrewal@babson.edu Maastricht, The Netherlands J. of the Acad. Mark. Sci. (2020) 48:24–42 25 will be assisted by an AI agent that monitors tele- insights about not only the ultimate promise of AI, but also conversations in real time. For example, using advanced voice the pathway and timelines along which AI is likely to develop. analysis capabilities, an AI agent might be able to infer from a This paper addresses the issues above, building not only from customer’s tone that an unmentioned issue remains a problem a review of literature across marketing (and more generally, and provide real-time feedback to guide the (human) business), psychology, sociology, computer science, and ro- salesperson’s next approach. In this sense, AI could augment botics, but also from extensive interactions with practitioners. salespersons’ capabilities, but it also might trigger unintended Second, the preceding examples highlight mostly positive negative consequences, especially if customers feel uncom- consequences of AI, without detailing the widespread, justifi- fortable about AI monitoring conversations. Also, in the fu- able concerns associated with their use. Technologists such as ture, firms may primarily use AI bots, which—in some Elon Musk believe that AI is “dangerous” (Metz 2018). AI cases—function as well as human salespeople, to make initial might not deliver on all its promises, due to the challenges it contact with sales prospects. But the danger remains that if introduces related to data privacy, algorithmic biases, and customers discover that they are interacting with a bot, they ethics (Larson 2019). may become uncomfortable, triggering negative We argue that the marketing discipline should take a lead consequences. role in addressing these questions, because arguably it has the Third, the business model currently used by online retailers most to gain from AI. In an analysis of more than 400 AI use generally requires customers to place orders, after which the cases, across 19 industries and 9 business functions, online retailer ships the products (the shopping-then-shipping McKinsey & Co. indicates that the greatest potential value model—Agrawal et al. 2018;Gansetal. 2017). With AI, of AI pertains to domains related to marketing and sales online retailers may be able to predict what customers will (Chui et al. 2018), through impacts on marketing activities want; assuming that these predictions achieve high accuracy, such as next-best offers to customers (Davenport et al. retailers might transition to a shipping-then-shopping business 2011), programmatic buying of digital ads (Parekh 2018), model. That is, retailers will use AI to identify customers’ and predictive lead scoring (Harding 2017). The impact of preferences and ship items to customers without a formal or- AI varies by industry; the impact of AI on marketing is highest der, with customers having the option to return what they do in industries such as consumer packaged goods, retail, bank- notneed(Agrawaletal. 2018;Gans et al. 2017). This shift ing, and travel. These industries inherently involve frequent would transform retailers’ marketing strategies, business contact with large numbers of customers, and produce vast models, and customer behaviors (e.g., information search). amounts of customer transaction data and customer attribute Businesses like Birchbox, Stitch Fix and Trendy Butler al- data. Further, information from external sources, such as so- ready use AI to try to predict what their customers want, with cial media or reports by data brokers, can augment these data. varying levels of success. Thereafter, AI can be leveraged to analyze such data and de- The three use cases (above) illustrate why so many aca- liver personalized recommendations (relating to next product demics and practitioners anticipate that AI will change the to buy, optimal price etc.) in real time (Mehta et al. 2018). face of marketing strategies and customers’ behaviors. In fact, Yet marketing literature related to AI is relatively sparse, a survey by Salesforce shows that AI will be the technology prompting this effort to propose a framework that describes most adopted by marketers in the coming years (Columbus both where AI stands today and how it is likely to evolve. 2019). The necessary factors to allow AI to deliver on its Marketers plan to use AI in areas like segmentation and ana- promises may be in place already; it has been stated that “this lytics (related to marketing strategy) and messaging, person- very moment is the great inflection point of history” (Reese alization and predictive behaviors (linked to customer behav- 2018, p. 38). Yet this argument can be challenged. First, the iors) (Columbus 2019). Thus, we also propose an agenda for technological capability required to execute the preceding ex- future research, in which we delineate how AI may affect amples remains inadequate. By way of an exemplar, self- marketing strategies and customer behaviors. In so doing, driving cars are not ready for deployment (Lowy 2016), we respond to mounting calls that AI be studied not only by as—amongst other things—currently self-driving cars cannot those in computer science, but also studied by those who can handle bad weather conditions. Predictive analytics also need integrate and incorporate insights from psychology, econom- to improve substantially before retailers can adopt shipping- ics and other social sciences (Rahwan et al. 2019;also see then-shopping practices that avoid substantial product returns Burrows 2019). and the associated negative affect. Putting all this together, it appears that marketing managers and researchers need Introduction to artificial intelligence Miller (2016) outlines the difference between an AI bot and a chatbot. In Researchers propose that AI “refers to programs, algorithms, brief, chatbots rely on (relatively) simple algorithms, whereas AI bots have greater capabilities, incorporating complex algorithms and NLP. systems and machines that demonstrate intelligence” (Shankar 26 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 1 Select use cases (in the order in which they appear in the paper) Industry or Usage Context (specific firm or AI application) Description AI in driverless cars (e.g., Tesla) In the future, AI-enabled cars may allow for car journeys without any driver input, with the potential to significantly impact various industries (e.g., insurance, taxi services) and customer behaviors (e.g., whether they still buy cars). Online retailing AI (e.g., Birchbox) AI will enable better predictions for what customers want, which may cause firms to move away from a shopping-then-shipping business model and toward a shipping-then-shopping business model. Fashion-related AI (e.g., Stitch Fix) AI applications support stylists, who curate a set of clothing items for customers. Stitch Fix’s AI analyzes both numeric and image/other non-numeric data. Sales AI (e.g. Conversica) AI bots can automate parts of the sales process, augmenting the capabilities of existing sales teams. There may be backlash if customers know (upfront) that they are chatting with an AI bot (even if the AI bot is otherwise capable) Customer service robots (e.g., Rock’em Robots with task-automating AI respond to relatively simple customer and Sock’em; Pepper) service requests (e.g., making cocktails). Emotional support AI (e.g., Replika) AI aims to provide emotional support to customers by asking meaningful questions, offering social support, and adjusting to users’ linguistic syntax. In-car AI (e.g., Affectiva) In-car AI that analyzes driver data (e.g., facial expression) to evaluate drivers’ emotional and cognitive states. Customer screening AI (e.g. Kanetix) AI used to identify customers who should be provided incentives to buy insurance (and avoid those who (1) are already likely to buy and (2) those unlikely to buy). Business process AI (e.g., IBM Interact) AI used for multiple (simple) applications, such as customized offers (e.g., Bank of Montreal). Retail store AI (e.g., Café X, Lowebot, Robots that can serve as coffee baristas, respond to simple customer service 84.51, Bossa Nova) requests in Lowe’s stores, and identifying misshelved items in grocery stores. Security AI (e.g., Knightscope’s K5) Security robots patrol in offices or malls, equipped with superior sensing capabilities (e.g., thermal cameras). Spiritual support AI (e.g., BlessU-2; Xian’er) Customizable robot priest/monk offering blessings in different languages to the user. Companion robot AI (e.g., Harmony from Realbotix) Customizable robot companion, which promises reduced loneliness to the user. 2018,p. vi), is “manifested by machines that exhibit aspects of or “reading” documents to extract key provisions using natu- human intelligence” (Huang and Rust 2018,p. 155), andin- ral language processing. Second, AI can gain insights from volves machines mimicking “intelligent human behavior” vast volumes of customer and transaction data, involving not (Syam and Sharma 2018, p. 136). It relies on several key just numeric but also text, voice, image, and facial expression technologies, such as machine learning, natural language pro- data. Using AI-enabled analytics, firms then can predict what cessing, rule-based expert systems, neural networks, deep a customer is likely to buy, anticipate credit fraud before it learning, physical robots, and robotic process automation happens, or deploy targeted digital advertising in real time. (Davenport 2018). By employing these tools, AI provides a For example, stylists working at Stitch Fix, a clothing and means to “interpret external data correctly, learn from such styling service, use AI to identify which clothing styles will data, and exhibit flexible adaptation” (Kaplan and Haenlein best suit different customers. The underlying AI integrates 2019, p. 17). Another way to describe AI depends not on its data provided by customers’ expressed preferences, their underlying technology but rather its marketing and business Pinterest boards, handwritten notes, similar customers’ pref- applications, such as automating business processes, gaining erences, and general style trends. Finally, AI can engage cus- insights from data, or engaging customers and employees tomers, before and after the sale. The Conversica AI bot works (Davenport and Ronanki 2018). We build on this latter per- to move customer transactions along the marketing pipeline, spective. A listing of this research is provided in Table 2. and the AI bot used by 1–800-Flowers provides both sales and First, to automate business processes, AI algorithms per- customer service support. AI bots offer advantages beyond form well-defined tasks with little or no human intervention, just 24/7 availability. Not only do these AI bots have lower such as transferring data from email or call centers into error rates, but also they free up human agents to deal with recordkeeping systems (updating customer files), replacing more complex cases. Further, AI bot deployment can be lost ATM cards, implementing simple market transactions, scaled up or down as needed, when demand ebbs or flows. J. of the Acad. Mark. Sci. (2020) 48:24–42 27 As these descriptions suggest, AI offers the potential to AI capabilities forward, from task automation to context increase revenues and reduce costs. Revenues may increase awareness (e.g., Ghahramani 2015;Mnihetal. 2015). through improved marketing decisions (e.g., pricing, promo- Context awareness is a form of intelligence that requires ma- tions, product recommendations, enhanced customer engage- chines and algorithms to “learn how to learn” and extend ment); costs may decline due to the automation of simple beyond their initial programming by humans. Such AI appli- marketing tasks, customer service, and (structured) market cations can address complex, idiosyncratic tasks by applying transactions. Furthermore, the above discussions indicate that holistic thinking and context-specific responses (Huang and rather than replacing humans, firms generally are using AI to Rust 2018). However, such capabilities remain distant; a 2016 augment their human employees’ capabilities, such as when survey of AI researchers indicated there was only a 50% Stitch Fix uses AI to augment its stylists’ efforts to make chance of achieving context awareness (or its equivalent) by appropriate choices for clients (Gaudin 2016). This point 2050 (Müller and Bostrom 2016). Building on the above aligns well with sentiments expressed by Ginni Rometty, the point, Reese (2018, p. 61) cautions that such AI “does not CEO of IBM, who proposed that AI would not lead to a world currently exist… nor is there agreement … if it is possible.” of man “versus” machine but rather a world of man “plus” Nevertheless, this capability constitutes the goal of AI devel- machines (Carpenter 2015). opments, as predicted by compelling examples from science fiction, such as Jarvis from the Iron Man movies or Karen from Spider Man–Homecoming; both AI can understand A framework for understanding artificial new and complex contexts and create solutions therein. intelligence The differences between task automation and context awareness map onto concepts of narrow versus general AI Building on insights from marketing (and more generally (Baum et al. 2011; Kaplan and Haenlein 2019; Reese 2018). business), social sciences (e.g., psychology, sociology), and As Kaplan and Haenlein (2019) state, both narrow and general computer science/robotics, we propose a framework to help AI may equal or outperform human performance, but narrow customers and firms anticipate how AI is likely to evolve. We AI is focused on a specific domain and cannot learn to extend consider three AI-related dimensions: levels of intelligence, into new domains, whereas general AI can extend into new task type, and whether the AI is embedded in a robot. domains. It is important to clarify that although in this paper we Level of intelligence consider two levels of intelligence (task automation vs. con- text awareness), ideally levels of intelligence are best concep- Task automation versus context awareness Davenport and tualized as a continuum. Some AI applications have moved Kirby (2016) contrast task automation with context aware- beyond task automation but still fall well short of context ness. The former involves AI applications that are standard- awareness, such as Google’s DeepMind AlphaGo (which beat ized, or rule based, such that they require consistency and the the world’s best Go player), the AI poker player Libratus, and imposition of logic (Huang and Rust 2018). For example, Replika. These applications represent substantial advances, IBM’s Deep Blue applied standardized rules and “brute force” yet state-of-the-art AI still is closer to task automation algorithms to beat the best human chess player. Such AI is best (Davenport 2018). suited to contexts with clear rules and predictable outcomes, like chess. On the cruise ship Symphony of the Seas,two Overview of extant research Research into the psychology of robots, Rock ‘em and Sock ‘em, make cocktails for customers. automation (Longoni et al. 2019), examines how customers Elsewhere, the robot Pepper can provide frontline greetings, may respond to AI. Notwithstanding the fact that AI and IBM’s Watson can provide credit scoring and tax prepa- may be more accurate and/ or more reliable than ration assistance. Notwithstanding that these AI applications humans, customers have reservations about AI, and involve fairly structured contexts, many firms struggle to im- these reservations tend to increase as AI moves towards plement even these AI applications and rely on specialized context awareness. In turn, these increased reservations businesses like Infinia ML and Noodle, or consulting firms negatively impact the propensity to adopt AI, propensity like Accenture or Deloitte, to develop and set up initial AI to use AI, etc. A listing of such research is shown in Table 2. initiatives. Moving forward, we discuss (separately) issues relating to AI In contrast, context awareness continues to be developed, adoption and AI usage. and researchers in computer science are working on moving Reese (2018, p. 61) cautions that this type of AI is in no way “easy AI.” Replika (replica.ai) aims to serve as an AI friend, programmed to ask ques- To clarify, businesses like Infinia ML etc. also provide support moving tions about you and your life that are “meaningful,” and to offer emotional forward, when the firm initiates more advanced AI initiatives. support (French 2018;alsosee Hassler 2018). 28 J. of the Acad. Mark. Sci. (2020) 48:24–42 AI adoption Customers appear to hold AI to a higher standard know that messages are more effective when the perceived than is normatively appropriate (Gray 2017), as exemplified characteristics of the message source and the contents of the by the case of driverless cars. Customers should adopt AI if its actual message match, communication from AI should be use leads to significantly fewer accidents; instead, customers more effective when it highlights how rather than why in its impose higher standards and seek zero accidents from AI. messaging (regulatory construal fit; Lee et al. 2009;Motyka Understanding the roots of this excessive caution is important. et al. 2014). In line with the above, Kim and Duhachek (2018) A preliminary hypothesis suggests that customers trust AI showed that a message from an AI application is more persua- less, and so hold AI to a higher standard, because they believe sive when the message is about how to use a product, rather that AI cannot “feel” (Gray 2017). than why to use this product. This is because customers doubt Task characteristics also influence AI adoption. To the ex- whether AI can “understand” the importance of engaging in tent a task appears subjective, involving intuition or affect, certain consumption behaviors. customers likely are even less comfortable with AI (Castelo Next we pivot to factors that impact the propensity of cus- 2019). Research confirms that customers are less willing to tomers to engage with AI. Examining the case of medical use AI for tasks involving subjectivity, intuition, and affect, decision making, Longoni et al. (2019)show that customers’ because they perceive AI as lacking the affective capability or reservations are due to their concerns about uniqueness ne- empathy needed to perform such tasks (Castelo et al. 2018). glect (i.e., the AI is perceived as less able to identify and relate Tasks differ in their consequences; choosing a movie is with customers’ unique features). Further, building from prior relatively less consequential, but steering a car may involve work (Şimşek and Yalınçetin 2010; also see Haslam et al. more consequences. Using AI for consequential tasks is per- 2005), Longoni et al. (2019) show that these reservations are ceived as involving more risk, in turn reducing adoption in- more for customers who have higher scores on the ‘personal tentions. Early work has found support for this hypothesis, sense of uniqueness’ scale. In other work on how customers more so among more conservative consumers for whom risks engage with AI, Luo et al. (2019) examined how (potential) are more salient (Castelo et al. 2018; Castelo and Ward 2016). customers engage with AI bots. In reality, AI bots can be as Finally, customer characteristics may also impact AI adop- effective as trained salespersons, and 4x as effective as inex- tion. We build from two points: (1) when outcomes are con- perienced salespersons. However, if it is disclosed that the sequential, this increases perceptions of risk (Bettman 1973), customer is conversing with an AI bot, purchase rates drop and (2) women perceive more risk in general (Gustafsod by 75%. Linked to points made prior in this paper, because 1998) and take on less risk (Byrnes et al. 1999). Hence, early customers perceive the AI bot as less empathetic, they are curt work has found that women (vs. men) are less likely to adopt when interacting with AI bots, and so purchase less. AI, especially when outcomes are consequential (Castelo and Ward 2016). Moving beyond demographics, other factors also Task type impact the extent of AI adoption, e.g., to the extent a task is salient to a customer’s identity, the customer may be less like- Task type refers to whether the AI application analyzes num- ly to adopt AI (Castelo 2019). To elaborate, if a certain con- bers versus non-numeric data (e.g., text, voice, images, or sumption activity is central to a customer’s identity, then the facial expressions). These different data types all provide in- customer likes to take credit for consumption outcomes puts for decision making, but analyzing numbers is substan- (Leung et al. 2018). Some customers perceive that using AI tially easier than analyzing other data forms. Practitioners, for these consumption activities is tantamount to cheating, and such as senior managers from Infinia ML, formulate this cat- this hinders the attribution of credit post-consumption. egorization slightly differently, noting that data that can be Therefore, if an activity is central to a customer’s identity, then organized into tabular formats are significantly easier to ana- the customer may be less likely to adopt AI (for this activity). lyze than those data that cannot. In our discussions with em- ployees of Stitch Fix, we gained further clarity on this point. AI usage Moving past adoption issues, we note some usage Stitch Fix elicits data from customers using both direct ques- considerations, including how AI should communicate with tions about their preferences (which can be put in tabular customers. Customers do not associate AI applications with formats) and indirect elicitations from customers’ Pinterest autonomous goals (Kim and Duhachek 2018); for example, pages and likes. Stitch Fix uses proprietary AI algorithms to customers do not believe Google’s AlphaGo has the self- analyze the latter, non-numeric data and regards these data as driven goal to be a national Go champion. Rather, they believe very useful, because it has learned that customers cannot al- that this AI application is programmed to play the game Go. ways articulate their preferences on numeric scales. Consistent with this perception, customers are more likely to The distinction in the above paragraph is critical, because focus on “how” (rather than “why”) the AI application per- much data is non-tabular in form, and so being able to com- forms; implying that when engaging with AI, customers will prehend and analyze such data significantly enhances the im- be in a low level construal mindset. From extant research, we pact of AI. Many AI applications have started to analyze text, J. of the Acad. Mark. Sci. (2020) 48:24–42 29 voice, image, and face data inputs. These data inputs are ini- individuals without any protective barrier, travel with tially in non-numeric formats, but are often translated into individuals, etc.). numerical formats, e.g., pixel brightness values, relating to images. Applications that can process such data inputs in- Overview of extant research Prior research (Table 2) indicates clude, for example (1) IPSoft, which processes words spoken that using robots offer substantial advantages, especially in to customer agents to interpret what customers want (2) cases involving customer interactions. As prior work indi- Affectiva, which is working on in-car AI that can sense driver cates, customers form more personal bonds with robots than emotion and fatigue and switch control to an autonomous AI, with AI that lack any physical embodiment. For example, and (3) Cloverleaf’s shelfPoint, installed on retail store individuals enjoy interacting with a physically present robot shelves, which examines customers’ facial expressions to an- than with either a robot simulation (on a computer) or a robot alyze their emotional responses at the point of purchase. presented via teleconference (Wainer et al. 2006). Further, Although currently AI’s abilities to comprehend and analyze customers empathize with robots. When individuals are asked such non-numeric data formats remain somewhat limited, de- to administer pain—via electric shocks—to a (physically pres- veloping this ability will be critical for the full realization of ent) robot or a robot simulation, both of which go on to display the power of AI, and computer scientists are working towards marks indicating pain after being subjected to an shocked, improving AI capabilities in this regard (e.g., LeCun et al. individuals empathized more with the physically present robot 2015; You et al. 2016). (Kwak et al. 2013). Finally, customers interacted longer with a Separate to the above, it is worth pointing out that the robot diet coach than with either a virtually present diet coach ability to analyze unstructured data may be limited by legacy or a diet diary in a paper form (Kidd and Breazeal 2008). infrastructures. A senior manager in Infinia ML indicated that Other studies find that customers demonstrate reciprocity- often data is stored in formats and structures less amenable to based perceptions, e.g., they express more positive percep- AI deployment. Also, Kroger has an AI application that auto- tions of a care robot that asks for help and then returns this mates visual inspection of out-of-stock items on its grocery help by offering a favor (Lammer et al. 2014). In a prisoner’s shelves. In an interview with one of the authors of this paper, a dilemma experiment, participants exhibited similar reciprocity Kroger data scientist reported that the proper functioning of levels toward both robot partners and human partners Kroger’s AI application requires hardware upgrades; specifi- (Sandoval et al. 2016), and their reciprocity towards the robot cally, it needs to upgrade its cameras to higher resolution partner increased even more if the robot provided early signs levels if the AI application is to work properly. of cooperation (vs. random behavior). Noting the benefits of embedding AI in robots, work in robotics is examining how AI in robots best to improve not only the physical capability of robots but also the robot–AI interface (e.g., Adami 2015; Kober et al. Virtuality-reality continuum Most AI is virtual in form. For 2013; Steels and Brooks 2018). Further, to take advantage of example, Replika is available on smartphones, and Libratus the preference for physical embodiment, some vendors of vir- uses a digital platform. However, AI can also be embedded in tual agents (or bots) try to present these agents as having a a real entity or robot form, with some elements of physical physical form. IPsoft’s virtual agent, for example, is called embodiment. The extent to which a form is virtual versus Amelia and is often represented by a lifelike avatar image embodied reflects its position on the Milgram virtuality– and voice. reality continuum (Milgram et al. 1995). In this sense, re- However, other research shows that customers’ discomfort searchers and practitioners should conceive of virtual and real with AI is accentuated when the AI application is embedded in forms not as distinct categories but rather as endpoints on a a robot. As robots appear more humanlike, they become more continuum, within which AI entities are spread out. An AI like unnerving, in line with the uncanny valley hypothesis (UVH; Conversica is purely virtual, with no physical embodiment— Mori 1970). UVH arises because the appearance of robots although some companies that use virtual AI do give it names. “prompts attributions of mind. In particular, we suggest that In contrast, an AI application embedded in a robot barista machines become unnerving when people ascribe to them (e.g., Tipsy Robot in Las Vegas) appears somewhere on the experience (the capacity to feel and sense), rather than agency continuum between virtuality and reality, because it has some (the capacity to act and do)” (Gray and Wegner 2012,p. 125). physical embodiment; however, that embodiment can only Such factors may hinder AI adoption. operate in a narrow range and on a specific task (making a drink). Finally, the AI embedded in proposed multifunctional, Masahiro Mori wrote an influential paper arguing that making robots look companion robots (that today remain under development) more human is beneficial, but only up to a certain point, after which such robots elicit negative reactions. Thus, reactions become negative as robots would entail substantially more reality, featuring both physical move. From somewhat human to human-like. Thereafter, if robots look per- embodiment and the capacity to operate in wide range fectly human, reactions turn positive. The valley reflects these trends, as reac- of contexts (specifically, share physical proximity with tions initially becoming more negative, then turn positive. 30 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 2 Select extant research (in the order in which they appear in the paper) Paper Domain Dimension Takeaways Agrawal et al. (2018) BUS Artificial intelligence (AI) reduces the cost of prediction. Gans et al. (2017)BUS Rahwan et al. (2019) CS/R To best understand AI, bring in insights from not only computer science, but also other disciplines Shankar (2018)MKTG AI “refers to programs, algorithms, systems and machines Huang and Rust (2018)MKTG that demonstrate intelligence” (Shankar 2018,p. vi),is “manifested by machines that exhibit aspects of human intelligence” (Huang and Rust 2018, p. 155), involves machines mimicking “intelligent human behavior” (Syam and Sharma 2018, p. 136), and provides means to “interpret external data correctly, learn from such data, and exhibit flexible adaptation” (Kaplan and Haenlein 2019,p. 17). Syam and Sharma (2018)MKTG HuangandRust(2018) - Mechanical and analytical intelligences involve simple, rule-based tasks. Intuitive and empathetic intelligences involve complex tasks requiring empathy, holistic thinking and context-specific responses. Kaplan and Haenlein (2019) MKTG Kaplan and Haenlein (2019) – Used the terms narrow versus general AI. Narrow AI somewhat maps onto mechanical and analytical intelligences, whereas general AI maps onto intuitive and empathetic intelligences. Davenport and Ronanki (2018) BUS LVLINT Another way to describe AI is by stating its marketing and business outcomes, such as automating business processes, gaining insights from data, or engaging customers and employees Davenport and Kirby (2016) BUS LVLINT Contrasts task automation with context awareness. The former involves AI applications that are standardized, or rule based (akin to narrow AI). The latter is a form of intelligence that requires machines and algorithms to ‘learn how to learn’ and extend beyond their initial programming (akin to general AI). Ghahramani (2015) CS/R LVLINT How machines can learn from experience, using probabilistic machine learning. Mnih et al. (2015) CS/R LVLINT How artificial agents can learn to generalize from past experience to new situations, using reinforcement learning. Müller and Bostrom (2016) BUS LVLINT Artificial general intelligence (AGI) is a hypothetical technology that would be the equivalent of a human intelligence in terms of its flexibility and capability of performing and learning a vast range of tasks (similar to context awareness). In a survey of AI researchers, the median estimate was for a 50% chance of achieving an AGI by 2050 and a 90% chance of achieving one by 2075. Reese (2018) BUS LVLINT Defines narrow versus general AI and analytical AI versus humanized AI; Baum et al. (2011) SOC LVLINT both contrasts are very similar to the contrast between task automation versus context awareness. Reese (2018) cautions that AGI does not exist, and that there is no guarantee that it ever will. Davenport (2018) BUS LVLINT The state-of-the-art AI is closer to task automation than context awareness. Gray (2017) PSY LVLINT Customers appear to hold AI to a higher standard than is normatively appropriate. A preliminary hypothesis suggests that customers trust AI less, and so hold AI to a higher standard, because they believe that AI cannot “feel”. Castelo (2019) MKTG LVLINT To the extent a task appears subjective, involving intuition or affect, customers Builds from: Castelo et al. (2018)MKTG likely are less comfortable with AI (Castelo 2019). Customers are less willing to use AI for tasks involving subjectivity, intuition, and affect, because they perceive AI as lacking the affective capability or empathy needed to perform such tasks (Castelo et al. 2018). Castelo and Ward (2016) MKTG LVLINT Using AI for consequential tasks is perceived as involving more risk, in Builds from: Bettman (1973)MKTG turn reducing adoption intentions. This is more so amongst (1) conservative consumers, for whom risks are more salient, (2) women, who perceive Gustafsod (1998)PSY more risk in general, and take on less risk. Byrnes et al. (1999)PSY J. of the Acad. Mark. Sci. (2020) 48:24–42 31 Table 2 (continued) Paper Domain Dimension Takeaways Leung et al. (2018) MKTG LVLINT If a certain consumption activity is central to a customer’s identity, the customer would like to take credit for consumption outcomes. Some customers perceive that using AI for these consumption activities is tantamount to cheating, and this hinders the attribution of credit post-consumption. Hence if an activity is central to a customer’s identity, then the customer may be less likely to adopt AI for this activity. Kim and Duhachek (2018) MKTG LVLINT Customers do not associate AI applications with autonomous goals Builds from: Lee et al. (2009)MKTG (Kim and Duhachek 2018). In line with this perception, customers are more likely to focus on “how” (rather than “why”) the AI application Motyka et al. (2014)MKTG performs; implying that when engaging with AI, customers will be in a low level construal mindset. Because persuasion is more effective when the perceived characteristics of the persuasion source and the persuasion message match, communication from AI should be more effective when it highlights how rather than why in its messaging (regulatory construal fit; Lee et al. 2009; Motyka et al. 2014). AI persuasion messages are more effective in persuading consumers to buy the recommended product or services when the message highlights “how” to use the product rather than “why” to use the product. These effects are because customers doubt whether AI can understand “why” it is important for customers to engage in certain behaviors. Longoni et al. (2019) MKTG LVLINT Examining the case of medical decision making, Longoni et al. (2019) propose that customers’ reservations are due to their concerns about Builds from: Şimşek PSY uniqueness neglect (i.e., the AI is perceived as less able to identify and and Yalınçetin (2010) relate with customers’ unique features). Further, building from prior Haslam et al. (2005)PSY work (Şimşek and Yalınçetin 2010; also see Haslam et al. 2005), Longoni et al. (2019) propose that these reservations would be more for customers who havehigherscores onthe ‘personal sense of uniqueness’ scale (Şimşek and Yalınçetin 2010). Luo et al. (2019) MKTG LVLINT Examines how (potential) customers engage with AI bots. In reality, AI bots can be as effective as trained salespersons, and 4 times effective as inexperienced salespersons. However, if it is disclosed that the customer is conversing with an AI bot, purchase rates reduce by 75%. Because customers perceive the AI bot as less empathetic, they are curt when interacting with AI bots, and so purchase less. Ties into themes from Castelo et al. (2018). LeCun et al. (2015) CS/R TSKTYPE How deep learning has improved the state-of-the-art in speech and visual object recognition. You et al. (2016) CS/R TSKTYPE How using a new algorithm improves visual object recognition. Milgram et al. (1995) PSY ROBOT Proposes the virtuality-reality continuum. Wainer et al. (2006) CS/R ROBOT Interacting with a physical robot is perceived as more enjoyable than either interacting with a simulated robot on a computer or interacting with a real robot presented through teleconferencing. Kwak et al. (2013) CS/R ROBOT When asked to administer electric shocks to a (physical) robot or a simulated robot on a computer screen, individuals empathized more with the (physical) robot. Kidd and Breazeal (2008) CS/R ROBOT Interactions were longer with a robot diet coach than either a virtual diet coach or a pen-and-paper diet diary. Lammer et al. (2014) CS/R ROBOT Individuals express reciprocity towards robots. Adami (2015) CS/R ROBOT With suitable machine learning algorithms, robots can learn from past experiences. Kober et al. (2013) CS/R ROBOT Reinforcement learning can work for robots embedded with suitable machine learning algorithms. Mori (1970) PSY ROBOT Making robots look more human is beneficial, but only up to a certain point, after which such robots elicit negative reactions (UVH). Gray and Wegner (2012) PSY ROBOT Machines are perceived as more unnerving when individuals ascribe to machines the capacity to feel, rather than capacity to do. 32 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 2 (continued) Paper Domain Dimension Takeaways Mende et al. (2019) MKTG ROBOT Interactions with robots trigger discomfort (linked to UVH) and so further trigger compensatory behaviors. Boyd and Holton (2018) SOC ROBOT Will the combination of robotics and AI lead to an unprecedented social transformation? Pedersen et al. (2018) SOC ROBOTS Outlines the issues surrounding use of social robots in medical treatment, care facilities, and private homes. Also, outlines ethical concerns. André et al. (2018) MKTG LVLINT Because AI facilitates data-driven, micro-targeting marketing offerings, customers should view such offerings favorably, because it reduces search costs. Yet this could undermine customers’ perceived autonomy, with implications for their subsequent evaluations and choices. Aguirre et al. (2015) MKTG LVLINT Proposes the privacy–personalization paradox, whereby individuals balance privacy concerns against the benefits of personalized recommendations. Wang and Kosinski (2018) PSY TSKTYPE How to use deep neural networks to identify sexual orientation, merely by analyzing facial images MKTG Marketing; BUS Business; PSY Psychology; SOC Sociology; CS/ R Computer Science/ Robotics. Dimension: LVLINT levels of intelligence; TSKTYPE task type; ROBOTwhetherAIin robots Moving beyond AI adoption, we pivot to how customers that in the lower half of Fig. 1, we do not distinguish between interact with robots with embedded AI. Early research sug- numeric and non-numeric data, because context awareness– gests that interactions with AI-embedded robots trigger dis- capable AI likely will be able to handle any types of data. comfort (linked to the UVH) and so further trigger (negative) The first four use cases, associated with short to medium compensatory behaviors, like buying of status goods, or eating term developments, involve task automation (see Fig. 1). more food (Mende et al. 2019). From a theory perspective, this work not only shows the downsides of anthropomorphism Cell 1: Controller of numerical data The first cell in Fig. 1 (especially in the case of robots), but also the existence of reflects what AI can do very well, namely, statistical analyses compensatory consumption specifically linked to robots. of numeric data using machine learning. A typical use case is More broadly, sociologists ponder how AI (and specifically the application of AI to optimize prices (Antonio 2018). robots with embedded AI) might transform economy and so- Pricing strategies must balance two competing concerns; that ciety (Boyd and Holton 2018). For example, cloud-based the price is low enough to attract customers versus high technology facilitates deep learning in robots, which can learn enough to enable the firm to earn sufficient profits. Firms from human agents through repeated interactions. use AI to analyze vast amounts of numeric data (including Sociologists particularly note ways that robots may en- less intuitive predictor variables) to both set optimal prices ter multiple aspects of social life, not only in (expected) and then change prices in real time. For example, Kanetix areas such as service and transportation, but also in helps Canadian customers find deals on car insurance by domains like the arts and music. allowing prospective buyers to compare and evaluate policies and rates offered by more than 50 providers. Scott Emberley, the Business Development Director of integrate.ai, which The current state and likely evolution of AI partnered with Kanetix to build an AI application, indicated that the goal was identify three sets of customers (1) those Short- and medium-term time horizon highly likely to buy, (2) those very unlikely to buy, and (3) those in-between. Thereafter, Kanetix would direct their ad- In Fig. 1, we combine all the above considerations to depict vertising towards these “in-between” customers, which would the current state of AI and its likely evolution. The upper half provide the greatest returns, and not expend efforts on those of Fig. 1 (four cells) relates to task automation and thus the either very likely to buy or very unlikely to buy. With four likely state of AI in the short to medium time horizon. The years of data, integrate.ai developed a machine learning model lower half of Fig. 1 (two cells) relates to context awareness that could identify such customers. Five months later, Kanetix applications that are only likely in the long term (if at all), due estimated 2.3 times return on its AI investment, and a more to the constraints associated with the current state of AI. Note than 20% increase in sales among previously undecided cus- tomers. In another example, the Bank of Montreal (BMO) uses IBM Interact to analyze customer data across all its chan- Compensatory consumption is consumption “motivated by a desire to offset nels and identify personalized product offerings. If a customer or reduce a self-discrepancy” (Mandel et al. 2017,p. 134). J. of the Acad. Mark. Sci. (2020) 48:24–42 33 has been exploring mortgages on BMO’s site and later calls location exists. Instead, customers fill out style surveys, pro- the contact center, IBM Interact prioritizes the list of available vide their physical measurements, evaluate sample styles, cre- mortgage offers for the contact center service agent—in effect ate links to their Pinterest boards, and send in personal notes. augmenting agents’ capabilities and facilitating more relevant As may be expected, customers have trouble explicating their customer conversations. exact style preferences using words and numbers, but their pins and likes can be (better) indicators of their preferences. Cell 2: Controller of data Efforts to analyze non-numeric data Stitch Fix’s proprietary machine learning algorithms examine offer the potential to improve understanding of what cus- numbers, words, and Pinterest pins, then summarize the find- tomers want, and firms’ customer service. Some AI applica- ings for the company’s fashion stylists, who in turn select tions can analyze non-numeric data (in some cases, after con- suitable clothing to send to each customer. The above example version to numeric data), primarily using speech and image illustrates the need to suitably balance AI input and human recognition capabilities achieved with deep learning neural input; senior managers from Stitch Fix told us that—in their networks (Chui et al. 2018). For example, Conversica AI, as experience—their AI works best when it augments the manifested in a virtual AI assistant named Angie, sends out- (human) stylists’ capabilities. bound emails to up to 30,000 leads per month, then interprets Noting that the AI applications in companies like the responses to identify the most promising leads (Power Conversica and Stitch Fix use all types of data (i.e., use nu- 2017). Angie engages in initial conversation with the pros- meric data and non-numeric data), we term the AI applications pect, and then routes to most promising leads to a (human) in this cell as reflecting “Controller of Data.” salesperson. In effect, Conversica’s AI augments salespersons’ capabilities. In a pilot test with a telecommuni- Cell 3: Numerical data robot This cell is similar to cell 1, cations company called Century Link, Angie appropriately except that it incorporates AI embedded in a robotic form, understood more than 95% of emails received (and sent the and so these AI applications can best be described as robots rest to human agents for interpretation), and Century Link that process numerical data inputs. Such robots are well suited earned a 20-fold return on its investments in Angie. to retail environments with well-structured operations. At The Stitch Fix’s business model offers another example. As Café X, for example, a robot barista can serve up to 120 we noted, Stitch Fix delivers apparel directly to customers coffees per hour (Hochman 2018). Each robotic barista fea- (Wilson et al. 2016), without requiring the customers to actu- tures a $25,000, six-axis animatronic arm. Customers place ally engage in a formal shopping task. No Stitch Fix retail orders on a kiosk touchscreen (or via an app), so all inputs Fig. 1 AI framework Task Digital form Robot form automation Analyze 1 – Controller of 3 – Numerical Data Robot technologies, numbers Numerical Data deployed currently or to Business Use Case Business Use Case be deployed in Kanetix Café X the short to IBM Tipsy Robot medium term Analyze 2 – Controller of Data 4 – Data Robot text, voice, Business Use Case Business Use Case faces, Conversica Lowebot images Stitch Fix 84.51/ Kroger Replika Walmart/ Bossa Nova K5 from Knightscape Context Digital form Robot form awareness Analyze 5 – Data Virtuoso 6 – Robot Expert technologies numbers, that may be text, Example Use Case Example Use Case deployed in the voice, Jarvis Dorian long term faces, image 34 J. of the Acad. Mark. Sci. (2020) 48:24–42 are numeric. As in a regular coffeehouse, customers can select human input (Lashinsky 2019). Put simply, the dream of get- various options: latte or espresso, with different amounts of ting into a driverless car outside in one city, falling asleep, and froth, and various ingredients such as organic Swedish oat waking up in another city is not reality and may not be achieved milk. The goal is not to replace baristas, but rather to augment anytime soon. Even the less consequential forms of AI remain baristas’ capabilities by taking over more routine operations. problematic. Google’s AlphaGo Zero might have successfully The Cafe X robot barista augments the capabilities of the learned the complex game of Go in a short period, using adver- human barista, who can then focus on providing high- sarial networks that pit two (competing) AI systems against quality customer service, and also facilitating what the com- each other so that they can learn; yet in this case, the outcome pany calls “coffee education” (e.g., managing tastings). space was very well defined. Furthermore, all these AI systems received significant training data. In contrast, the outcome Cell 4: Data robot This cell is similar to cell 2, except that the spaces (i.e., business domains) for most likely AI applications robotic form can process all types of data (not just numeric are poorly defined, and relevant training data is hard to obtain. data). For example, the Lowebot at Lowe’sHome These points reiterate the challenges of moving from task au- Improvement stores (Hullinger 2016) can scan a product held tomation to context awareness. As such, the use cases we pres- up by a customer (or listen to the customer speak the name of ent for the last two cells are hypothetical, and this section is the desired product), confirm whether the item is in stock, and deliberately brief, reflecting that our discussion is more aspira- then roll along with the customer to the exact spot in the store tional than descriptive of any near-term reality. where he or she can find the product. This task requires com- prehension and examination of both numeric and non-numeric Cell 5: Data virtuoso Advanced AI could be embedded in a data, as well as an indoor navigation capability, which repre- digital form, as exemplified by the AI Jarvis in Iron Man sents a significant advance over the capabilities embodied in movies. Jarvis has advanced data capabilities that can examine the Café X robot. Using the Lowebot augments the capabili- multiple data types. Perhaps most notably, Jarvis adapts to ties of Lowe’s human sales associates, allowing focus on more new contexts, beyond those for which it has been trained, such complex customer service requests. as when it hides from the more advanced AI Ultron and finds Other retailers have similar applications. Our discussions with ways to thwart Ultron’s hacking attempts. Futurists would senior managers at 84.51 indicate that they are working with have us believe that such AI will emerge in the long term, Kroger to implement in-store robots that can identify misshelved with strong predictive abilities for customers’ preferences or out-of-stock items. In another example, Walmart has partnered and high capability levels for managing customer service. with Bossa Nova Robotics to deploy robots in its stores to scan Thus, the term virtuoso seems appropriate for such AI. shelves. The goal appears to be to get robots to perform tasks that repeat and are predictable, enabling (human) associates to focus Cell 6: Robot experts An advanced AI also could be embedded on serving customers (Avalos 2018). in a robot form, such as the AI Dorian from the television Finally, security robots, such as the K5 from Knightscope, show Almost Human. Dorian’s advanced capabilities include roam offices and malls at night. These robots have better sens- facial recognition, bio scans, analyses of non-numeric stimuli ing capabilities than humans, because they incorporate ther- such as DNA, speed-reading, speaking multiple languages, mal cameras and other high-technology sensing tools. Here and taking the temperature of fluids using his finger. Like again, the objective is to augment human security guards’ Jarvis, Dorian can adapt to a variety of new contexts. capabilities (Robinson 2017). Futurists predict that such robot experts will emerge in the long term to serve as companions that meet various customer Long-term time horizon needs (e.g., in-home service, home security, medical support). Such robots even might be able to bond emotionally with For completeness, we also examine what might happen when (human) customers, and potentially replace human partners AI applications incorporate context awareness, as summarized and animal partners. in the two cells in the lower half of Fig. 1. We reiterate that there is no indication that such developments will occur in the short or medium term, as exemplified by the case of driverless Agenda for future research cars. Tesla has removed any “self-driving” labels from its website, noting that these labels were causing confusion Having described AI and presented a framework to better (Hawkins 2019). The CEO of Waymo admits that driverless understand it, we pivot to outlining some important areas for cars are unable to drive in poor weather conditions without future research. These include how firms may need to change their marketing strategy, how customers’ behaviors will be impacted, and issues relevant to policymakers. We outline This consulting firm is a subsidiary of Kroger and provides retail insights to Kroger and its partners; it has strong analytics and AI capabilities. these areas in Fig. 2, linking these to the cells in Fig. 1. J. of the Acad. Mark. Sci. (2020) 48:24–42 35 AI and marketing strategy how best to make predictions for RNPs, research can also examine how best to combine AI-driven insights with human Predictive ability Because AI can help firms predict what cus- judgment. tomers will buy, using AI should lead to substantial improve- AI is expected to play an important role in predicting not ments in predictive ability. Contingent on levels of predictive only what customers want to buy, but also what price to accuracy, firms may even substantially change their business charge, and whether price promotions should be offered models, providing goods and services to customers on an on- (Shankar 2018). Price and price promotions are important going basis based on data and predictions about their needs. drivers of sales (Biswas et al. 2013;Guha etal. 2018), and Multiple research opportunities thus emerge, related to differ- so are an important area of research for marketing researchers. ent customer purchase behaviors and marketing strategies. Thus, an important area for future research relates to how AI One especially important research area may relate to how well can be best used to predict what prices are optimal and wheth- prediction AI–driven algorithms may extend to forecasting er or not price promotions should be offered. demand for really new products (RNPs; described in Zhao Another important research avenue pertains to allocations et al. 2012). AI algorithms probably have good predictive of advertising resources. Much advertising focuses on devel- ability for incrementally new products; the open question is oping customer awareness and driving customers’ informa- whether they will have good predictive ability for RNPs. For tion search. Would these advertising dollars be required AI algorithms to do so would presumably require data on in the future, wherein firms may be able to better pre- RNPs that would be used in training machine learning models; dict customers’ preferences, and thus would not need to this is often not readily available. Further, when examining advertise as much? Fig. 2 Research agenda for AI. Task Digital form Robot form Notes: As noted in the text, the automation Analyze 1 – Controller of Numerical 3 – Numerical Data numbers Data Robot sales AI application will be more effective if it can process both Research Agenda Research Agenda numeric and non-numeric data, Predictive ability (MS) similar to Controller of and hence is more related to the AI adoption (CB) Numerical Data cell Controller of Data cell. This is - negative response to AI + more likely for more advanced - state and trait moderators Affective responses to robots, and so more likely to be AI usage (CB) robots (CB) relevant to robots able to handle - primed mindset non-numeric data (notably voice), - Post AI issues (CB) and hence more related to perhaps - perceived loss of the Data Robot cell, but more so autonomy to the Robot Expert cell - state and trait moderators Data privacy (P) Bias (P) Ethics (P) Analyze 2 – Controller of Data 4 – Data Robot text, voice, Research Agenda Research Agenda faces, similar to Controller of similar to Controller of images Numerical Data cell Data cell + + Sales (MS) UVH (CB) AI adoption for spiritual well- Loss of human being (CB) connectedness (CB) Context Digital form Robot form awareness Analyze 5 – Data Virtuoso 6 – Robot Expert numbers, text, Research Agenda Research Agenda voice, similar to Controller of Data similar to Data Robot faces, cell cell image Notes: MS = marketing strategy; CB = consumer behavior; P = public policy. 36 J. of the Acad. Mark. Sci. (2020) 48:24–42 Sales and AI As we discussed with regard to Conversica, AI Clothing suggestions from stylists who “swiped” on the app may alter all stages of the sales process, from prospecting to similarly to particular customers elicited more positive re- pre-approach to presentation to follow-up (Singh et al. 2019; sponses from the customers (i.e., both qualitative feedback Syam and Sharma 2018). Thus, a wide variety of research about the stylist and increased sales of clothes curated by that questions arise: stylist). When implementing AI, firms thus may achieve better outcomes if they let their data scientists spend some amount of & Can AI analyze customer communication and other cus- time on unauthorized “pet projects,” a research and develop- tomer information (e.g., social media posts) in ways to ment practice already in place in firms like 3 M (Shum and Lin devise future communications that are more persuasive 2007). Researching the best way to implement AI, to take or increase engagement? advantage of both expected and unexpected benefits, is a fruit- & Can AI provide real-time feedback to salespeople to help ful area for research. them improve their sales pitches, based on assessments of customers’ verbal and facial responses? Modeling the evolution of AI Finally, firms need to develop & How might AI combine text and other communication realistic expectations, because “in the short run, AI will pro- inputs (e.g., voice data), actual customer behavior, and vide evolutionary benefits; in the long run, it is likely to be other information (e.g., behaviors of similar customers) revolutionary” (Davenport 2018, p. 7). That is, the benefits of to predict repurchases? This effort demands non-numeric AI could be overestimated in the short term but data, in line with cells 2, 4, 5 and 6. underestimated in the long term, a point (sometimes called & Considering Luo et al.’s(2019) findings, how should Amara’s Law) in accordance with Gartner’shype cycle model firms deploy AI sales bots effectively? of how new technologies evolve (Dedehayir and Steinert 2016; also see van Lente et al. 2013;Shankar 2018). This view Answering these questions could help firms design sales to is popular among practitioners, according to our personal dis- take the most advantage of AI. cussions and interviews with various senior managers. Will In addition, firms need to consider how they (re)organize the evolution of AI reflect this model, or will its evolution their sales and innovation processes. These points are not differ and more closely map onto models that also integrate listed in Fig. 2, as they do not tie neatly into the cells shown more traditional innovation models (e.g., Roger’smodel, the in Fig. 1. Bass model)? Research that tests which innovation model best predicts AI evolutions will be useful. Sales process In the presence of AI, how should sales be or- ganized and what skills will salespeople need? First, how best AI and customer behavior to structure the sales organization wherein organizational components include both AI bots and human salespeople. New technologies often alter customer behavior (e.g., Secondly, how should the firm manage the tradeoff between Giebelhausen et al. 2014; Groom et al. 2011; Hoffman and AI focusing on customers’ expressed needs versus salespeople Novak 2018;Moon 2003), and we expect that AI will do so as being relatively better able to manage issues like customer well. We propose three research topics, related to AI adoption, stewardship. Lastly, will salespeople be able to be trained/ to AI usage, and post-adoption issues. be able to manage customers’ concerns relating to AI, specif- ically issues related to data privacy and ethics. It is clear that AI adoption As a general point, due to a wide variety of fac- sales processes will require innovation related not only to AI tors, customers view AI negatively, which is a barrier to adop- technologies, but also in job design and skills (Barro and tion. As noted, these negative views often stem from cus- Davenport 2019). tomers’ sense that AI is unable to feel (Castelo et al. 2018; Gray 2017) orthatAIisrelativelyless ableto identify whatis AI innovation process Because the impact of AI is uncertain, unique about each customer (Longoni et al. 2019). Also Luo firms need to figure out how best to (continually) develop AI. et al. (2019) suggest that customers perceive AI bots as being In our discussions with senior managers at Stitch Fix, they less empathetic. Customers also are less likely to adopt AI in indicated that the company encourages its data scientists to consequential domains (Castelo et al. 2018; Castelo and Ward pursue projects on their own (Colson 2018), such that they 2016) and for tasks salient to their identity (Castelo 2019; continually engage in preliminary testing of new project ideas. Leung et al. 2018). One Stitch Fix data scientist created a Tinder-like app called Thus, an important area for future research, important from Style Shuffle, to allow users to indicate preferences for various the standpoint of both research and practice, would be to ex- clothing styles. This app not only informed stylists about cus- amine how best to mitigate the impact of the above. Initial tomers’ preferences (the expected benefit) but also helped brainstorming with fellow researchers and with practitioners match stylists with specific customers (an unexpected benefit). suggests that positioning AI as a learning (artificial) organism, J. of the Acad. Mark. Sci. (2020) 48:24–42 37 or else positioning the AI application as one that combines AI When AI is embedded in robots, the robots likely have and human inputs (as in Stitch Fix), may help partially miti- important roles in customers’ lives, functioning as frontline gate the impact of the points above. Longoni et al. (2019) service providers (Wirtz et al. 2018), companions, nannies, propose that offering customers the opportunity to slightly or pet replacements. In addition to the UVH-related challenges modify the AI may get these customers to look past unique- documented previously, some research results suggest ness neglect, and focus more on the benefits of personaliza- that interactions with AI-embedded robots trigger dis- tion. This too may be a way to mitigate the points raised prior. comfort and compensatory behaviors (Mende et al. 2019). It The discomfort with AI is accentuated in case the AI ap- is important to determine when customers perceive AI- plication is embedded in a robot. As robots become more embedded robots negatively and whether these perceptions humanlike, then due to the UVH, customers find these robots may improve over time. unnerving. Such factors may hinder AI adoption and deserve Finally, if customers’ ideal preferences actually differ from study. An interesting moderator of this effect—worth investi- their past behaviors (e.g., customers trying to stop eat- gating—may be whether the AI form is perceived by cus- ing unhealthy foods), AI might make it harder for them tomers as a servant or partner; UVH effects may be stronger to find and move toward their preferred options, by if AI achieves partner status. Also deserving of study is other only presenting them with choices reflecting their past ways of mitigating such effects. Early efforts in this direction behaviors. The widespread “retargeting” of digital ads is one involve trying to prime empathy, by convincing customers example of this phenomenon. How to train AI to best manage that robots have some ability to see things from the customers’ this issue? viewpoint, and (also) have some ability to feel sympathy for the customer if the customer were suffering (Castelo 2019). Post-adoption The downstream consequences of AI adoption Other possible methods could relate to anthropomorphizing also suggest some relevant research topics. In particular, cus- the AI, as this may persuade customers that the AI has some- tomers might perceive a loss of autonomy if AI can substan- what more empathy (this point needs to balance with concerns tially predict their preferences. In theory, because AI facilitates about the UVH). data-driven, micro-targeting marketing offerings (e.g., Gans Sociologists appear especially interested in how robots et al. 2017; Luo et al. 2019), customers should view offerings with embedded AI might make inroads into society (Boyd more favorably, because it reduces their search costs. Yet it and Holton 2018, p. 338), noting that “complexities arise also could undermine customers’ perceived autonomy, with when cultural preferences associated with human as opposed implications for their evaluations and choices (André et al. to machine delivery of personal services are considered. Do 2018). If customers learn that an AI algorithm can predict their … consumers find social robots acceptable?” Broadly speak- preferred choices, they may deliberately choose a non- ing, research can address how attitudes toward robots vary by preferred option, to reaffirm their autonomy (André et al. culture (Li et al. 2010). Beyond concerns associated with cul- 2018; Schrift et al. 2017). Such considerations evoke a variety ture, it may be pertinent to examine which other trait factors of research questions. For example, which factors determine determine whether customers are willing to have their hair whether (and how much) customers value perceived autono- styled by robots or accept childcare/elderly care services de- my in AI-mediated choice settings? In this regard, it may be livered by robots (Pedersen et al. 2018). In addition to physical helpful to examine individual difference variables, such as well-being considerations, some sociologists suggest that ro- culture and whether customers regard AI as a servant or part- bots may assist with spiritual well-being (Fleming 2019), as ner. Research also might address state factors, such as the exemplified by the robot priest BlessU-2 (Sherwood 2017) product type; perceived autonomy may be less relevant for and Buddhist monk Xian’er (Andrews 2016). Understanding utilitarian product choices than for hedonic ones, because of how robots with embedded AI can assist in various ways, differential links to customers’ identity. beyond improving customers’ physical well-being, is a good Also, there is a generalized fear of a loss of human con- area for research. nectedness, if humans form bonds with robots with embedded AI. The popular press (e.g., Marr 2019) stokes concerns about AI usage When customers interact with an AI application, it robots with embedded AI becoming popular (over humans) as might prime a low-level construal mindset (Kim and partners. Robots like Harmony (by Realbotix) appear promis- Duhachek 2018). Research should determine what other ing in this regard, able to take on different personae and dis- mindsets might be primed by AI, e.g., AI may prime preven- play some expression. But such robots arguably could be tion focus among customers for whom AI is a relatively new damaging to society at large, by increasing social isolation, technology. Related insights would have implications for how reducing the incidence of marriage, or reducing birthrates— the AI application should communicate with the customer, which is critical for countries like Japan, where birthrates are because communication exerts stronger impacts when it fits already low. This point suggests some interesting research with (a primed) mindset. opportunities. 38 J. of the Acad. Mark. Sci. (2020) 48:24–42 AI and policy issues choices. Does the trade-off depend on the product category or the level of the customer’s trust in the firm, for example? Finally, AI is of interest to policymakers. We note three broad Also, how would this trade-off shift over time? areas in which policymakers seek to ensure that firms strike a suitable balance between their own commercial interests and Bias The potential algorithmic bias embedded in AI applica- the interests of customers: data privacy, bias, and ethics. All tions could stem from multiple causes (Villasenor 2019), in- three of these areas generally align with all cells in Fig. 2. cluding the data sets that inform AI. For example, Amazon abandoned a tool that used AI intelligence to rate job appli- Data privacy Today, the combination of AI and big data im- cants, in part because it discriminated against female appli- plies that firms know much about their customers (Wilson cants (Weissman 2018). This bias emerged because the train- 2018). Hence, two issues deserve research attention. First, ing data sets used to develop the algorithm were based on data customers worry about the privacy of their data (Martin and relating to previous applicants, who were predominantly men. Murphy 2017; Martin et al. 2017). Privacy is complicated Exacerbating the issue, many AI algorithms are opaque black (Tucker 2018), for three reasons: (1) the low cost of storage boxes, so it is difficult to isolate which exact factors these implies that data may exist substantially longer than was algorithms consider. Testing for whether there is bias in AI intended, (2) data may be repackaged and reused for rationales applications is an important topic. different than those intended, and (3) data for a certain indi- In addition, AI may not be able to distinguish attributes that vidual may contain information about other individuals. could induce potential bias. Villasenor (2019)argues asfollows. Policy related to data privacy requires balancing two compet- In general, it may not be offensive when insurance companies ing priorities. Too little protection means that customers may treat men and women differently, with one set of premiums for not adopt AI-related applications; too much regulation may male drivers and another set of premiums for female drivers. strangle innovation. Does this mean that it is acceptable for AI to calculate auto Second, important research questions pertain to whether insurance premiums based on religion? Many would argue that data privacy management efforts should be driven by legal basing auto insurance premiums on religion is offensive, but regulations or self-regulation, in that “it is not clear yet if from the point of view of an AI algorithm designed to “slice- market driven incentives will be sufficient for firms to adopt and-dice” data in every way possible, the distinction between policies that favor consumers or whether regulatory oversight using gender versus religion, as a basis to determining auto is required to ensure a fair outcome for consumers” (Verhoef insurance rates, may not be obvious. All this suggests that issues et al. 2017, p. 7). Cultural perspectives on data privacy also relating to bias remain a non-trivial problem (Knight 2017). vary, which is an important consideration; some have sug- gested that the lack of data privacy in China, for example, is Ethics Finally, AI developers must grapple with ethics; we consistent with Confucian cultural ideals (Smith 2019). highlight two issues. First, data privacy choices may reflect Third, we need insights into how best to acknowledge and afirm’s strategy (e.g., if it wants to be perceived as a trusted address privacy concerns at the moment data are collected, as firm; Martin and Murphy 2017; also see Goldfarb and Tucker well as how to manage data privacy failures (e.g., data 2013) but also could be driven by ethical concerns. In this breaches). Amazon already sells doorbells with cameras (the sense, research should address “how can normative ethical Ring device) and may be planning to add facial identification theory pave the way for what organizations should be doing AI to the devices (Fowler 2019). Customers thus may become to exceed consumer privacy expectations, as well as to over- concerned if Amazon has access to data recorded through comply with legal mandates in order to preserve their ability to Ring, which it could use or sell. Neighbors also might protest self-regulate” (Martin and Murphy 2017, p. 152). A related if Ring cameras record their front yard activities without their research topic might involve examining how ethical concerns permission. Also, the data from Ring arguably could be about AI vary across cultures. subpoenaed by law enforcement agencies or obtained illegally Second, firms choose to deploy AI by defining which prob- by hackers. Such issues reflect topics for further research. lems the AI will tackle. For example, two Stanford researchers Finally, we consider the privacy–personalization paradox used deep neural networks to identify people’ssexualorien- (Aguirre et al. 2015). Customers must balance privacy con- tation, merely by analyzing facial images (Wang and Kosinski cerns against the benefits of personalized recommendations 2018). The deep neural network tools (vs. human judges) were and offers. Important questions relate to how customers deter- better able to differentiate between gay and straight men. mine the optimal trade-off, including which individual differ- However, the work raised ethical concerns, in that many ar- ence variables and state variables might moderate their gued that this AI-based technology may be used by spouses on their partners (if they suspected their partners were closet- ed), or—more frighteningly—may be used by certain govern- Firms are aware of this, and are taking steps to suitably respond (Deloitte Insights, as reported in Schatsky et al. 2019) ments to “out” and then prosecute certain populations (Levin J. of the Acad. Mark. Sci. (2020) 48:24–42 39 collection and trust-building strategies on online advertisement ef- 2017). An important topic for research thus is to address up- fectiveness. JournalofRetailing, 91(1), 34–49. front the types of applications for which AI should be used for André, Q., Carmon, Z., Wertenbroch, K., Crum, A., Frank, D., Goldstein, (or, should not be used for). W., et al. (2018). Consumer choice and autonomy in the age of artificial intelligence and big data. Customer Needs and Solutions, 5(1–2), 28–37. Andrews, T. (2016). Meet the robot monk spreading the teachings of Conclusion Buddhism around China. Washington Post, April 27. Retrieved February 11, 2019 from https://www.washingtonpost.com/news/ This paper outlines a framework to understand how AI will morning-mix/wp/2016/04/27/meet-the-robot-monk-spreading-the- teachings-of-buddhism-around-china/?utm_term=.fed52d90bff3. impact the future of marketing, specifically to outline how AI Accessed 11 Feb 2019 may influence marketing strategies and customer behaviors. Antonio, V. (2018). How AI is changing sales. Harvard Business Review, We build on prior work, as well as build from extensive inter- July 30. Retrieved February 11, 2019 from https://hbr.org/2018/07/ actions with practitioners. First, we develop a multidimension- how-ai-is-changing-sales. al framework for the evolution of AI, noting the importance of Avalos, G. (2018). Walmart tests shelf-scanning robots in Bay Area. The Mercury News, March 20. Retrieved February 11, 2019 from https:// dimensions pertaining to intelligence levels, task types, and www.mercurynews.com/2018/03/20/walmart-tests-shelf-scanning- whether the AI is embedded in a physical robot. In so doing, robots-bay-area/. we provide the first attempt to integrate all three dimensions in Barro, S., & Davenport, T. H. (2019). People and machines: Partners in a single framework. We also make two (cautionary) points. innovation. MIT Sloan Management Review, 60(4), 22–28. Baum, S. D., Goertzel, B., & Goertzel, T. G. (2011). How long until First, the short to medium term impacts of AI may be more human-level AI? Results from an expert assessment. limited than the popular press would suggest. Second, we Technological Forecasting and Social Change, 78(1), 185–195. suggest that AI will be more effective if it is deployed in ways Bettman, J. (1973). Perceived risk and its components: A model and that augment (rather than replace) human managers. empirical test. JournalofMarketing, 10(2), 184–190. To examine the full scope of the impact of AI, we propose a Biswas, A., Bhowmick, S., Guha, A., & Grewal, D. (2013). Consumer evaluations of sale prices: Role of the subtraction principle. Journal research agenda covering three broad areas: (1) how firms’ of Marketing, 77(4), 49–66. marketing strategies will change, (2) how customers’ behav- Boyd, R., & Holton, R. J. (2018). Technology, innovation, employment iors will change, and (3) issues related to data privacy, bias, and power: Does robotics and artificial intelligence really mean and ethics. This research agenda warrants consideration social transformation? Journal of Sociology, 54(3), 331–345. Burrows, L. (2019). The science of the artificial. Retrieved June 12, 2019 by academia, firms, and policy experts, with the recog- from https://www.seas.harvard.edu/news/2019/05/science-of- nition that although AI already has had some impact on artificial. marketing, it will exert substantially more impact in the Byrnes, J. P., Miller, D. C., & Schafer, W. D. (1999). Gender differences future, and so there is much still to learn. We hope that in risk taking: A meta-analysis. Psychological Bulletin, 125(3), 367. this research agenda motivates and guides continued re- Carpenter, J. (2015). IBM’s Virginia Rometty tells NU grads: Technology will enhance us. Retrieved February 11, 2019 from https://www. search into AI. chicagotribune.com/bluesky/originals/ct-northwestern-virginia- rometty-ibm-bsi-20150619-story.html. Acknowledgements The authors thank Eric Colson (Stitch Fix), James Castelo, N. (2019). Blurring the line between human and machine: Kotecki (Infinia ML) and Charu Kohli (CognitiveScale), who shared their Marketing artificial intelligence (doctoral dissertation). Retrieved time, materials and insights with the authors as this paper developed. from Columbia University Academic Commons. https://doi.org/ Thomas Davenport and Dhruv Grewal dedicate this article to the memory 10.7916/d8-k7vk-0s40. of their dear friend and colleague, Dean Bala Iyer of Babson University. Castelo, N., & Ward, A. (2016). Political affiliation moderates attitudes towards artificial intelligence. In P. Moreau & S. Puntoni (Eds.), NA Open Access This article is distributed under the terms of the Creative - advances in consumer research (pp. 723–723). Duluth, MN: Commons Attribution 4.0 International License (http:// Association for Consumer Research. creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Castelo, N., Bos, M., & Lehman, D. (2018). Consumer adoption of al- distribution, and reproduction in any medium, provided you give appro- gorithms that blur the line between human and machine. Graduate priate credit to the original author(s) and the source, provide a link to the School of Business: Columbia University Working Paper. Creative Commons license, and indicate if changes were made. Chui, M., Manyika, J., Miremadi, M., Henke, N., Chung, R., Nel, P., & Malhotra, S. (2018). Notes from the AI frontier: Applications and value of deep learning. McKinsey global institute discussion paper, References April 2018. Retrieved June 12, 2019 from https://www.mckinsey. com/featured-insights/artificial-intelligence/notes-from-the-ai- frontier-applications-and-value-of-deep-learning. Adami, C. (2015). Artificial intelligence: Robots with instincts. Nature, Colson, E. (2018). Curiosity-driven data science. Harvard Business 521(7553), 426–427. Review, November 27. Retrieved February 11, 2019 from https:// Agrawal, A., Gans, J. S., & Goldfarb, A. (2018). Prediction machines: hbr.org/2018/11/curiosity-driven-data-science The simple economics of artificial intelligence. Harvard Business Columbus, L. (2019). 10 charts that will change your perspective of AI in School Press. marketing. Forbes, July 07. Retrieved on July 09, 2019 from https:// Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., & Wetzels, M. (2015). www.forbes.com/sites/louiscolumbus/2019/07/07/10-charts-that- Unraveling the personalization paradox: The effect of information will-change-your-perspective-of-ai-in-marketing/amp/ 40 J. of the Acad. Mark. Sci. (2020) 48:24–42 Davenport, T. H. (2018). The AI advantage: How to put the artificial Hawkins, A. (2019). No, Elon, the navigate on autopilot feature is not intelligence revolution to work. MIT Press. ‘full self driving’. Retrieved February 11, 2019 from https://www. theverge.com/2019/1/30/18204427/tesla-autopilot-elon-musk-full- Davenport, T. H., & Kirby, J. (2016). Just how smart are smart machines? self-driving-confusion. MIT Sloan Management Review, 57(3), 21–25. Hayes, A. (2015). The unintended consequences of self-driving cars. Davenport, T. H., & Ronanki, R. (2018). Artificial intelligence for the real Retrieved February 11, 2019 from https://www.investopedia.com/ world. Harvard Business Review, 96(1), 108–116. articles/investing/090215/unintended-consequences-selfdriving- Davenport, T. H., Dalle Mule, L., & Lucker, J. (2011). Know what your cars.asp. customers want before they do. Harvard Business Review, 89(12), Hochman, D. (2018). This $25,000 robotic arm wants to put your 84–92. Starbucks barista out of business. Retrieved February 11, 2019 from Dedehayir, O., & Steinert, M. (2016). The hype cycle model: A review https://www.cnbc.com/2018/05/08/this-25000-robot-wants-to-put- and future directions. Technological Forecasting and Social your-starbucks-barista-out-of-business.html. Change, 108,28–41. Hoffman, D. L., & Novak, T. P. (2018). Consumer and object experience Fleming, P. (2019). Robots and organization studies: Why robots might in the internet of things: An assemblage theory approach. Journal of not want to steal your job. Organization Studies, 40(1), 23–38. Consumer Research, 44(6), 1178–1204. Fowler, G. (2019). The doorbells have eyes: The privacy battles brewing Huang, M. H., & Rust, R. T. (2018). Artificial intelligence in service. over home security cameras. Retrieved February 11, 2019 from Journal of Service Research, 21(2), 155–172. https://www.sltrib.com/news/business/2019/02/01/doorbells-have- Hullinger, J. (2016). What the Lowe’s robot will do for you – and the eyes/. future of retail. Retrieved February 11, 2019 from http://campfire- French, K. (2018). Your new best friend: AI chatbot. Retrieved February capital.com/retail-innovation/sales-channel-innovation/what-the- 13, 2019 from https://futurism.com/ai-chatbot-meaningful- lowes-robot-will-do-for-you-and-the-future-of-retail/. conversation. Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’sthe Gans, J., Agrawal, A., & Goldfarb, A. (2017). How AI will change strat- fairest in the land? On the interpretations, illustrations, and implica- egy: A thought experiment. Harvard business review online. tions of artificial intelligence. Business Horizons, 62(1), 15–25. Retrieved February 11, 2019 from https://hbr.org/product/how-ai- Kidd, C. D., & Breazeal C. (2008, September). Robots at home: will-change-strategy-a-thought-experiment/H03XDI-PDF-ENG. Understanding long-term human-robot interaction. In Proceedings Gaudin, S. (2016). At stitch fix, data scientists and a.I. become personal of the IEEE International Conference of Intelligent Robot Systems stylists. Retrieved February 11, 2019 from https://www. (pp. 3230–3235). computerworld.com/article/3067264/artificial-intelligence/at-stitch- Kim, T., & Duhachek, A. (2018). The impact of artificial agents on per- fix-data-scientists-and-ai-become-personal-stylists.html. suasion: A construal level account. ACR Asia-Pacific Advances. Ghahramani, Z. (2015). Probabilistic machine learning and artificial in- Knight, W. (2017). Forget killer robots – bias is the real AI danger. telligence. Nature, 521(7553), 452–459. Retrieved February 11, 2019 from https://www.technologyreview. Giebelhausen, M., Robinson, S. G., Sirianni, N. J., & Brady, M. K. com/s/608986/forget-killer-robotsbias-is-the-real-ai-danger/. (2014). Touch versus tech: When technology functions as a barrier Kober, J., Bagnell, J. A., & Peters, J. (2013). Reinforcement learning in or a benefit to service encounters. Journal of Marketing, 78(4), 113– robotics: A survey. The International Journal of Robotics Research, 32(11), 1238–1274. Goldfarb, A., & Tucker, C. (2013). Why managing consumer privacy can Kwak, S., Kim, Y., Kim, E., Shin, C., & Cho, K. (2013). What makes be an opportunity. MIT Sloan Management Review, 54(3), 10–12. people empathize with an emotional robot? The impact of agency Gray, K. (2017). AI can be a troublesome teammate. Harvard Business and physical embodiment on human empathy for a robot. In Review, July 20. Retrieved February 11, 2019 from https://hbr.org/ Proceedings of the IEEE International. Symposium Robot Human 2017/07/ai-can-be-a-troublesome-teammate. –185). Interaction Community (pp. 180 Gray, K., & Wegner, D. M. (2012). Feeling robots and human zombies: Lammer, L., Huber, A., Weiss, A., Vincze, M. (2014). Mutual care: How Mind perception and the uncanny valley. Cognition, 125(1), 125– older adults react when they should help their care robot. In Proceedings of the International Symposium of Social Study AI Groom, V., Srinivasan, V., Bethel, C. L., Murphy, R., Dole, L., & Nass, C. Simulation Behavior (pp. 3-4). (2011, May). Responses to robot social roles and social role framing. Larson, K. (2019). Data privacy and AI ethics stepped to the fore in 2018. In Proceedings of the IEEE International Conference on Retrieved February 11 from https://medium.com/@Smalltofeds/ Collaboration Technologies and Systems (CTS) (pp. 194-203). data-privacy-and-ai-ethics-stepped-to-the-fore-in-2018- Guha, A., Biswas, A., Grewal, D., Verma, S., Banerjee, S., & Nordfält, J. 4e0207f28210. (2018). Reframing the discount as a comparison against the sale Lashinsky, A. (2019). Artificial intelligence: Separating the hype from the price: Does it make the discount more attractive? Journal of reality. Retrieved Feburary 11, 2019 from http://fortune.com/2019/ Marketing Research, 55(3), 339–351. 01/22/artificial-intelligence-ai-reality/. Gustafsod, P. E. (1998). Gender Differences in risk perception: LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, Theoretical and methodological perspectives. Risk Analysis, 18(6), 521(7553), 436–444. 805–811. Lee, A. Y., Keller, P. A., & Sternthal, B. (2009). Value from regulatory Harding, K. (2017). AI and machine learning for predictive data scoring. construal fit: The persuasive impact of fit between consumer goals Retrieved February 11, 2019 from https://www.objectiveit.com/ and message concreteness. Journal of Consumer Research, 36(5), blog/use-ai-and-machine-learning-for-predictive-lead-scoring on 735–747. 13 February 2019. Leung, E., Paolacci, G., & Puntoni, S. (2018). Man versus machine: Haslam, N., Bain, P., Douge, L., Lee, M., & Bastian, B. (2005). More Resisting automation in identity-based consumer behavior. Journal human than you: Attributing humanness to self and others. Journal of Marketing Research, 55(6), 818–831. of Personality and Social Psychology, 89(6), 937–950. Levin, S. (2017). New AI can guess whether you’re gay or straight from a Hassler, C. A. (2018). Meet Replika, the AI Bot that wants to be your best photograph. Retrieved February 11, 2019 from https://www. friend. Retrieved February 11, 2019 from https://www.popsugar. theguardian.com/technology/2017/sep/07/new-artificial- com/news/Replika-Bot-AI-App-Review-Interview-Eugenia- intelligence-can-tell-whether-youre-gay-or-straight-from-a- Kuyda-44216396. photograph. J. of the Acad. Mark. Sci. (2020) 48:24–42 41 Li, D., Rau, P. P., & Li, Y. (2010). A cross-cultural study: Effect of robot Pedersen, I., Reid, S., & Aspevig, K. (2018). Developing social robots for appearance and task. International Journal of Social Robotics, 2(2), aging populations: A literature review of recent academic sources. 175–186. Sociology Compass, 12(6). Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to Power, B. (2017). How AI is streamlining marketing and sales. Harvard Medical Artificial Intelligence. Journal of Consumer Research Business Review, June 12. Retrieved February 11, 2019 from (forthcoming). https://hbr.org/2017/06/how-ai-is-streamlining-marketing-and- Lowy, J. (2016). Self-driving cars are ‘absolutely not’ ready for deploy- sales. ment. Retrieved February 11, 2019 from https://www.pbs.org/ Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J. F., newshour/science/self-driving-cars-are-absolutely-not-ready-for- Breazeal, C., et al. (2019). Machine behavior. Nature, 568(7753), deployment. 477–486. Luo, X., Tong, S., Fang, Z. & Zhe, Q. (2019). Machines vs. humans: The Reese, B. (2018). The fourth age: Smart robots, conscious computers and impact of chatbot disclosure on consumer purchases. Unpublished the future of humanity. New York, NY: Atria Books. working paper. Robinson, M. (2017). The tech giants of Silicon Valley are starting to rely Mandel, N., Rucker, D. D., Levav, J., & Galinsky, A. D. (2017). The on crime-fighting robots for security. Retrieved February 11, 2019 compensatory consumer behavior model: How self-discrepancies from https://www.businessinsider.com/knightscope-security-robots- drive consumer behavior. Journal of Consumer Psychology, 27(1), microsoft-uber-2017-5. 133–146. Sandoval, E. B., Brandstetter, J., Obaid, M., & Bartneck, C. (2016). Marr, B. (2019). How robots, IoT and artificial intelligence are changing Reciprocity in human-robot interaction: A quantitative approach how humans have sex. Retrieved June 12, 2019 from https://www. through the prisoner’s dilemma and the ultimatum game. forbes.com/sites/bernardmarr/2019/04/01/how-robots-iot-and- International Journal of Social Robotics, 8(2), 303–317. artificial-intelligence-are-changing-how-humans-have-sex/# Schatsky, D., Katyal, V., Iyengar, S., & Chauhan, R. (2019). Why enter- 3679d398329c. prises shouldn’t wait for AI regulation. Retrieved July 11, 2019 from Martin, K. D., & Murphy, P. E. (2017). The role of data privacy in https://www2.deloitte.com/insights/us/en/focus/signals-for- marketing. Journal of the Academy of Marketing Science, 45(2), strategists/ethical-artificial-intelligence.html 135–155. Schrift, R. Y., Wertenbroch, K., André, Q., & Frank, D.H. (2017). Martin, K. D., Borah, A., & Palmatier, R. W. (2017). Data privacy: Threatening free will. Presentation at the Symposium on Alienation Effects on customer and firm performance. Journal of Marketing, and Meaning in Production and Consumption, Technische 81(1), 36–58. Universität München. Mehta, N., Detroja, P., & Agashe, A. (2018). Amazon changes prices on Shankar, V. (2018). How artificial intelligence (AI) is reshaping retailing. its products about every 10 minutes — here's how and why they do Journal of Retailing, 94(4), vi–xi. it. Retrieved February 11, 2019 from https://www.businessinsider. Sherwood, H. (2017). Robot priest unveiled in Germany to mark 500 com/amazon-price-changes-2018-8?international=true&r=US&IR= years since reformation. The Guardian, may 20. Retrieved T. February 11, 2019 from https://www.theguardian.com/technology/ Mende, M., Scott, M. L., van Doorn, J., Grewal, D., & Shanks, I. (2019). 2017/may/30/robot-priest-blessu-2-germany-reformation- Service robots rising: How humanoid robots influence service ex- exhibition. periences and food consumption. Journal of Marketing Research, Shum, P., & Lin, G. (2007). A world class new product development best 56(4), 535–556. practices model. International Journal of Production Research, Metz, C. (2018). Mark Zuckerberg, Elon musk and the feud over killer 45(7), 1609–1629. robots. Retrieved February 11, 2019 from https://www.nytimes. Şimşek, Ö. F., & Yalınçetin, B. (2010). I feel unique, therefore I am: The com/2018/06/09/technology/elon-musk-mark-zuckerberg-artificial- development and preliminary validation of the personal sense of intelligence.html. uniqueness (PSU) scale. Personality and Individual Differences, Milgram, P., Takemura, H., Utsumi, A., & Kishino, F. (1995, December). 49(6), 576–581. Augmented reality: A class of displays on the reality-virtuality con- Singh, J., Flaherty, K., Sohi, R. S., Deeter-Schmelz, D., Habel, J., Le tinuum. Telemanipulator and telepresence technologies, 2351,282– Meunier-FitzHugh, K., et al. (2019). Sales profession and profes- sionals in the age of digitization and artificial intelligence technolo- Miller, G. (2016). Bots, chatbots and artificial intelligence, what the evo- gies: Concepts, priorities, and questions. Journal of Personal Selling lution is all (aBot) about. Retrieved June 12, 2019 from https:// & Sales Management,1–21. chatbotnewsdaily.com/bots-chatbots-and-artificial-intelligence- Smith, C. (2019) Understanding China’s Confucian edge in the global AI what-the-evolution-is-all-abot-about-a7e148dd067d. race, Quillette, Feb. 14, 2019, Retrieved from https://quillette.com/ Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., 2019/02/14/understanding-chinas-confucian-edge-in-the-global-ai- Bellemare, M. G., et al. (2015). Human-level control through deep race/ on July 6, 2019. reinforcement learning. Nature, 518(7540), 529–533. Steels, L., & Brooks, R. (2018). The artificial life route to artificial intel- Moon, Y. (2003). Don’t blame the computer: When self-disclosure mod- ligence: Building embodied, situated agents. Routledge. erates the self-serving bias. Journal of Consumer Psychology, 13(1– Syam, N., & Sharma, A. (2018). Waiting for a sales renaissance in the 2), 125–137. fourth industrial revolution: Machine learning and artificial intelli- Mori, M. (1970). The Uncanny Valley. Energy, 7(4), 33–35. gence in sales research and practice. Industrial Marketing Motyka, S., Grewal, D., Puccinelli, N. M., Roggeveen, A. L., Avnet, T., Management, 69,135–146. Daryanto, A., et al. (2014). Regulatory fit: A meta-analytic synthe- Tucker, C. (2018). Privacy, algorithms, and artificial intelligence. In The sis. Journal of Consumer Psychology, 24(3), 394–410. Economics of Artificial Intelligence: An Agenda. University of Müller, V. C., & Bostrom, N. (2016). Future progress in artificial intelli- Chicago Press. gence: A survey of expert opinion. In Fundamental issues of artifi- van Lente, H., Spitters, C., & Peine, A. (2013). Comparing technological cial intelligence (pp. 555–572). Springer, Cham. hype cycles: Towards a theory. Technological Forecasting and Parekh, J. (2018). Why Programmatic provides a better digital marketing Social Change, 80(8), 1615–1628. landscape. Retrieved February 13, 2019 from https://www.adweek. com/programmatic/why-programmatic-provides-a-better-digital- Verhoef, P. C., Stephen, A. T., Kannan, P. K., Luo, X., Abhishek, V., marketing-landscape/. Andrews, M., et al. (2017). Consumer connectivity in a complex, 42 J. of the Acad. Mark. Sci. (2020) 48:24–42 technology-enabled, and mobile-oriented world with smart prod- zdnet.com/article/big-privacy-the-data-privacy-compact-for-the- ucts. Journal of Interactive Marketing, 40,1–8. era-of-big-data-and-ai/. Villasenor, J. (2019). Artificial intelligence and bias: Four key challenges. Wilson J., Daugherty P., & Shukla P. (2016). How one clothing company Retrieved February 11, 2019 from https://www.brookings.edu/blog/ blends AI and human expertise. Harvard Business Review, techtank/2019/01/03/artificial-intelligence-and-bias-four-key- November 21. Retrieved February 11 from https://hbr.org/2016/11/ challenges/. how-one-clothing-company-blends-ai-and-human-expertise. Wainer, J., Feil-Seifer, D. J., Shell, D. A., & Mataric, M. J. (2006, Wirtz, J., Patterson, P. G., Kunz, W. H., Gruber, T., Lu, V. N., Paluch, S., September). The role of physical embodiment in human-robot inter- & Martins, A. (2018). Brave new world: Service robots in the front- action. In Proceedings of the IEEE International Symposium of line. Journal of Service Management, 29(5), 907–931. Robot Human Interaction Communication (pp.6-8). You, Q., Jin,H., Wang,Z., Fang,C., & Luo,J.(2016). Image captioning Wang, Y., & Kosinski, M. (2018). Deep neural networks are more accu- with semantic attention. In Proceedings of the IEEE conference on rate than humans at detecting sexual orientation from facial images. computer vision and pattern recognition (pp. 4651-4659). Journal of Personality and Social Psychology, 114(2), 246–257. Zhao, M., Hoeffler, S., & Dahl, D. W. (2012). Imagination difficulty and Weissman, J. (2018). Amazon created a hiring tool using a.I. it immedi- new product evaluation. Journal of Product Innovation ately started discriminating against women. October 10. Retrieved Management, 29(S1), 76–90. February 11, 2019 from https://slate.com/business/2018/10/ amazon-artificial-intelligence-hiring-discrimination-women.html. Publisher’snote Springer Nature remains neutral with regard to Wilson, S. (2018). Big privacy: The data privacy compact for the era of jurisdictional claims in published maps and institutional affiliations. big data and AI. Retrieved February 11, 2019 from https://www. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of the Academy of Marketing Science Springer Journals

How artificial intelligence will change the future of marketing

Download PDF
19 pages

Loading next page...
 
/lp/springer-journals/how-artificial-intelligence-will-change-the-future-of-marketing-aVaZ5DGpva

References (113)

Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2019
Subject
Business and Management; Business and Management, general; Marketing; Social Sciences, general
ISSN
0092-0703
eISSN
1552-7824
DOI
10.1007/s11747-019-00696-0
Publisher site
See Article on Publisher Site

Abstract

In the future, artificial intelligence (AI) is likely to substantially change both marketing strategies and customer behaviors. Building from not only extant research but also extensive interactions with practice, the authors propose a multidimensional framework for understanding the impact of AI involving intelligence levels, task types, and whether AI is embedded in a robot. Prior research typically addresses a subset of these dimensions; this paper integrates all three into a single framework. Next, the authors propose a research agenda that addresses not only how marketing strategies and customer behaviors will change in the future, but also highlights important policy questions relating to privacy, bias and ethics. Finally, the authors suggest AI will be more effective if it augments (rather than replaces) human managers. . . . . . Keywords Artificial intelligence Marketing strategy Robots Privacy Bias Ethics AI is going to make our lives better in the future. ride-sharing businesses must evolve to avoid being marginal- —Mark Zuckerberg, CEO, Facebook ized by AI-enabled transportation models; demand for auto- mobile insurance (from individual customers) and breathaly- zers (fewer people will drive, especially after drinking) will likely diminish, whereas demand for security systems that Introduction protect cars from being hacked will increase (Hayes 2015). Driverless vehicles could also impact the attractiveness of real In the future, artificial intelligence (AI) appears likely to in- estate, because (1) driverless cars can move at faster speeds, fluence marketing strategies, including business models, sales and so commute times will reduce, and (2) commute times processes, and customer service options, as well as customer will be more productive for passengers, who can safely work behaviors. These impending transformations might be best while being driven to their destination. As such, far flung understood using three illustrative cases from diverse indus- suburbs may become more attractive, vis-à-vis the case today. tries (see Table 1). First, in the transportation industry, driver- Second, AI will affect sales processes in various industries. less, AI-enabled cars may be just around the corner, promising Most salespeople still rely on a telephone call (or equivalent) to alter both business models and customer behavior. Taxi and as a critical part of the sales process. In the future, salespeople Thomas Davenport, Abhijit Guha, Dhruv Grewal and Timna Bressgott contributed to the writing of the paper. Mark Houston served as accepting Editor for this article. * Timna Bressgott Department of Technology, Operations, and Information t.bressgott@maastrichtuniversity.nl Management, Babson College, Babson Park, MA 02457, USA Department of Marketing, Darla Moore School of Business, Thomas Davenport University of South Carolina, Columbia, SC 29208, USA tdavenport@babson.edu Department of Marketing, Babson College, Babson Abhijit Guha Park, MA 02457, USA abhijit.guha@moore.sc.edu Department of Marketing and Supply Chain Management, Dhruv Grewal Maastricht University, Tongersestraat 53, 6211, LM dgrewal@babson.edu Maastricht, The Netherlands J. of the Acad. Mark. Sci. (2020) 48:24–42 25 will be assisted by an AI agent that monitors tele- insights about not only the ultimate promise of AI, but also conversations in real time. For example, using advanced voice the pathway and timelines along which AI is likely to develop. analysis capabilities, an AI agent might be able to infer from a This paper addresses the issues above, building not only from customer’s tone that an unmentioned issue remains a problem a review of literature across marketing (and more generally, and provide real-time feedback to guide the (human) business), psychology, sociology, computer science, and ro- salesperson’s next approach. In this sense, AI could augment botics, but also from extensive interactions with practitioners. salespersons’ capabilities, but it also might trigger unintended Second, the preceding examples highlight mostly positive negative consequences, especially if customers feel uncom- consequences of AI, without detailing the widespread, justifi- fortable about AI monitoring conversations. Also, in the fu- able concerns associated with their use. Technologists such as ture, firms may primarily use AI bots, which—in some Elon Musk believe that AI is “dangerous” (Metz 2018). AI cases—function as well as human salespeople, to make initial might not deliver on all its promises, due to the challenges it contact with sales prospects. But the danger remains that if introduces related to data privacy, algorithmic biases, and customers discover that they are interacting with a bot, they ethics (Larson 2019). may become uncomfortable, triggering negative We argue that the marketing discipline should take a lead consequences. role in addressing these questions, because arguably it has the Third, the business model currently used by online retailers most to gain from AI. In an analysis of more than 400 AI use generally requires customers to place orders, after which the cases, across 19 industries and 9 business functions, online retailer ships the products (the shopping-then-shipping McKinsey & Co. indicates that the greatest potential value model—Agrawal et al. 2018;Gansetal. 2017). With AI, of AI pertains to domains related to marketing and sales online retailers may be able to predict what customers will (Chui et al. 2018), through impacts on marketing activities want; assuming that these predictions achieve high accuracy, such as next-best offers to customers (Davenport et al. retailers might transition to a shipping-then-shopping business 2011), programmatic buying of digital ads (Parekh 2018), model. That is, retailers will use AI to identify customers’ and predictive lead scoring (Harding 2017). The impact of preferences and ship items to customers without a formal or- AI varies by industry; the impact of AI on marketing is highest der, with customers having the option to return what they do in industries such as consumer packaged goods, retail, bank- notneed(Agrawaletal. 2018;Gans et al. 2017). This shift ing, and travel. These industries inherently involve frequent would transform retailers’ marketing strategies, business contact with large numbers of customers, and produce vast models, and customer behaviors (e.g., information search). amounts of customer transaction data and customer attribute Businesses like Birchbox, Stitch Fix and Trendy Butler al- data. Further, information from external sources, such as so- ready use AI to try to predict what their customers want, with cial media or reports by data brokers, can augment these data. varying levels of success. Thereafter, AI can be leveraged to analyze such data and de- The three use cases (above) illustrate why so many aca- liver personalized recommendations (relating to next product demics and practitioners anticipate that AI will change the to buy, optimal price etc.) in real time (Mehta et al. 2018). face of marketing strategies and customers’ behaviors. In fact, Yet marketing literature related to AI is relatively sparse, a survey by Salesforce shows that AI will be the technology prompting this effort to propose a framework that describes most adopted by marketers in the coming years (Columbus both where AI stands today and how it is likely to evolve. 2019). The necessary factors to allow AI to deliver on its Marketers plan to use AI in areas like segmentation and ana- promises may be in place already; it has been stated that “this lytics (related to marketing strategy) and messaging, person- very moment is the great inflection point of history” (Reese alization and predictive behaviors (linked to customer behav- 2018, p. 38). Yet this argument can be challenged. First, the iors) (Columbus 2019). Thus, we also propose an agenda for technological capability required to execute the preceding ex- future research, in which we delineate how AI may affect amples remains inadequate. By way of an exemplar, self- marketing strategies and customer behaviors. In so doing, driving cars are not ready for deployment (Lowy 2016), we respond to mounting calls that AI be studied not only by as—amongst other things—currently self-driving cars cannot those in computer science, but also studied by those who can handle bad weather conditions. Predictive analytics also need integrate and incorporate insights from psychology, econom- to improve substantially before retailers can adopt shipping- ics and other social sciences (Rahwan et al. 2019;also see then-shopping practices that avoid substantial product returns Burrows 2019). and the associated negative affect. Putting all this together, it appears that marketing managers and researchers need Introduction to artificial intelligence Miller (2016) outlines the difference between an AI bot and a chatbot. In Researchers propose that AI “refers to programs, algorithms, brief, chatbots rely on (relatively) simple algorithms, whereas AI bots have greater capabilities, incorporating complex algorithms and NLP. systems and machines that demonstrate intelligence” (Shankar 26 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 1 Select use cases (in the order in which they appear in the paper) Industry or Usage Context (specific firm or AI application) Description AI in driverless cars (e.g., Tesla) In the future, AI-enabled cars may allow for car journeys without any driver input, with the potential to significantly impact various industries (e.g., insurance, taxi services) and customer behaviors (e.g., whether they still buy cars). Online retailing AI (e.g., Birchbox) AI will enable better predictions for what customers want, which may cause firms to move away from a shopping-then-shipping business model and toward a shipping-then-shopping business model. Fashion-related AI (e.g., Stitch Fix) AI applications support stylists, who curate a set of clothing items for customers. Stitch Fix’s AI analyzes both numeric and image/other non-numeric data. Sales AI (e.g. Conversica) AI bots can automate parts of the sales process, augmenting the capabilities of existing sales teams. There may be backlash if customers know (upfront) that they are chatting with an AI bot (even if the AI bot is otherwise capable) Customer service robots (e.g., Rock’em Robots with task-automating AI respond to relatively simple customer and Sock’em; Pepper) service requests (e.g., making cocktails). Emotional support AI (e.g., Replika) AI aims to provide emotional support to customers by asking meaningful questions, offering social support, and adjusting to users’ linguistic syntax. In-car AI (e.g., Affectiva) In-car AI that analyzes driver data (e.g., facial expression) to evaluate drivers’ emotional and cognitive states. Customer screening AI (e.g. Kanetix) AI used to identify customers who should be provided incentives to buy insurance (and avoid those who (1) are already likely to buy and (2) those unlikely to buy). Business process AI (e.g., IBM Interact) AI used for multiple (simple) applications, such as customized offers (e.g., Bank of Montreal). Retail store AI (e.g., Café X, Lowebot, Robots that can serve as coffee baristas, respond to simple customer service 84.51, Bossa Nova) requests in Lowe’s stores, and identifying misshelved items in grocery stores. Security AI (e.g., Knightscope’s K5) Security robots patrol in offices or malls, equipped with superior sensing capabilities (e.g., thermal cameras). Spiritual support AI (e.g., BlessU-2; Xian’er) Customizable robot priest/monk offering blessings in different languages to the user. Companion robot AI (e.g., Harmony from Realbotix) Customizable robot companion, which promises reduced loneliness to the user. 2018,p. vi), is “manifested by machines that exhibit aspects of or “reading” documents to extract key provisions using natu- human intelligence” (Huang and Rust 2018,p. 155), andin- ral language processing. Second, AI can gain insights from volves machines mimicking “intelligent human behavior” vast volumes of customer and transaction data, involving not (Syam and Sharma 2018, p. 136). It relies on several key just numeric but also text, voice, image, and facial expression technologies, such as machine learning, natural language pro- data. Using AI-enabled analytics, firms then can predict what cessing, rule-based expert systems, neural networks, deep a customer is likely to buy, anticipate credit fraud before it learning, physical robots, and robotic process automation happens, or deploy targeted digital advertising in real time. (Davenport 2018). By employing these tools, AI provides a For example, stylists working at Stitch Fix, a clothing and means to “interpret external data correctly, learn from such styling service, use AI to identify which clothing styles will data, and exhibit flexible adaptation” (Kaplan and Haenlein best suit different customers. The underlying AI integrates 2019, p. 17). Another way to describe AI depends not on its data provided by customers’ expressed preferences, their underlying technology but rather its marketing and business Pinterest boards, handwritten notes, similar customers’ pref- applications, such as automating business processes, gaining erences, and general style trends. Finally, AI can engage cus- insights from data, or engaging customers and employees tomers, before and after the sale. The Conversica AI bot works (Davenport and Ronanki 2018). We build on this latter per- to move customer transactions along the marketing pipeline, spective. A listing of this research is provided in Table 2. and the AI bot used by 1–800-Flowers provides both sales and First, to automate business processes, AI algorithms per- customer service support. AI bots offer advantages beyond form well-defined tasks with little or no human intervention, just 24/7 availability. Not only do these AI bots have lower such as transferring data from email or call centers into error rates, but also they free up human agents to deal with recordkeeping systems (updating customer files), replacing more complex cases. Further, AI bot deployment can be lost ATM cards, implementing simple market transactions, scaled up or down as needed, when demand ebbs or flows. J. of the Acad. Mark. Sci. (2020) 48:24–42 27 As these descriptions suggest, AI offers the potential to AI capabilities forward, from task automation to context increase revenues and reduce costs. Revenues may increase awareness (e.g., Ghahramani 2015;Mnihetal. 2015). through improved marketing decisions (e.g., pricing, promo- Context awareness is a form of intelligence that requires ma- tions, product recommendations, enhanced customer engage- chines and algorithms to “learn how to learn” and extend ment); costs may decline due to the automation of simple beyond their initial programming by humans. Such AI appli- marketing tasks, customer service, and (structured) market cations can address complex, idiosyncratic tasks by applying transactions. Furthermore, the above discussions indicate that holistic thinking and context-specific responses (Huang and rather than replacing humans, firms generally are using AI to Rust 2018). However, such capabilities remain distant; a 2016 augment their human employees’ capabilities, such as when survey of AI researchers indicated there was only a 50% Stitch Fix uses AI to augment its stylists’ efforts to make chance of achieving context awareness (or its equivalent) by appropriate choices for clients (Gaudin 2016). This point 2050 (Müller and Bostrom 2016). Building on the above aligns well with sentiments expressed by Ginni Rometty, the point, Reese (2018, p. 61) cautions that such AI “does not CEO of IBM, who proposed that AI would not lead to a world currently exist… nor is there agreement … if it is possible.” of man “versus” machine but rather a world of man “plus” Nevertheless, this capability constitutes the goal of AI devel- machines (Carpenter 2015). opments, as predicted by compelling examples from science fiction, such as Jarvis from the Iron Man movies or Karen from Spider Man–Homecoming; both AI can understand A framework for understanding artificial new and complex contexts and create solutions therein. intelligence The differences between task automation and context awareness map onto concepts of narrow versus general AI Building on insights from marketing (and more generally (Baum et al. 2011; Kaplan and Haenlein 2019; Reese 2018). business), social sciences (e.g., psychology, sociology), and As Kaplan and Haenlein (2019) state, both narrow and general computer science/robotics, we propose a framework to help AI may equal or outperform human performance, but narrow customers and firms anticipate how AI is likely to evolve. We AI is focused on a specific domain and cannot learn to extend consider three AI-related dimensions: levels of intelligence, into new domains, whereas general AI can extend into new task type, and whether the AI is embedded in a robot. domains. It is important to clarify that although in this paper we Level of intelligence consider two levels of intelligence (task automation vs. con- text awareness), ideally levels of intelligence are best concep- Task automation versus context awareness Davenport and tualized as a continuum. Some AI applications have moved Kirby (2016) contrast task automation with context aware- beyond task automation but still fall well short of context ness. The former involves AI applications that are standard- awareness, such as Google’s DeepMind AlphaGo (which beat ized, or rule based, such that they require consistency and the the world’s best Go player), the AI poker player Libratus, and imposition of logic (Huang and Rust 2018). For example, Replika. These applications represent substantial advances, IBM’s Deep Blue applied standardized rules and “brute force” yet state-of-the-art AI still is closer to task automation algorithms to beat the best human chess player. Such AI is best (Davenport 2018). suited to contexts with clear rules and predictable outcomes, like chess. On the cruise ship Symphony of the Seas,two Overview of extant research Research into the psychology of robots, Rock ‘em and Sock ‘em, make cocktails for customers. automation (Longoni et al. 2019), examines how customers Elsewhere, the robot Pepper can provide frontline greetings, may respond to AI. Notwithstanding the fact that AI and IBM’s Watson can provide credit scoring and tax prepa- may be more accurate and/ or more reliable than ration assistance. Notwithstanding that these AI applications humans, customers have reservations about AI, and involve fairly structured contexts, many firms struggle to im- these reservations tend to increase as AI moves towards plement even these AI applications and rely on specialized context awareness. In turn, these increased reservations businesses like Infinia ML and Noodle, or consulting firms negatively impact the propensity to adopt AI, propensity like Accenture or Deloitte, to develop and set up initial AI to use AI, etc. A listing of such research is shown in Table 2. initiatives. Moving forward, we discuss (separately) issues relating to AI In contrast, context awareness continues to be developed, adoption and AI usage. and researchers in computer science are working on moving Reese (2018, p. 61) cautions that this type of AI is in no way “easy AI.” Replika (replica.ai) aims to serve as an AI friend, programmed to ask ques- To clarify, businesses like Infinia ML etc. also provide support moving tions about you and your life that are “meaningful,” and to offer emotional forward, when the firm initiates more advanced AI initiatives. support (French 2018;alsosee Hassler 2018). 28 J. of the Acad. Mark. Sci. (2020) 48:24–42 AI adoption Customers appear to hold AI to a higher standard know that messages are more effective when the perceived than is normatively appropriate (Gray 2017), as exemplified characteristics of the message source and the contents of the by the case of driverless cars. Customers should adopt AI if its actual message match, communication from AI should be use leads to significantly fewer accidents; instead, customers more effective when it highlights how rather than why in its impose higher standards and seek zero accidents from AI. messaging (regulatory construal fit; Lee et al. 2009;Motyka Understanding the roots of this excessive caution is important. et al. 2014). In line with the above, Kim and Duhachek (2018) A preliminary hypothesis suggests that customers trust AI showed that a message from an AI application is more persua- less, and so hold AI to a higher standard, because they believe sive when the message is about how to use a product, rather that AI cannot “feel” (Gray 2017). than why to use this product. This is because customers doubt Task characteristics also influence AI adoption. To the ex- whether AI can “understand” the importance of engaging in tent a task appears subjective, involving intuition or affect, certain consumption behaviors. customers likely are even less comfortable with AI (Castelo Next we pivot to factors that impact the propensity of cus- 2019). Research confirms that customers are less willing to tomers to engage with AI. Examining the case of medical use AI for tasks involving subjectivity, intuition, and affect, decision making, Longoni et al. (2019)show that customers’ because they perceive AI as lacking the affective capability or reservations are due to their concerns about uniqueness ne- empathy needed to perform such tasks (Castelo et al. 2018). glect (i.e., the AI is perceived as less able to identify and relate Tasks differ in their consequences; choosing a movie is with customers’ unique features). Further, building from prior relatively less consequential, but steering a car may involve work (Şimşek and Yalınçetin 2010; also see Haslam et al. more consequences. Using AI for consequential tasks is per- 2005), Longoni et al. (2019) show that these reservations are ceived as involving more risk, in turn reducing adoption in- more for customers who have higher scores on the ‘personal tentions. Early work has found support for this hypothesis, sense of uniqueness’ scale. In other work on how customers more so among more conservative consumers for whom risks engage with AI, Luo et al. (2019) examined how (potential) are more salient (Castelo et al. 2018; Castelo and Ward 2016). customers engage with AI bots. In reality, AI bots can be as Finally, customer characteristics may also impact AI adop- effective as trained salespersons, and 4x as effective as inex- tion. We build from two points: (1) when outcomes are con- perienced salespersons. However, if it is disclosed that the sequential, this increases perceptions of risk (Bettman 1973), customer is conversing with an AI bot, purchase rates drop and (2) women perceive more risk in general (Gustafsod by 75%. Linked to points made prior in this paper, because 1998) and take on less risk (Byrnes et al. 1999). Hence, early customers perceive the AI bot as less empathetic, they are curt work has found that women (vs. men) are less likely to adopt when interacting with AI bots, and so purchase less. AI, especially when outcomes are consequential (Castelo and Ward 2016). Moving beyond demographics, other factors also Task type impact the extent of AI adoption, e.g., to the extent a task is salient to a customer’s identity, the customer may be less like- Task type refers to whether the AI application analyzes num- ly to adopt AI (Castelo 2019). To elaborate, if a certain con- bers versus non-numeric data (e.g., text, voice, images, or sumption activity is central to a customer’s identity, then the facial expressions). These different data types all provide in- customer likes to take credit for consumption outcomes puts for decision making, but analyzing numbers is substan- (Leung et al. 2018). Some customers perceive that using AI tially easier than analyzing other data forms. Practitioners, for these consumption activities is tantamount to cheating, and such as senior managers from Infinia ML, formulate this cat- this hinders the attribution of credit post-consumption. egorization slightly differently, noting that data that can be Therefore, if an activity is central to a customer’s identity, then organized into tabular formats are significantly easier to ana- the customer may be less likely to adopt AI (for this activity). lyze than those data that cannot. In our discussions with em- ployees of Stitch Fix, we gained further clarity on this point. AI usage Moving past adoption issues, we note some usage Stitch Fix elicits data from customers using both direct ques- considerations, including how AI should communicate with tions about their preferences (which can be put in tabular customers. Customers do not associate AI applications with formats) and indirect elicitations from customers’ Pinterest autonomous goals (Kim and Duhachek 2018); for example, pages and likes. Stitch Fix uses proprietary AI algorithms to customers do not believe Google’s AlphaGo has the self- analyze the latter, non-numeric data and regards these data as driven goal to be a national Go champion. Rather, they believe very useful, because it has learned that customers cannot al- that this AI application is programmed to play the game Go. ways articulate their preferences on numeric scales. Consistent with this perception, customers are more likely to The distinction in the above paragraph is critical, because focus on “how” (rather than “why”) the AI application per- much data is non-tabular in form, and so being able to com- forms; implying that when engaging with AI, customers will prehend and analyze such data significantly enhances the im- be in a low level construal mindset. From extant research, we pact of AI. Many AI applications have started to analyze text, J. of the Acad. Mark. Sci. (2020) 48:24–42 29 voice, image, and face data inputs. These data inputs are ini- individuals without any protective barrier, travel with tially in non-numeric formats, but are often translated into individuals, etc.). numerical formats, e.g., pixel brightness values, relating to images. Applications that can process such data inputs in- Overview of extant research Prior research (Table 2) indicates clude, for example (1) IPSoft, which processes words spoken that using robots offer substantial advantages, especially in to customer agents to interpret what customers want (2) cases involving customer interactions. As prior work indi- Affectiva, which is working on in-car AI that can sense driver cates, customers form more personal bonds with robots than emotion and fatigue and switch control to an autonomous AI, with AI that lack any physical embodiment. For example, and (3) Cloverleaf’s shelfPoint, installed on retail store individuals enjoy interacting with a physically present robot shelves, which examines customers’ facial expressions to an- than with either a robot simulation (on a computer) or a robot alyze their emotional responses at the point of purchase. presented via teleconference (Wainer et al. 2006). Further, Although currently AI’s abilities to comprehend and analyze customers empathize with robots. When individuals are asked such non-numeric data formats remain somewhat limited, de- to administer pain—via electric shocks—to a (physically pres- veloping this ability will be critical for the full realization of ent) robot or a robot simulation, both of which go on to display the power of AI, and computer scientists are working towards marks indicating pain after being subjected to an shocked, improving AI capabilities in this regard (e.g., LeCun et al. individuals empathized more with the physically present robot 2015; You et al. 2016). (Kwak et al. 2013). Finally, customers interacted longer with a Separate to the above, it is worth pointing out that the robot diet coach than with either a virtually present diet coach ability to analyze unstructured data may be limited by legacy or a diet diary in a paper form (Kidd and Breazeal 2008). infrastructures. A senior manager in Infinia ML indicated that Other studies find that customers demonstrate reciprocity- often data is stored in formats and structures less amenable to based perceptions, e.g., they express more positive percep- AI deployment. Also, Kroger has an AI application that auto- tions of a care robot that asks for help and then returns this mates visual inspection of out-of-stock items on its grocery help by offering a favor (Lammer et al. 2014). In a prisoner’s shelves. In an interview with one of the authors of this paper, a dilemma experiment, participants exhibited similar reciprocity Kroger data scientist reported that the proper functioning of levels toward both robot partners and human partners Kroger’s AI application requires hardware upgrades; specifi- (Sandoval et al. 2016), and their reciprocity towards the robot cally, it needs to upgrade its cameras to higher resolution partner increased even more if the robot provided early signs levels if the AI application is to work properly. of cooperation (vs. random behavior). Noting the benefits of embedding AI in robots, work in robotics is examining how AI in robots best to improve not only the physical capability of robots but also the robot–AI interface (e.g., Adami 2015; Kober et al. Virtuality-reality continuum Most AI is virtual in form. For 2013; Steels and Brooks 2018). Further, to take advantage of example, Replika is available on smartphones, and Libratus the preference for physical embodiment, some vendors of vir- uses a digital platform. However, AI can also be embedded in tual agents (or bots) try to present these agents as having a a real entity or robot form, with some elements of physical physical form. IPsoft’s virtual agent, for example, is called embodiment. The extent to which a form is virtual versus Amelia and is often represented by a lifelike avatar image embodied reflects its position on the Milgram virtuality– and voice. reality continuum (Milgram et al. 1995). In this sense, re- However, other research shows that customers’ discomfort searchers and practitioners should conceive of virtual and real with AI is accentuated when the AI application is embedded in forms not as distinct categories but rather as endpoints on a a robot. As robots appear more humanlike, they become more continuum, within which AI entities are spread out. An AI like unnerving, in line with the uncanny valley hypothesis (UVH; Conversica is purely virtual, with no physical embodiment— Mori 1970). UVH arises because the appearance of robots although some companies that use virtual AI do give it names. “prompts attributions of mind. In particular, we suggest that In contrast, an AI application embedded in a robot barista machines become unnerving when people ascribe to them (e.g., Tipsy Robot in Las Vegas) appears somewhere on the experience (the capacity to feel and sense), rather than agency continuum between virtuality and reality, because it has some (the capacity to act and do)” (Gray and Wegner 2012,p. 125). physical embodiment; however, that embodiment can only Such factors may hinder AI adoption. operate in a narrow range and on a specific task (making a drink). Finally, the AI embedded in proposed multifunctional, Masahiro Mori wrote an influential paper arguing that making robots look companion robots (that today remain under development) more human is beneficial, but only up to a certain point, after which such robots elicit negative reactions. Thus, reactions become negative as robots would entail substantially more reality, featuring both physical move. From somewhat human to human-like. Thereafter, if robots look per- embodiment and the capacity to operate in wide range fectly human, reactions turn positive. The valley reflects these trends, as reac- of contexts (specifically, share physical proximity with tions initially becoming more negative, then turn positive. 30 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 2 Select extant research (in the order in which they appear in the paper) Paper Domain Dimension Takeaways Agrawal et al. (2018) BUS Artificial intelligence (AI) reduces the cost of prediction. Gans et al. (2017)BUS Rahwan et al. (2019) CS/R To best understand AI, bring in insights from not only computer science, but also other disciplines Shankar (2018)MKTG AI “refers to programs, algorithms, systems and machines Huang and Rust (2018)MKTG that demonstrate intelligence” (Shankar 2018,p. vi),is “manifested by machines that exhibit aspects of human intelligence” (Huang and Rust 2018, p. 155), involves machines mimicking “intelligent human behavior” (Syam and Sharma 2018, p. 136), and provides means to “interpret external data correctly, learn from such data, and exhibit flexible adaptation” (Kaplan and Haenlein 2019,p. 17). Syam and Sharma (2018)MKTG HuangandRust(2018) - Mechanical and analytical intelligences involve simple, rule-based tasks. Intuitive and empathetic intelligences involve complex tasks requiring empathy, holistic thinking and context-specific responses. Kaplan and Haenlein (2019) MKTG Kaplan and Haenlein (2019) – Used the terms narrow versus general AI. Narrow AI somewhat maps onto mechanical and analytical intelligences, whereas general AI maps onto intuitive and empathetic intelligences. Davenport and Ronanki (2018) BUS LVLINT Another way to describe AI is by stating its marketing and business outcomes, such as automating business processes, gaining insights from data, or engaging customers and employees Davenport and Kirby (2016) BUS LVLINT Contrasts task automation with context awareness. The former involves AI applications that are standardized, or rule based (akin to narrow AI). The latter is a form of intelligence that requires machines and algorithms to ‘learn how to learn’ and extend beyond their initial programming (akin to general AI). Ghahramani (2015) CS/R LVLINT How machines can learn from experience, using probabilistic machine learning. Mnih et al. (2015) CS/R LVLINT How artificial agents can learn to generalize from past experience to new situations, using reinforcement learning. Müller and Bostrom (2016) BUS LVLINT Artificial general intelligence (AGI) is a hypothetical technology that would be the equivalent of a human intelligence in terms of its flexibility and capability of performing and learning a vast range of tasks (similar to context awareness). In a survey of AI researchers, the median estimate was for a 50% chance of achieving an AGI by 2050 and a 90% chance of achieving one by 2075. Reese (2018) BUS LVLINT Defines narrow versus general AI and analytical AI versus humanized AI; Baum et al. (2011) SOC LVLINT both contrasts are very similar to the contrast between task automation versus context awareness. Reese (2018) cautions that AGI does not exist, and that there is no guarantee that it ever will. Davenport (2018) BUS LVLINT The state-of-the-art AI is closer to task automation than context awareness. Gray (2017) PSY LVLINT Customers appear to hold AI to a higher standard than is normatively appropriate. A preliminary hypothesis suggests that customers trust AI less, and so hold AI to a higher standard, because they believe that AI cannot “feel”. Castelo (2019) MKTG LVLINT To the extent a task appears subjective, involving intuition or affect, customers Builds from: Castelo et al. (2018)MKTG likely are less comfortable with AI (Castelo 2019). Customers are less willing to use AI for tasks involving subjectivity, intuition, and affect, because they perceive AI as lacking the affective capability or empathy needed to perform such tasks (Castelo et al. 2018). Castelo and Ward (2016) MKTG LVLINT Using AI for consequential tasks is perceived as involving more risk, in Builds from: Bettman (1973)MKTG turn reducing adoption intentions. This is more so amongst (1) conservative consumers, for whom risks are more salient, (2) women, who perceive Gustafsod (1998)PSY more risk in general, and take on less risk. Byrnes et al. (1999)PSY J. of the Acad. Mark. Sci. (2020) 48:24–42 31 Table 2 (continued) Paper Domain Dimension Takeaways Leung et al. (2018) MKTG LVLINT If a certain consumption activity is central to a customer’s identity, the customer would like to take credit for consumption outcomes. Some customers perceive that using AI for these consumption activities is tantamount to cheating, and this hinders the attribution of credit post-consumption. Hence if an activity is central to a customer’s identity, then the customer may be less likely to adopt AI for this activity. Kim and Duhachek (2018) MKTG LVLINT Customers do not associate AI applications with autonomous goals Builds from: Lee et al. (2009)MKTG (Kim and Duhachek 2018). In line with this perception, customers are more likely to focus on “how” (rather than “why”) the AI application Motyka et al. (2014)MKTG performs; implying that when engaging with AI, customers will be in a low level construal mindset. Because persuasion is more effective when the perceived characteristics of the persuasion source and the persuasion message match, communication from AI should be more effective when it highlights how rather than why in its messaging (regulatory construal fit; Lee et al. 2009; Motyka et al. 2014). AI persuasion messages are more effective in persuading consumers to buy the recommended product or services when the message highlights “how” to use the product rather than “why” to use the product. These effects are because customers doubt whether AI can understand “why” it is important for customers to engage in certain behaviors. Longoni et al. (2019) MKTG LVLINT Examining the case of medical decision making, Longoni et al. (2019) propose that customers’ reservations are due to their concerns about Builds from: Şimşek PSY uniqueness neglect (i.e., the AI is perceived as less able to identify and and Yalınçetin (2010) relate with customers’ unique features). Further, building from prior Haslam et al. (2005)PSY work (Şimşek and Yalınçetin 2010; also see Haslam et al. 2005), Longoni et al. (2019) propose that these reservations would be more for customers who havehigherscores onthe ‘personal sense of uniqueness’ scale (Şimşek and Yalınçetin 2010). Luo et al. (2019) MKTG LVLINT Examines how (potential) customers engage with AI bots. In reality, AI bots can be as effective as trained salespersons, and 4 times effective as inexperienced salespersons. However, if it is disclosed that the customer is conversing with an AI bot, purchase rates reduce by 75%. Because customers perceive the AI bot as less empathetic, they are curt when interacting with AI bots, and so purchase less. Ties into themes from Castelo et al. (2018). LeCun et al. (2015) CS/R TSKTYPE How deep learning has improved the state-of-the-art in speech and visual object recognition. You et al. (2016) CS/R TSKTYPE How using a new algorithm improves visual object recognition. Milgram et al. (1995) PSY ROBOT Proposes the virtuality-reality continuum. Wainer et al. (2006) CS/R ROBOT Interacting with a physical robot is perceived as more enjoyable than either interacting with a simulated robot on a computer or interacting with a real robot presented through teleconferencing. Kwak et al. (2013) CS/R ROBOT When asked to administer electric shocks to a (physical) robot or a simulated robot on a computer screen, individuals empathized more with the (physical) robot. Kidd and Breazeal (2008) CS/R ROBOT Interactions were longer with a robot diet coach than either a virtual diet coach or a pen-and-paper diet diary. Lammer et al. (2014) CS/R ROBOT Individuals express reciprocity towards robots. Adami (2015) CS/R ROBOT With suitable machine learning algorithms, robots can learn from past experiences. Kober et al. (2013) CS/R ROBOT Reinforcement learning can work for robots embedded with suitable machine learning algorithms. Mori (1970) PSY ROBOT Making robots look more human is beneficial, but only up to a certain point, after which such robots elicit negative reactions (UVH). Gray and Wegner (2012) PSY ROBOT Machines are perceived as more unnerving when individuals ascribe to machines the capacity to feel, rather than capacity to do. 32 J. of the Acad. Mark. Sci. (2020) 48:24–42 Table 2 (continued) Paper Domain Dimension Takeaways Mende et al. (2019) MKTG ROBOT Interactions with robots trigger discomfort (linked to UVH) and so further trigger compensatory behaviors. Boyd and Holton (2018) SOC ROBOT Will the combination of robotics and AI lead to an unprecedented social transformation? Pedersen et al. (2018) SOC ROBOTS Outlines the issues surrounding use of social robots in medical treatment, care facilities, and private homes. Also, outlines ethical concerns. André et al. (2018) MKTG LVLINT Because AI facilitates data-driven, micro-targeting marketing offerings, customers should view such offerings favorably, because it reduces search costs. Yet this could undermine customers’ perceived autonomy, with implications for their subsequent evaluations and choices. Aguirre et al. (2015) MKTG LVLINT Proposes the privacy–personalization paradox, whereby individuals balance privacy concerns against the benefits of personalized recommendations. Wang and Kosinski (2018) PSY TSKTYPE How to use deep neural networks to identify sexual orientation, merely by analyzing facial images MKTG Marketing; BUS Business; PSY Psychology; SOC Sociology; CS/ R Computer Science/ Robotics. Dimension: LVLINT levels of intelligence; TSKTYPE task type; ROBOTwhetherAIin robots Moving beyond AI adoption, we pivot to how customers that in the lower half of Fig. 1, we do not distinguish between interact with robots with embedded AI. Early research sug- numeric and non-numeric data, because context awareness– gests that interactions with AI-embedded robots trigger dis- capable AI likely will be able to handle any types of data. comfort (linked to the UVH) and so further trigger (negative) The first four use cases, associated with short to medium compensatory behaviors, like buying of status goods, or eating term developments, involve task automation (see Fig. 1). more food (Mende et al. 2019). From a theory perspective, this work not only shows the downsides of anthropomorphism Cell 1: Controller of numerical data The first cell in Fig. 1 (especially in the case of robots), but also the existence of reflects what AI can do very well, namely, statistical analyses compensatory consumption specifically linked to robots. of numeric data using machine learning. A typical use case is More broadly, sociologists ponder how AI (and specifically the application of AI to optimize prices (Antonio 2018). robots with embedded AI) might transform economy and so- Pricing strategies must balance two competing concerns; that ciety (Boyd and Holton 2018). For example, cloud-based the price is low enough to attract customers versus high technology facilitates deep learning in robots, which can learn enough to enable the firm to earn sufficient profits. Firms from human agents through repeated interactions. use AI to analyze vast amounts of numeric data (including Sociologists particularly note ways that robots may en- less intuitive predictor variables) to both set optimal prices ter multiple aspects of social life, not only in (expected) and then change prices in real time. For example, Kanetix areas such as service and transportation, but also in helps Canadian customers find deals on car insurance by domains like the arts and music. allowing prospective buyers to compare and evaluate policies and rates offered by more than 50 providers. Scott Emberley, the Business Development Director of integrate.ai, which The current state and likely evolution of AI partnered with Kanetix to build an AI application, indicated that the goal was identify three sets of customers (1) those Short- and medium-term time horizon highly likely to buy, (2) those very unlikely to buy, and (3) those in-between. Thereafter, Kanetix would direct their ad- In Fig. 1, we combine all the above considerations to depict vertising towards these “in-between” customers, which would the current state of AI and its likely evolution. The upper half provide the greatest returns, and not expend efforts on those of Fig. 1 (four cells) relates to task automation and thus the either very likely to buy or very unlikely to buy. With four likely state of AI in the short to medium time horizon. The years of data, integrate.ai developed a machine learning model lower half of Fig. 1 (two cells) relates to context awareness that could identify such customers. Five months later, Kanetix applications that are only likely in the long term (if at all), due estimated 2.3 times return on its AI investment, and a more to the constraints associated with the current state of AI. Note than 20% increase in sales among previously undecided cus- tomers. In another example, the Bank of Montreal (BMO) uses IBM Interact to analyze customer data across all its chan- Compensatory consumption is consumption “motivated by a desire to offset nels and identify personalized product offerings. If a customer or reduce a self-discrepancy” (Mandel et al. 2017,p. 134). J. of the Acad. Mark. Sci. (2020) 48:24–42 33 has been exploring mortgages on BMO’s site and later calls location exists. Instead, customers fill out style surveys, pro- the contact center, IBM Interact prioritizes the list of available vide their physical measurements, evaluate sample styles, cre- mortgage offers for the contact center service agent—in effect ate links to their Pinterest boards, and send in personal notes. augmenting agents’ capabilities and facilitating more relevant As may be expected, customers have trouble explicating their customer conversations. exact style preferences using words and numbers, but their pins and likes can be (better) indicators of their preferences. Cell 2: Controller of data Efforts to analyze non-numeric data Stitch Fix’s proprietary machine learning algorithms examine offer the potential to improve understanding of what cus- numbers, words, and Pinterest pins, then summarize the find- tomers want, and firms’ customer service. Some AI applica- ings for the company’s fashion stylists, who in turn select tions can analyze non-numeric data (in some cases, after con- suitable clothing to send to each customer. The above example version to numeric data), primarily using speech and image illustrates the need to suitably balance AI input and human recognition capabilities achieved with deep learning neural input; senior managers from Stitch Fix told us that—in their networks (Chui et al. 2018). For example, Conversica AI, as experience—their AI works best when it augments the manifested in a virtual AI assistant named Angie, sends out- (human) stylists’ capabilities. bound emails to up to 30,000 leads per month, then interprets Noting that the AI applications in companies like the responses to identify the most promising leads (Power Conversica and Stitch Fix use all types of data (i.e., use nu- 2017). Angie engages in initial conversation with the pros- meric data and non-numeric data), we term the AI applications pect, and then routes to most promising leads to a (human) in this cell as reflecting “Controller of Data.” salesperson. In effect, Conversica’s AI augments salespersons’ capabilities. In a pilot test with a telecommuni- Cell 3: Numerical data robot This cell is similar to cell 1, cations company called Century Link, Angie appropriately except that it incorporates AI embedded in a robotic form, understood more than 95% of emails received (and sent the and so these AI applications can best be described as robots rest to human agents for interpretation), and Century Link that process numerical data inputs. Such robots are well suited earned a 20-fold return on its investments in Angie. to retail environments with well-structured operations. At The Stitch Fix’s business model offers another example. As Café X, for example, a robot barista can serve up to 120 we noted, Stitch Fix delivers apparel directly to customers coffees per hour (Hochman 2018). Each robotic barista fea- (Wilson et al. 2016), without requiring the customers to actu- tures a $25,000, six-axis animatronic arm. Customers place ally engage in a formal shopping task. No Stitch Fix retail orders on a kiosk touchscreen (or via an app), so all inputs Fig. 1 AI framework Task Digital form Robot form automation Analyze 1 – Controller of 3 – Numerical Data Robot technologies, numbers Numerical Data deployed currently or to Business Use Case Business Use Case be deployed in Kanetix Café X the short to IBM Tipsy Robot medium term Analyze 2 – Controller of Data 4 – Data Robot text, voice, Business Use Case Business Use Case faces, Conversica Lowebot images Stitch Fix 84.51/ Kroger Replika Walmart/ Bossa Nova K5 from Knightscape Context Digital form Robot form awareness Analyze 5 – Data Virtuoso 6 – Robot Expert technologies numbers, that may be text, Example Use Case Example Use Case deployed in the voice, Jarvis Dorian long term faces, image 34 J. of the Acad. Mark. Sci. (2020) 48:24–42 are numeric. As in a regular coffeehouse, customers can select human input (Lashinsky 2019). Put simply, the dream of get- various options: latte or espresso, with different amounts of ting into a driverless car outside in one city, falling asleep, and froth, and various ingredients such as organic Swedish oat waking up in another city is not reality and may not be achieved milk. The goal is not to replace baristas, but rather to augment anytime soon. Even the less consequential forms of AI remain baristas’ capabilities by taking over more routine operations. problematic. Google’s AlphaGo Zero might have successfully The Cafe X robot barista augments the capabilities of the learned the complex game of Go in a short period, using adver- human barista, who can then focus on providing high- sarial networks that pit two (competing) AI systems against quality customer service, and also facilitating what the com- each other so that they can learn; yet in this case, the outcome pany calls “coffee education” (e.g., managing tastings). space was very well defined. Furthermore, all these AI systems received significant training data. In contrast, the outcome Cell 4: Data robot This cell is similar to cell 2, except that the spaces (i.e., business domains) for most likely AI applications robotic form can process all types of data (not just numeric are poorly defined, and relevant training data is hard to obtain. data). For example, the Lowebot at Lowe’sHome These points reiterate the challenges of moving from task au- Improvement stores (Hullinger 2016) can scan a product held tomation to context awareness. As such, the use cases we pres- up by a customer (or listen to the customer speak the name of ent for the last two cells are hypothetical, and this section is the desired product), confirm whether the item is in stock, and deliberately brief, reflecting that our discussion is more aspira- then roll along with the customer to the exact spot in the store tional than descriptive of any near-term reality. where he or she can find the product. This task requires com- prehension and examination of both numeric and non-numeric Cell 5: Data virtuoso Advanced AI could be embedded in a data, as well as an indoor navigation capability, which repre- digital form, as exemplified by the AI Jarvis in Iron Man sents a significant advance over the capabilities embodied in movies. Jarvis has advanced data capabilities that can examine the Café X robot. Using the Lowebot augments the capabili- multiple data types. Perhaps most notably, Jarvis adapts to ties of Lowe’s human sales associates, allowing focus on more new contexts, beyond those for which it has been trained, such complex customer service requests. as when it hides from the more advanced AI Ultron and finds Other retailers have similar applications. Our discussions with ways to thwart Ultron’s hacking attempts. Futurists would senior managers at 84.51 indicate that they are working with have us believe that such AI will emerge in the long term, Kroger to implement in-store robots that can identify misshelved with strong predictive abilities for customers’ preferences or out-of-stock items. In another example, Walmart has partnered and high capability levels for managing customer service. with Bossa Nova Robotics to deploy robots in its stores to scan Thus, the term virtuoso seems appropriate for such AI. shelves. The goal appears to be to get robots to perform tasks that repeat and are predictable, enabling (human) associates to focus Cell 6: Robot experts An advanced AI also could be embedded on serving customers (Avalos 2018). in a robot form, such as the AI Dorian from the television Finally, security robots, such as the K5 from Knightscope, show Almost Human. Dorian’s advanced capabilities include roam offices and malls at night. These robots have better sens- facial recognition, bio scans, analyses of non-numeric stimuli ing capabilities than humans, because they incorporate ther- such as DNA, speed-reading, speaking multiple languages, mal cameras and other high-technology sensing tools. Here and taking the temperature of fluids using his finger. Like again, the objective is to augment human security guards’ Jarvis, Dorian can adapt to a variety of new contexts. capabilities (Robinson 2017). Futurists predict that such robot experts will emerge in the long term to serve as companions that meet various customer Long-term time horizon needs (e.g., in-home service, home security, medical support). Such robots even might be able to bond emotionally with For completeness, we also examine what might happen when (human) customers, and potentially replace human partners AI applications incorporate context awareness, as summarized and animal partners. in the two cells in the lower half of Fig. 1. We reiterate that there is no indication that such developments will occur in the short or medium term, as exemplified by the case of driverless Agenda for future research cars. Tesla has removed any “self-driving” labels from its website, noting that these labels were causing confusion Having described AI and presented a framework to better (Hawkins 2019). The CEO of Waymo admits that driverless understand it, we pivot to outlining some important areas for cars are unable to drive in poor weather conditions without future research. These include how firms may need to change their marketing strategy, how customers’ behaviors will be impacted, and issues relevant to policymakers. We outline This consulting firm is a subsidiary of Kroger and provides retail insights to Kroger and its partners; it has strong analytics and AI capabilities. these areas in Fig. 2, linking these to the cells in Fig. 1. J. of the Acad. Mark. Sci. (2020) 48:24–42 35 AI and marketing strategy how best to make predictions for RNPs, research can also examine how best to combine AI-driven insights with human Predictive ability Because AI can help firms predict what cus- judgment. tomers will buy, using AI should lead to substantial improve- AI is expected to play an important role in predicting not ments in predictive ability. Contingent on levels of predictive only what customers want to buy, but also what price to accuracy, firms may even substantially change their business charge, and whether price promotions should be offered models, providing goods and services to customers on an on- (Shankar 2018). Price and price promotions are important going basis based on data and predictions about their needs. drivers of sales (Biswas et al. 2013;Guha etal. 2018), and Multiple research opportunities thus emerge, related to differ- so are an important area of research for marketing researchers. ent customer purchase behaviors and marketing strategies. Thus, an important area for future research relates to how AI One especially important research area may relate to how well can be best used to predict what prices are optimal and wheth- prediction AI–driven algorithms may extend to forecasting er or not price promotions should be offered. demand for really new products (RNPs; described in Zhao Another important research avenue pertains to allocations et al. 2012). AI algorithms probably have good predictive of advertising resources. Much advertising focuses on devel- ability for incrementally new products; the open question is oping customer awareness and driving customers’ informa- whether they will have good predictive ability for RNPs. For tion search. Would these advertising dollars be required AI algorithms to do so would presumably require data on in the future, wherein firms may be able to better pre- RNPs that would be used in training machine learning models; dict customers’ preferences, and thus would not need to this is often not readily available. Further, when examining advertise as much? Fig. 2 Research agenda for AI. Task Digital form Robot form Notes: As noted in the text, the automation Analyze 1 – Controller of Numerical 3 – Numerical Data numbers Data Robot sales AI application will be more effective if it can process both Research Agenda Research Agenda numeric and non-numeric data, Predictive ability (MS) similar to Controller of and hence is more related to the AI adoption (CB) Numerical Data cell Controller of Data cell. This is - negative response to AI + more likely for more advanced - state and trait moderators Affective responses to robots, and so more likely to be AI usage (CB) robots (CB) relevant to robots able to handle - primed mindset non-numeric data (notably voice), - Post AI issues (CB) and hence more related to perhaps - perceived loss of the Data Robot cell, but more so autonomy to the Robot Expert cell - state and trait moderators Data privacy (P) Bias (P) Ethics (P) Analyze 2 – Controller of Data 4 – Data Robot text, voice, Research Agenda Research Agenda faces, similar to Controller of similar to Controller of images Numerical Data cell Data cell + + Sales (MS) UVH (CB) AI adoption for spiritual well- Loss of human being (CB) connectedness (CB) Context Digital form Robot form awareness Analyze 5 – Data Virtuoso 6 – Robot Expert numbers, text, Research Agenda Research Agenda voice, similar to Controller of Data similar to Data Robot faces, cell cell image Notes: MS = marketing strategy; CB = consumer behavior; P = public policy. 36 J. of the Acad. Mark. Sci. (2020) 48:24–42 Sales and AI As we discussed with regard to Conversica, AI Clothing suggestions from stylists who “swiped” on the app may alter all stages of the sales process, from prospecting to similarly to particular customers elicited more positive re- pre-approach to presentation to follow-up (Singh et al. 2019; sponses from the customers (i.e., both qualitative feedback Syam and Sharma 2018). Thus, a wide variety of research about the stylist and increased sales of clothes curated by that questions arise: stylist). When implementing AI, firms thus may achieve better outcomes if they let their data scientists spend some amount of & Can AI analyze customer communication and other cus- time on unauthorized “pet projects,” a research and develop- tomer information (e.g., social media posts) in ways to ment practice already in place in firms like 3 M (Shum and Lin devise future communications that are more persuasive 2007). Researching the best way to implement AI, to take or increase engagement? advantage of both expected and unexpected benefits, is a fruit- & Can AI provide real-time feedback to salespeople to help ful area for research. them improve their sales pitches, based on assessments of customers’ verbal and facial responses? Modeling the evolution of AI Finally, firms need to develop & How might AI combine text and other communication realistic expectations, because “in the short run, AI will pro- inputs (e.g., voice data), actual customer behavior, and vide evolutionary benefits; in the long run, it is likely to be other information (e.g., behaviors of similar customers) revolutionary” (Davenport 2018, p. 7). That is, the benefits of to predict repurchases? This effort demands non-numeric AI could be overestimated in the short term but data, in line with cells 2, 4, 5 and 6. underestimated in the long term, a point (sometimes called & Considering Luo et al.’s(2019) findings, how should Amara’s Law) in accordance with Gartner’shype cycle model firms deploy AI sales bots effectively? of how new technologies evolve (Dedehayir and Steinert 2016; also see van Lente et al. 2013;Shankar 2018). This view Answering these questions could help firms design sales to is popular among practitioners, according to our personal dis- take the most advantage of AI. cussions and interviews with various senior managers. Will In addition, firms need to consider how they (re)organize the evolution of AI reflect this model, or will its evolution their sales and innovation processes. These points are not differ and more closely map onto models that also integrate listed in Fig. 2, as they do not tie neatly into the cells shown more traditional innovation models (e.g., Roger’smodel, the in Fig. 1. Bass model)? Research that tests which innovation model best predicts AI evolutions will be useful. Sales process In the presence of AI, how should sales be or- ganized and what skills will salespeople need? First, how best AI and customer behavior to structure the sales organization wherein organizational components include both AI bots and human salespeople. New technologies often alter customer behavior (e.g., Secondly, how should the firm manage the tradeoff between Giebelhausen et al. 2014; Groom et al. 2011; Hoffman and AI focusing on customers’ expressed needs versus salespeople Novak 2018;Moon 2003), and we expect that AI will do so as being relatively better able to manage issues like customer well. We propose three research topics, related to AI adoption, stewardship. Lastly, will salespeople be able to be trained/ to AI usage, and post-adoption issues. be able to manage customers’ concerns relating to AI, specif- ically issues related to data privacy and ethics. It is clear that AI adoption As a general point, due to a wide variety of fac- sales processes will require innovation related not only to AI tors, customers view AI negatively, which is a barrier to adop- technologies, but also in job design and skills (Barro and tion. As noted, these negative views often stem from cus- Davenport 2019). tomers’ sense that AI is unable to feel (Castelo et al. 2018; Gray 2017) orthatAIisrelativelyless ableto identify whatis AI innovation process Because the impact of AI is uncertain, unique about each customer (Longoni et al. 2019). Also Luo firms need to figure out how best to (continually) develop AI. et al. (2019) suggest that customers perceive AI bots as being In our discussions with senior managers at Stitch Fix, they less empathetic. Customers also are less likely to adopt AI in indicated that the company encourages its data scientists to consequential domains (Castelo et al. 2018; Castelo and Ward pursue projects on their own (Colson 2018), such that they 2016) and for tasks salient to their identity (Castelo 2019; continually engage in preliminary testing of new project ideas. Leung et al. 2018). One Stitch Fix data scientist created a Tinder-like app called Thus, an important area for future research, important from Style Shuffle, to allow users to indicate preferences for various the standpoint of both research and practice, would be to ex- clothing styles. This app not only informed stylists about cus- amine how best to mitigate the impact of the above. Initial tomers’ preferences (the expected benefit) but also helped brainstorming with fellow researchers and with practitioners match stylists with specific customers (an unexpected benefit). suggests that positioning AI as a learning (artificial) organism, J. of the Acad. Mark. Sci. (2020) 48:24–42 37 or else positioning the AI application as one that combines AI When AI is embedded in robots, the robots likely have and human inputs (as in Stitch Fix), may help partially miti- important roles in customers’ lives, functioning as frontline gate the impact of the points above. Longoni et al. (2019) service providers (Wirtz et al. 2018), companions, nannies, propose that offering customers the opportunity to slightly or pet replacements. In addition to the UVH-related challenges modify the AI may get these customers to look past unique- documented previously, some research results suggest ness neglect, and focus more on the benefits of personaliza- that interactions with AI-embedded robots trigger dis- tion. This too may be a way to mitigate the points raised prior. comfort and compensatory behaviors (Mende et al. 2019). It The discomfort with AI is accentuated in case the AI ap- is important to determine when customers perceive AI- plication is embedded in a robot. As robots become more embedded robots negatively and whether these perceptions humanlike, then due to the UVH, customers find these robots may improve over time. unnerving. Such factors may hinder AI adoption and deserve Finally, if customers’ ideal preferences actually differ from study. An interesting moderator of this effect—worth investi- their past behaviors (e.g., customers trying to stop eat- gating—may be whether the AI form is perceived by cus- ing unhealthy foods), AI might make it harder for them tomers as a servant or partner; UVH effects may be stronger to find and move toward their preferred options, by if AI achieves partner status. Also deserving of study is other only presenting them with choices reflecting their past ways of mitigating such effects. Early efforts in this direction behaviors. The widespread “retargeting” of digital ads is one involve trying to prime empathy, by convincing customers example of this phenomenon. How to train AI to best manage that robots have some ability to see things from the customers’ this issue? viewpoint, and (also) have some ability to feel sympathy for the customer if the customer were suffering (Castelo 2019). Post-adoption The downstream consequences of AI adoption Other possible methods could relate to anthropomorphizing also suggest some relevant research topics. In particular, cus- the AI, as this may persuade customers that the AI has some- tomers might perceive a loss of autonomy if AI can substan- what more empathy (this point needs to balance with concerns tially predict their preferences. In theory, because AI facilitates about the UVH). data-driven, micro-targeting marketing offerings (e.g., Gans Sociologists appear especially interested in how robots et al. 2017; Luo et al. 2019), customers should view offerings with embedded AI might make inroads into society (Boyd more favorably, because it reduces their search costs. Yet it and Holton 2018, p. 338), noting that “complexities arise also could undermine customers’ perceived autonomy, with when cultural preferences associated with human as opposed implications for their evaluations and choices (André et al. to machine delivery of personal services are considered. Do 2018). If customers learn that an AI algorithm can predict their … consumers find social robots acceptable?” Broadly speak- preferred choices, they may deliberately choose a non- ing, research can address how attitudes toward robots vary by preferred option, to reaffirm their autonomy (André et al. culture (Li et al. 2010). Beyond concerns associated with cul- 2018; Schrift et al. 2017). Such considerations evoke a variety ture, it may be pertinent to examine which other trait factors of research questions. For example, which factors determine determine whether customers are willing to have their hair whether (and how much) customers value perceived autono- styled by robots or accept childcare/elderly care services de- my in AI-mediated choice settings? In this regard, it may be livered by robots (Pedersen et al. 2018). In addition to physical helpful to examine individual difference variables, such as well-being considerations, some sociologists suggest that ro- culture and whether customers regard AI as a servant or part- bots may assist with spiritual well-being (Fleming 2019), as ner. Research also might address state factors, such as the exemplified by the robot priest BlessU-2 (Sherwood 2017) product type; perceived autonomy may be less relevant for and Buddhist monk Xian’er (Andrews 2016). Understanding utilitarian product choices than for hedonic ones, because of how robots with embedded AI can assist in various ways, differential links to customers’ identity. beyond improving customers’ physical well-being, is a good Also, there is a generalized fear of a loss of human con- area for research. nectedness, if humans form bonds with robots with embedded AI. The popular press (e.g., Marr 2019) stokes concerns about AI usage When customers interact with an AI application, it robots with embedded AI becoming popular (over humans) as might prime a low-level construal mindset (Kim and partners. Robots like Harmony (by Realbotix) appear promis- Duhachek 2018). Research should determine what other ing in this regard, able to take on different personae and dis- mindsets might be primed by AI, e.g., AI may prime preven- play some expression. But such robots arguably could be tion focus among customers for whom AI is a relatively new damaging to society at large, by increasing social isolation, technology. Related insights would have implications for how reducing the incidence of marriage, or reducing birthrates— the AI application should communicate with the customer, which is critical for countries like Japan, where birthrates are because communication exerts stronger impacts when it fits already low. This point suggests some interesting research with (a primed) mindset. opportunities. 38 J. of the Acad. Mark. Sci. (2020) 48:24–42 AI and policy issues choices. Does the trade-off depend on the product category or the level of the customer’s trust in the firm, for example? Finally, AI is of interest to policymakers. We note three broad Also, how would this trade-off shift over time? areas in which policymakers seek to ensure that firms strike a suitable balance between their own commercial interests and Bias The potential algorithmic bias embedded in AI applica- the interests of customers: data privacy, bias, and ethics. All tions could stem from multiple causes (Villasenor 2019), in- three of these areas generally align with all cells in Fig. 2. cluding the data sets that inform AI. For example, Amazon abandoned a tool that used AI intelligence to rate job appli- Data privacy Today, the combination of AI and big data im- cants, in part because it discriminated against female appli- plies that firms know much about their customers (Wilson cants (Weissman 2018). This bias emerged because the train- 2018). Hence, two issues deserve research attention. First, ing data sets used to develop the algorithm were based on data customers worry about the privacy of their data (Martin and relating to previous applicants, who were predominantly men. Murphy 2017; Martin et al. 2017). Privacy is complicated Exacerbating the issue, many AI algorithms are opaque black (Tucker 2018), for three reasons: (1) the low cost of storage boxes, so it is difficult to isolate which exact factors these implies that data may exist substantially longer than was algorithms consider. Testing for whether there is bias in AI intended, (2) data may be repackaged and reused for rationales applications is an important topic. different than those intended, and (3) data for a certain indi- In addition, AI may not be able to distinguish attributes that vidual may contain information about other individuals. could induce potential bias. Villasenor (2019)argues asfollows. Policy related to data privacy requires balancing two compet- In general, it may not be offensive when insurance companies ing priorities. Too little protection means that customers may treat men and women differently, with one set of premiums for not adopt AI-related applications; too much regulation may male drivers and another set of premiums for female drivers. strangle innovation. Does this mean that it is acceptable for AI to calculate auto Second, important research questions pertain to whether insurance premiums based on religion? Many would argue that data privacy management efforts should be driven by legal basing auto insurance premiums on religion is offensive, but regulations or self-regulation, in that “it is not clear yet if from the point of view of an AI algorithm designed to “slice- market driven incentives will be sufficient for firms to adopt and-dice” data in every way possible, the distinction between policies that favor consumers or whether regulatory oversight using gender versus religion, as a basis to determining auto is required to ensure a fair outcome for consumers” (Verhoef insurance rates, may not be obvious. All this suggests that issues et al. 2017, p. 7). Cultural perspectives on data privacy also relating to bias remain a non-trivial problem (Knight 2017). vary, which is an important consideration; some have sug- gested that the lack of data privacy in China, for example, is Ethics Finally, AI developers must grapple with ethics; we consistent with Confucian cultural ideals (Smith 2019). highlight two issues. First, data privacy choices may reflect Third, we need insights into how best to acknowledge and afirm’s strategy (e.g., if it wants to be perceived as a trusted address privacy concerns at the moment data are collected, as firm; Martin and Murphy 2017; also see Goldfarb and Tucker well as how to manage data privacy failures (e.g., data 2013) but also could be driven by ethical concerns. In this breaches). Amazon already sells doorbells with cameras (the sense, research should address “how can normative ethical Ring device) and may be planning to add facial identification theory pave the way for what organizations should be doing AI to the devices (Fowler 2019). Customers thus may become to exceed consumer privacy expectations, as well as to over- concerned if Amazon has access to data recorded through comply with legal mandates in order to preserve their ability to Ring, which it could use or sell. Neighbors also might protest self-regulate” (Martin and Murphy 2017, p. 152). A related if Ring cameras record their front yard activities without their research topic might involve examining how ethical concerns permission. Also, the data from Ring arguably could be about AI vary across cultures. subpoenaed by law enforcement agencies or obtained illegally Second, firms choose to deploy AI by defining which prob- by hackers. Such issues reflect topics for further research. lems the AI will tackle. For example, two Stanford researchers Finally, we consider the privacy–personalization paradox used deep neural networks to identify people’ssexualorien- (Aguirre et al. 2015). Customers must balance privacy con- tation, merely by analyzing facial images (Wang and Kosinski cerns against the benefits of personalized recommendations 2018). The deep neural network tools (vs. human judges) were and offers. Important questions relate to how customers deter- better able to differentiate between gay and straight men. mine the optimal trade-off, including which individual differ- However, the work raised ethical concerns, in that many ar- ence variables and state variables might moderate their gued that this AI-based technology may be used by spouses on their partners (if they suspected their partners were closet- ed), or—more frighteningly—may be used by certain govern- Firms are aware of this, and are taking steps to suitably respond (Deloitte Insights, as reported in Schatsky et al. 2019) ments to “out” and then prosecute certain populations (Levin J. of the Acad. Mark. Sci. (2020) 48:24–42 39 collection and trust-building strategies on online advertisement ef- 2017). An important topic for research thus is to address up- fectiveness. JournalofRetailing, 91(1), 34–49. front the types of applications for which AI should be used for André, Q., Carmon, Z., Wertenbroch, K., Crum, A., Frank, D., Goldstein, (or, should not be used for). W., et al. (2018). Consumer choice and autonomy in the age of artificial intelligence and big data. Customer Needs and Solutions, 5(1–2), 28–37. Andrews, T. (2016). Meet the robot monk spreading the teachings of Conclusion Buddhism around China. Washington Post, April 27. Retrieved February 11, 2019 from https://www.washingtonpost.com/news/ This paper outlines a framework to understand how AI will morning-mix/wp/2016/04/27/meet-the-robot-monk-spreading-the- teachings-of-buddhism-around-china/?utm_term=.fed52d90bff3. impact the future of marketing, specifically to outline how AI Accessed 11 Feb 2019 may influence marketing strategies and customer behaviors. Antonio, V. (2018). How AI is changing sales. Harvard Business Review, We build on prior work, as well as build from extensive inter- July 30. Retrieved February 11, 2019 from https://hbr.org/2018/07/ actions with practitioners. First, we develop a multidimension- how-ai-is-changing-sales. al framework for the evolution of AI, noting the importance of Avalos, G. (2018). Walmart tests shelf-scanning robots in Bay Area. The Mercury News, March 20. Retrieved February 11, 2019 from https:// dimensions pertaining to intelligence levels, task types, and www.mercurynews.com/2018/03/20/walmart-tests-shelf-scanning- whether the AI is embedded in a physical robot. In so doing, robots-bay-area/. we provide the first attempt to integrate all three dimensions in Barro, S., & Davenport, T. H. (2019). People and machines: Partners in a single framework. We also make two (cautionary) points. innovation. MIT Sloan Management Review, 60(4), 22–28. Baum, S. D., Goertzel, B., & Goertzel, T. G. (2011). How long until First, the short to medium term impacts of AI may be more human-level AI? Results from an expert assessment. limited than the popular press would suggest. Second, we Technological Forecasting and Social Change, 78(1), 185–195. suggest that AI will be more effective if it is deployed in ways Bettman, J. (1973). Perceived risk and its components: A model and that augment (rather than replace) human managers. empirical test. JournalofMarketing, 10(2), 184–190. To examine the full scope of the impact of AI, we propose a Biswas, A., Bhowmick, S., Guha, A., & Grewal, D. (2013). Consumer evaluations of sale prices: Role of the subtraction principle. Journal research agenda covering three broad areas: (1) how firms’ of Marketing, 77(4), 49–66. marketing strategies will change, (2) how customers’ behav- Boyd, R., & Holton, R. J. (2018). Technology, innovation, employment iors will change, and (3) issues related to data privacy, bias, and power: Does robotics and artificial intelligence really mean and ethics. This research agenda warrants consideration social transformation? Journal of Sociology, 54(3), 331–345. Burrows, L. (2019). The science of the artificial. Retrieved June 12, 2019 by academia, firms, and policy experts, with the recog- from https://www.seas.harvard.edu/news/2019/05/science-of- nition that although AI already has had some impact on artificial. marketing, it will exert substantially more impact in the Byrnes, J. P., Miller, D. C., & Schafer, W. D. (1999). Gender differences future, and so there is much still to learn. We hope that in risk taking: A meta-analysis. Psychological Bulletin, 125(3), 367. this research agenda motivates and guides continued re- Carpenter, J. (2015). IBM’s Virginia Rometty tells NU grads: Technology will enhance us. Retrieved February 11, 2019 from https://www. search into AI. chicagotribune.com/bluesky/originals/ct-northwestern-virginia- rometty-ibm-bsi-20150619-story.html. Acknowledgements The authors thank Eric Colson (Stitch Fix), James Castelo, N. (2019). Blurring the line between human and machine: Kotecki (Infinia ML) and Charu Kohli (CognitiveScale), who shared their Marketing artificial intelligence (doctoral dissertation). Retrieved time, materials and insights with the authors as this paper developed. from Columbia University Academic Commons. https://doi.org/ Thomas Davenport and Dhruv Grewal dedicate this article to the memory 10.7916/d8-k7vk-0s40. of their dear friend and colleague, Dean Bala Iyer of Babson University. Castelo, N., & Ward, A. (2016). Political affiliation moderates attitudes towards artificial intelligence. In P. Moreau & S. Puntoni (Eds.), NA Open Access This article is distributed under the terms of the Creative - advances in consumer research (pp. 723–723). Duluth, MN: Commons Attribution 4.0 International License (http:// Association for Consumer Research. creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Castelo, N., Bos, M., & Lehman, D. (2018). Consumer adoption of al- distribution, and reproduction in any medium, provided you give appro- gorithms that blur the line between human and machine. Graduate priate credit to the original author(s) and the source, provide a link to the School of Business: Columbia University Working Paper. Creative Commons license, and indicate if changes were made. Chui, M., Manyika, J., Miremadi, M., Henke, N., Chung, R., Nel, P., & Malhotra, S. (2018). Notes from the AI frontier: Applications and value of deep learning. McKinsey global institute discussion paper, References April 2018. Retrieved June 12, 2019 from https://www.mckinsey. com/featured-insights/artificial-intelligence/notes-from-the-ai- frontier-applications-and-value-of-deep-learning. Adami, C. (2015). Artificial intelligence: Robots with instincts. Nature, Colson, E. (2018). Curiosity-driven data science. Harvard Business 521(7553), 426–427. Review, November 27. Retrieved February 11, 2019 from https:// Agrawal, A., Gans, J. S., & Goldfarb, A. (2018). Prediction machines: hbr.org/2018/11/curiosity-driven-data-science The simple economics of artificial intelligence. Harvard Business Columbus, L. (2019). 10 charts that will change your perspective of AI in School Press. marketing. Forbes, July 07. Retrieved on July 09, 2019 from https:// Aguirre, E., Mahr, D., Grewal, D., de Ruyter, K., & Wetzels, M. (2015). www.forbes.com/sites/louiscolumbus/2019/07/07/10-charts-that- Unraveling the personalization paradox: The effect of information will-change-your-perspective-of-ai-in-marketing/amp/ 40 J. of the Acad. Mark. Sci. (2020) 48:24–42 Davenport, T. H. (2018). The AI advantage: How to put the artificial Hawkins, A. (2019). No, Elon, the navigate on autopilot feature is not intelligence revolution to work. MIT Press. ‘full self driving’. Retrieved February 11, 2019 from https://www. theverge.com/2019/1/30/18204427/tesla-autopilot-elon-musk-full- Davenport, T. H., & Kirby, J. (2016). Just how smart are smart machines? self-driving-confusion. MIT Sloan Management Review, 57(3), 21–25. Hayes, A. (2015). The unintended consequences of self-driving cars. Davenport, T. H., & Ronanki, R. (2018). Artificial intelligence for the real Retrieved February 11, 2019 from https://www.investopedia.com/ world. Harvard Business Review, 96(1), 108–116. articles/investing/090215/unintended-consequences-selfdriving- Davenport, T. H., Dalle Mule, L., & Lucker, J. (2011). Know what your cars.asp. customers want before they do. Harvard Business Review, 89(12), Hochman, D. (2018). This $25,000 robotic arm wants to put your 84–92. Starbucks barista out of business. Retrieved February 11, 2019 from Dedehayir, O., & Steinert, M. (2016). The hype cycle model: A review https://www.cnbc.com/2018/05/08/this-25000-robot-wants-to-put- and future directions. Technological Forecasting and Social your-starbucks-barista-out-of-business.html. Change, 108,28–41. Hoffman, D. L., & Novak, T. P. (2018). Consumer and object experience Fleming, P. (2019). Robots and organization studies: Why robots might in the internet of things: An assemblage theory approach. Journal of not want to steal your job. Organization Studies, 40(1), 23–38. Consumer Research, 44(6), 1178–1204. Fowler, G. (2019). The doorbells have eyes: The privacy battles brewing Huang, M. H., & Rust, R. T. (2018). Artificial intelligence in service. over home security cameras. Retrieved February 11, 2019 from Journal of Service Research, 21(2), 155–172. https://www.sltrib.com/news/business/2019/02/01/doorbells-have- Hullinger, J. (2016). What the Lowe’s robot will do for you – and the eyes/. future of retail. Retrieved February 11, 2019 from http://campfire- French, K. (2018). Your new best friend: AI chatbot. Retrieved February capital.com/retail-innovation/sales-channel-innovation/what-the- 13, 2019 from https://futurism.com/ai-chatbot-meaningful- lowes-robot-will-do-for-you-and-the-future-of-retail/. conversation. Kaplan, A., & Haenlein, M. (2019). Siri, Siri, in my hand: Who’sthe Gans, J., Agrawal, A., & Goldfarb, A. (2017). How AI will change strat- fairest in the land? On the interpretations, illustrations, and implica- egy: A thought experiment. Harvard business review online. tions of artificial intelligence. Business Horizons, 62(1), 15–25. Retrieved February 11, 2019 from https://hbr.org/product/how-ai- Kidd, C. D., & Breazeal C. (2008, September). Robots at home: will-change-strategy-a-thought-experiment/H03XDI-PDF-ENG. Understanding long-term human-robot interaction. In Proceedings Gaudin, S. (2016). At stitch fix, data scientists and a.I. become personal of the IEEE International Conference of Intelligent Robot Systems stylists. Retrieved February 11, 2019 from https://www. (pp. 3230–3235). computerworld.com/article/3067264/artificial-intelligence/at-stitch- Kim, T., & Duhachek, A. (2018). The impact of artificial agents on per- fix-data-scientists-and-ai-become-personal-stylists.html. suasion: A construal level account. ACR Asia-Pacific Advances. Ghahramani, Z. (2015). Probabilistic machine learning and artificial in- Knight, W. (2017). Forget killer robots – bias is the real AI danger. telligence. Nature, 521(7553), 452–459. Retrieved February 11, 2019 from https://www.technologyreview. Giebelhausen, M., Robinson, S. G., Sirianni, N. J., & Brady, M. K. com/s/608986/forget-killer-robotsbias-is-the-real-ai-danger/. (2014). Touch versus tech: When technology functions as a barrier Kober, J., Bagnell, J. A., & Peters, J. (2013). Reinforcement learning in or a benefit to service encounters. Journal of Marketing, 78(4), 113– robotics: A survey. The International Journal of Robotics Research, 32(11), 1238–1274. Goldfarb, A., & Tucker, C. (2013). Why managing consumer privacy can Kwak, S., Kim, Y., Kim, E., Shin, C., & Cho, K. (2013). What makes be an opportunity. MIT Sloan Management Review, 54(3), 10–12. people empathize with an emotional robot? The impact of agency Gray, K. (2017). AI can be a troublesome teammate. Harvard Business and physical embodiment on human empathy for a robot. In Review, July 20. Retrieved February 11, 2019 from https://hbr.org/ Proceedings of the IEEE International. Symposium Robot Human 2017/07/ai-can-be-a-troublesome-teammate. –185). Interaction Community (pp. 180 Gray, K., & Wegner, D. M. (2012). Feeling robots and human zombies: Lammer, L., Huber, A., Weiss, A., Vincze, M. (2014). Mutual care: How Mind perception and the uncanny valley. Cognition, 125(1), 125– older adults react when they should help their care robot. In Proceedings of the International Symposium of Social Study AI Groom, V., Srinivasan, V., Bethel, C. L., Murphy, R., Dole, L., & Nass, C. Simulation Behavior (pp. 3-4). (2011, May). Responses to robot social roles and social role framing. Larson, K. (2019). Data privacy and AI ethics stepped to the fore in 2018. In Proceedings of the IEEE International Conference on Retrieved February 11 from https://medium.com/@Smalltofeds/ Collaboration Technologies and Systems (CTS) (pp. 194-203). data-privacy-and-ai-ethics-stepped-to-the-fore-in-2018- Guha, A., Biswas, A., Grewal, D., Verma, S., Banerjee, S., & Nordfält, J. 4e0207f28210. (2018). Reframing the discount as a comparison against the sale Lashinsky, A. (2019). Artificial intelligence: Separating the hype from the price: Does it make the discount more attractive? Journal of reality. Retrieved Feburary 11, 2019 from http://fortune.com/2019/ Marketing Research, 55(3), 339–351. 01/22/artificial-intelligence-ai-reality/. Gustafsod, P. E. (1998). Gender Differences in risk perception: LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, Theoretical and methodological perspectives. Risk Analysis, 18(6), 521(7553), 436–444. 805–811. Lee, A. Y., Keller, P. A., & Sternthal, B. (2009). Value from regulatory Harding, K. (2017). AI and machine learning for predictive data scoring. construal fit: The persuasive impact of fit between consumer goals Retrieved February 11, 2019 from https://www.objectiveit.com/ and message concreteness. Journal of Consumer Research, 36(5), blog/use-ai-and-machine-learning-for-predictive-lead-scoring on 735–747. 13 February 2019. Leung, E., Paolacci, G., & Puntoni, S. (2018). Man versus machine: Haslam, N., Bain, P., Douge, L., Lee, M., & Bastian, B. (2005). More Resisting automation in identity-based consumer behavior. Journal human than you: Attributing humanness to self and others. Journal of Marketing Research, 55(6), 818–831. of Personality and Social Psychology, 89(6), 937–950. Levin, S. (2017). New AI can guess whether you’re gay or straight from a Hassler, C. A. (2018). Meet Replika, the AI Bot that wants to be your best photograph. Retrieved February 11, 2019 from https://www. friend. Retrieved February 11, 2019 from https://www.popsugar. theguardian.com/technology/2017/sep/07/new-artificial- com/news/Replika-Bot-AI-App-Review-Interview-Eugenia- intelligence-can-tell-whether-youre-gay-or-straight-from-a- Kuyda-44216396. photograph. J. of the Acad. Mark. Sci. (2020) 48:24–42 41 Li, D., Rau, P. P., & Li, Y. (2010). A cross-cultural study: Effect of robot Pedersen, I., Reid, S., & Aspevig, K. (2018). Developing social robots for appearance and task. International Journal of Social Robotics, 2(2), aging populations: A literature review of recent academic sources. 175–186. Sociology Compass, 12(6). Longoni, C., Bonezzi, A., & Morewedge, C. K. (2019). Resistance to Power, B. (2017). How AI is streamlining marketing and sales. Harvard Medical Artificial Intelligence. Journal of Consumer Research Business Review, June 12. Retrieved February 11, 2019 from (forthcoming). https://hbr.org/2017/06/how-ai-is-streamlining-marketing-and- Lowy, J. (2016). Self-driving cars are ‘absolutely not’ ready for deploy- sales. ment. Retrieved February 11, 2019 from https://www.pbs.org/ Rahwan, I., Cebrian, M., Obradovich, N., Bongard, J., Bonnefon, J. F., newshour/science/self-driving-cars-are-absolutely-not-ready-for- Breazeal, C., et al. (2019). Machine behavior. Nature, 568(7753), deployment. 477–486. Luo, X., Tong, S., Fang, Z. & Zhe, Q. (2019). Machines vs. humans: The Reese, B. (2018). The fourth age: Smart robots, conscious computers and impact of chatbot disclosure on consumer purchases. Unpublished the future of humanity. New York, NY: Atria Books. working paper. Robinson, M. (2017). The tech giants of Silicon Valley are starting to rely Mandel, N., Rucker, D. D., Levav, J., & Galinsky, A. D. (2017). The on crime-fighting robots for security. Retrieved February 11, 2019 compensatory consumer behavior model: How self-discrepancies from https://www.businessinsider.com/knightscope-security-robots- drive consumer behavior. Journal of Consumer Psychology, 27(1), microsoft-uber-2017-5. 133–146. Sandoval, E. B., Brandstetter, J., Obaid, M., & Bartneck, C. (2016). Marr, B. (2019). How robots, IoT and artificial intelligence are changing Reciprocity in human-robot interaction: A quantitative approach how humans have sex. Retrieved June 12, 2019 from https://www. through the prisoner’s dilemma and the ultimatum game. forbes.com/sites/bernardmarr/2019/04/01/how-robots-iot-and- International Journal of Social Robotics, 8(2), 303–317. artificial-intelligence-are-changing-how-humans-have-sex/# Schatsky, D., Katyal, V., Iyengar, S., & Chauhan, R. (2019). Why enter- 3679d398329c. prises shouldn’t wait for AI regulation. Retrieved July 11, 2019 from Martin, K. D., & Murphy, P. E. (2017). The role of data privacy in https://www2.deloitte.com/insights/us/en/focus/signals-for- marketing. Journal of the Academy of Marketing Science, 45(2), strategists/ethical-artificial-intelligence.html 135–155. Schrift, R. Y., Wertenbroch, K., André, Q., & Frank, D.H. (2017). Martin, K. D., Borah, A., & Palmatier, R. W. (2017). Data privacy: Threatening free will. Presentation at the Symposium on Alienation Effects on customer and firm performance. Journal of Marketing, and Meaning in Production and Consumption, Technische 81(1), 36–58. Universität München. Mehta, N., Detroja, P., & Agashe, A. (2018). Amazon changes prices on Shankar, V. (2018). How artificial intelligence (AI) is reshaping retailing. its products about every 10 minutes — here's how and why they do Journal of Retailing, 94(4), vi–xi. it. Retrieved February 11, 2019 from https://www.businessinsider. Sherwood, H. (2017). Robot priest unveiled in Germany to mark 500 com/amazon-price-changes-2018-8?international=true&r=US&IR= years since reformation. The Guardian, may 20. Retrieved T. February 11, 2019 from https://www.theguardian.com/technology/ Mende, M., Scott, M. L., van Doorn, J., Grewal, D., & Shanks, I. (2019). 2017/may/30/robot-priest-blessu-2-germany-reformation- Service robots rising: How humanoid robots influence service ex- exhibition. periences and food consumption. Journal of Marketing Research, Shum, P., & Lin, G. (2007). A world class new product development best 56(4), 535–556. practices model. International Journal of Production Research, Metz, C. (2018). Mark Zuckerberg, Elon musk and the feud over killer 45(7), 1609–1629. robots. Retrieved February 11, 2019 from https://www.nytimes. Şimşek, Ö. F., & Yalınçetin, B. (2010). I feel unique, therefore I am: The com/2018/06/09/technology/elon-musk-mark-zuckerberg-artificial- development and preliminary validation of the personal sense of intelligence.html. uniqueness (PSU) scale. Personality and Individual Differences, Milgram, P., Takemura, H., Utsumi, A., & Kishino, F. (1995, December). 49(6), 576–581. Augmented reality: A class of displays on the reality-virtuality con- Singh, J., Flaherty, K., Sohi, R. S., Deeter-Schmelz, D., Habel, J., Le tinuum. Telemanipulator and telepresence technologies, 2351,282– Meunier-FitzHugh, K., et al. (2019). Sales profession and profes- sionals in the age of digitization and artificial intelligence technolo- Miller, G. (2016). Bots, chatbots and artificial intelligence, what the evo- gies: Concepts, priorities, and questions. Journal of Personal Selling lution is all (aBot) about. Retrieved June 12, 2019 from https:// & Sales Management,1–21. chatbotnewsdaily.com/bots-chatbots-and-artificial-intelligence- Smith, C. (2019) Understanding China’s Confucian edge in the global AI what-the-evolution-is-all-abot-about-a7e148dd067d. race, Quillette, Feb. 14, 2019, Retrieved from https://quillette.com/ Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., 2019/02/14/understanding-chinas-confucian-edge-in-the-global-ai- Bellemare, M. G., et al. (2015). Human-level control through deep race/ on July 6, 2019. reinforcement learning. Nature, 518(7540), 529–533. Steels, L., & Brooks, R. (2018). The artificial life route to artificial intel- Moon, Y. (2003). Don’t blame the computer: When self-disclosure mod- ligence: Building embodied, situated agents. Routledge. erates the self-serving bias. Journal of Consumer Psychology, 13(1– Syam, N., & Sharma, A. (2018). Waiting for a sales renaissance in the 2), 125–137. fourth industrial revolution: Machine learning and artificial intelli- Mori, M. (1970). The Uncanny Valley. Energy, 7(4), 33–35. gence in sales research and practice. Industrial Marketing Motyka, S., Grewal, D., Puccinelli, N. M., Roggeveen, A. L., Avnet, T., Management, 69,135–146. Daryanto, A., et al. (2014). Regulatory fit: A meta-analytic synthe- Tucker, C. (2018). Privacy, algorithms, and artificial intelligence. In The sis. Journal of Consumer Psychology, 24(3), 394–410. Economics of Artificial Intelligence: An Agenda. University of Müller, V. C., & Bostrom, N. (2016). Future progress in artificial intelli- Chicago Press. gence: A survey of expert opinion. In Fundamental issues of artifi- van Lente, H., Spitters, C., & Peine, A. (2013). Comparing technological cial intelligence (pp. 555–572). Springer, Cham. hype cycles: Towards a theory. Technological Forecasting and Parekh, J. (2018). Why Programmatic provides a better digital marketing Social Change, 80(8), 1615–1628. landscape. Retrieved February 13, 2019 from https://www.adweek. com/programmatic/why-programmatic-provides-a-better-digital- Verhoef, P. C., Stephen, A. T., Kannan, P. K., Luo, X., Abhishek, V., marketing-landscape/. Andrews, M., et al. (2017). Consumer connectivity in a complex, 42 J. of the Acad. Mark. Sci. (2020) 48:24–42 technology-enabled, and mobile-oriented world with smart prod- zdnet.com/article/big-privacy-the-data-privacy-compact-for-the- ucts. Journal of Interactive Marketing, 40,1–8. era-of-big-data-and-ai/. Villasenor, J. (2019). Artificial intelligence and bias: Four key challenges. Wilson J., Daugherty P., & Shukla P. (2016). How one clothing company Retrieved February 11, 2019 from https://www.brookings.edu/blog/ blends AI and human expertise. Harvard Business Review, techtank/2019/01/03/artificial-intelligence-and-bias-four-key- November 21. Retrieved February 11 from https://hbr.org/2016/11/ challenges/. how-one-clothing-company-blends-ai-and-human-expertise. Wainer, J., Feil-Seifer, D. J., Shell, D. A., & Mataric, M. J. (2006, Wirtz, J., Patterson, P. G., Kunz, W. H., Gruber, T., Lu, V. N., Paluch, S., September). The role of physical embodiment in human-robot inter- & Martins, A. (2018). Brave new world: Service robots in the front- action. In Proceedings of the IEEE International Symposium of line. Journal of Service Management, 29(5), 907–931. Robot Human Interaction Communication (pp.6-8). You, Q., Jin,H., Wang,Z., Fang,C., & Luo,J.(2016). Image captioning Wang, Y., & Kosinski, M. (2018). Deep neural networks are more accu- with semantic attention. In Proceedings of the IEEE conference on rate than humans at detecting sexual orientation from facial images. computer vision and pattern recognition (pp. 4651-4659). Journal of Personality and Social Psychology, 114(2), 246–257. Zhao, M., Hoeffler, S., & Dahl, D. W. (2012). Imagination difficulty and Weissman, J. (2018). Amazon created a hiring tool using a.I. it immedi- new product evaluation. Journal of Product Innovation ately started discriminating against women. October 10. Retrieved Management, 29(S1), 76–90. February 11, 2019 from https://slate.com/business/2018/10/ amazon-artificial-intelligence-hiring-discrimination-women.html. Publisher’snote Springer Nature remains neutral with regard to Wilson, S. (2018). Big privacy: The data privacy compact for the era of jurisdictional claims in published maps and institutional affiliations. big data and AI. Retrieved February 11, 2019 from https://www.

Journal

Journal of the Academy of Marketing Science – Springer Journals

Published: Jan 10, 2020

There are no references for this article.