Retailers have started to deploy robots in stores (Forgan, 2020). In some cases, robots may help retailers gain efficiency by streamlining non–customer-facing functions (e.g., cleaning, inventory checking, price changes) and likely replacing the need for employees to do these tasks, while in other cases robots may directly perform customer-facing tasks (e.g., interacting with customers, providing promotional or product information), or assist sales associates with their customer-facing tasks.

Our commentary builds on the work of Rindfleisch et al. (2022, in this issue), and presents both a new conceptual framework and a forward-looking research agenda. Rindfleisch et al. (2022, in this issue) present the results of their interviews with a senior manager at Softbank (which manufacturers robots, including Whiz), and a senior manager at Daiei (Japanese supermarket chain, which uses Whiz). They determined that initially Daiei deployed Whiz mainly to execute cleaning functions, which was particularly valuable during the COVID-19 pandemic, when staffing was limited but cleanliness demands were especially acute. This deployment attracted substantial consumer interest, leading to various photos of Whiz appearing on social media sites. Over time, the retailer deployed Whiz to share sales promotions, and employees added personalized decorations to the robot in their individual stores to reflect local landmarks. Yet Whiz’s success also has a backstory, in that Softbank had initially introduced a robot called Pepper, designed to interact directly with customers and address their customer service queries (Softbank, 2022b). But due to Pepper’s limited functionality, Softbank shifted its focus to the cleaning robot Whiz (Nussey, 2021), which might suggest that while robots may be able to execute non-customer-facing tasks, they are ill-equipped to engage in customer-facing applications.

Beyond Pepper and Whiz, there have been many instances of retailers deploying robots. Considering first non-customer-facing domains, the Tally robot was deployed in Meijer stores, where it searches constantly for missing, mispriced, or misplaced products, and also assists with inventory updates. With its assistance, retailers are two to fifteen times more likely to identify out-of-stock items, than are retailers that do not rely on Tally (Ferguson, 2021). The Whiz robot cleans floors; some Hilton hotels earned nearly 2 × higher guest cleanliness scores after introducing it (Softbank, 2022a). Other robots deployed by retailers include (i) K5, which can patrol shopping malls and parking lots (Robinson, 2017), (ii) Millie, used in Woolworths stores (in Australia) to perform safety tasks, such as clean ups of spills, (iii) Amazon’s Kiva robots, located in distribution centers, that can perform sophisticated operations and keep track of millions of items (Forgan, 2020; Mims, 2018), and (iv) Walmart’s Bossa Nova robots, which track inventory (Nassauer & Cutter, 2019).

Yet some robot deployments also entail customer-facing domains, and it is these robots that are more widely reported in the popular press. LoweBot (at Lowe’s) helps customers find items in stores, after they request its assistance via either a voice command or by typing on its touchscreen (Morgan, 2020). Robot servers, such as Peanut, deliver food; using infrared technology, it can navigate its path, relying on lidar to detect any obstacles (Simon, 2021). Various versions of robot baristas include Café X, Rocky, Monty Café, and Rozum Café (e.g., Hochman, 2018). Despite excitement about such encounters, expectations about the robots’ capabilities still need to be tempered (Davenport et al., 2020); their limited functionalities mean they cannot (fully) function sans human support. For example, Peanut assists human servers by bringing orders to the table, but the servers still take orders and interact with customers (Simon, 2021). Table 1 lists exemplars of robots deployed by retailers.

Table 1 Exemplars of robot deployment in retail
Full size table

In this paper, we first provide an overview of pertinent AI and Robotics literature, predominantly from the marketing domain. Next, we develop and present a four-quadrant framework to understand the evolution and deployment of robotics in retail and service settings. Finally, we present an agenda for future research, to help stimulate and guide future examinations in this rapidly evolving field.

Literature review

We provide a summary of relevant prior research in Table 2. The summary pertains to either work in robotics or to work in artificial intelligence (hereafter, AI) that can inform how robots with AI capabilities will likely function, and influence customer, employee, and retailer/service provider performance. For a more comprehensive summary of robotic work, interested readers are referred to Mende et al. (2019).

Table 2 Literature review
Full size table

Artificial intelligence and robots

In their theoretical consideration of how AI might alter retailing, Guha et al. (2021) offer three main predictions and recommendations. First, early applications of AI should address non–customer-facing tasks because such deployments offer substantial value to retailers. In contrast, customer-facing interactions are relatively harder to control and more variable. Due to the technological capacities and limitations of today’s AI, such interactions increase the risk of service failures. Second, retailers and service providers should use AI to augment their employees’ capacities, rather than replacing them (Guha et al., 2021). In essence, retailers should look at balancing human employees’ and AI input, by positioning AI input as a complement to human input. With this approach, the retailers can overcome the limitations of current AI technology by leveraging human resources to intervene or smooth over potential service failures. Third, retailers need to develop realistic expectations.

As suggested by Davenport et al. (2020), the effects of robots are likely to be evolutionary, not revolutionary, despite some exaggerated claims in the popular press. That is, we currently have access to artificial narrow intelligence (ANI), not artificial general intelligence (AGI) (Guha et al., 2021). Although ANI performs better than humans in domains with structured data and predictable outcomes, it is less effective in novel domains. In contrast, AGI performs better in novel domains and on complex, idiosyncratic tasks (Huang & Rust, 2018), but it is not a near-term reality (Davenport et al., 2020). According to AI researchers, the odds of achieving AGI by 2050 are 50–50 at best. (Guha et al., 2021). Thus, the benefits of robots may well be overestimated for the near term (Davenport et al., 2020). It is likely easier for retailers to start using robots (i) for non–customer-facing applications, such as cleaning, checking inventory and prices, and (ii) in ways that could augment the human capabilities of their existing workforce.

Recent research on robots

Recent papers have highlighted some important points as regards the use and deployment of robots in retail and service settings. Noble et al. (2022) argue that, in retail and service settings, robots and humans should collaborate closely, rather than allowing robots to replace humans. They also identify an initial prioritization on “robots running warehouses or performing constant inventory assessments” (Noble et al., 2022, p. 202), reflecting the notion that robots should initially be deployed in non–customer-facing functions. In contrast though, studies of how robots might influence satisfaction tend to focus on customer-facing domains. Both Van Doorn et al. (2017) and Grewal et al. (2020b) posit that perceived warmth and perceived competence mediate robots’ impacts on customer satisfaction. Similarly, in a study of AI-powered voice assistants (Guha et al., 2022), the findings imply that perceived artificiality and perceived intelligence mediate the impact of these devices on customers’ continued usage intentions.

Deploying robots may also have negative outcomes. Mende et al. (2019) posit that robots can elicit perceptions of eeriness, with downstream effects on consumption, including inducing defensive consumption (e.g., eating more comfort food). As a broader point, Castelo et al. (2018) argue that very humanlike robots may prompt discomfort, in line with uncanny valley theory; this effect pertains to the sense of discomfort or eeriness experienced when robots are too humanoid (Grewal et al., 2020b). Such concerns seem more relevant in customer-facing domains, so again, the recommendation that retailers should focus on initially deploying AI (and robots) in non–customer-facing domains appears appropriate (Guha et al., 2021).

Proposed framework

How should retailers think about deploying robots? Building on contributions from Guha et al (2021) and Davenport et al. (2020), we propose a four-quadrant framework (Fig. 1) for guiding their deployment decisions. In this framework, we specify two key influences: whether the usage domain involves non– or customer-facing applications, and if retailers’ intent is to augment or replace human resources.

Fig. 1
figure 1

Deployment of robots in retail

Full size image

Usage domain

Guha et al. (2021) (see also Shankar, 2018) outline two pathways by which AI is likely to affect retailing: demand-side (e.g., in-store customer experience management) or supply-side (e.g., inventory optimization). But we argue that the usage domain goes beyond such a dichotomous split, and instead reflects a continuum that varies in the extent to which robots face customers. At one end of the continuum, robots like Tally, which exclusively track inventory, are fully non–customer-facing; Whiz is somewhat customer-facing (Rindfleisch et al. 2022, in this issue); and LoweBot is substantially customer-facing, because its primary functions relate to interacting with customers. Customer-facing interactions are relatively complex and variable. Because AI currently achieves only ANI (not AGI), there is high risk of service failures, and considering the direct interaction with customers, such service failures may have dire consequences.

Usage intent

Both Davenport et al. (2020) and Guha et al. (2021) suggest using AI, in its ANI form, to augment rather than replace human capability, reflecting ‘human plus machines’, rather than ‘humans versus machines’. In practice though, we note that some retailers choose to deploy robots with the intent that such robots augment human capability. For example, the Café X robot has the intent of augmenting human capability, with the robot focusing on coffee preparation, and with human associates focusing on order taking, order advice, and troubleshooting if there is service failure. In contrast, other retailers choose to deploy robots with the intention of largely replacing human capability. For example, Lowebot (and to an extent Pepper) looked to replace human capability, especially for relatively simple customer service tasks.

Four-quadrant framework

Non–customer-facing applications, augmenting human capability

This quadrant (Backstage Robot Assistant) relates to the proverbial low hanging fruit, involving immediate benefits at relatively low costs. For example, Tally checks prices and alerts employees to fix any incorrect price tags, and the K5 security robot alerts security personnel to any disturbances in parking lots. The backstage, non–customer-facing domain is relatively easier to operate in, and the goals set for the robot are relatively modest (they still require human assistance). Thus, these robots should be relatively effective in the limited tasks they are designed to perform, and retailers should embrace such assistance and deployments.

Non–customer-facing applications, substituting for humans

In this quadrant (Backstage Robot Worker), robots take on a more autonomous role in non-customer-facing domains. Although the non–customer-facing domain remains relatively easy to operate in, the usage intent (i.e., substituting for humans, and operating fairly autonomously) is ambitious, such that it is unclear whether current ANI robot technology can achieve this goal. Whiz operates reasonably autonomously, but even for the seemingly straightforward task of cleaning floors, it sometimes needs human support (e.g., if it gets stuck under furniture, or if it encounters stairs). To the extent possible, retailers should pursue deployments in this quadrant, but they also should be cautious and prioritize simpler tasks that can be executed by existing ANI technology.

Customer-facing applications, augmenting human capability

In this quadrant (Robot Coworker), retailers deploy robots to augment human capabilities in customer-facing domains, notwithstanding the challenges in customer-facing domains. For example, the Café X barista robot prepares various coffee brews, whereas human baristas engage customers, perform cleaning tasks, and refill coffee beans and other supplies. To the extent possible, retailers should pursue deployments in this quadrant, but we again call for caution, to ensure the robots are assigned tasks they can actually execute, given current ANI technology. Further, retailers should provide sufficient human support, such that a human employee can intervene to address any service failures.

Customer-facing applications, substituting for humans

In this final quadrant (involving Robot Associates), we examine whether retailers should deploy robots to operate somewhat autonomously, in customer-facing domains. We highlight two prominent, linked risks: (i) existing ANI technology struggles with operating autonomously in customer-facing domains, and (ii) if the robot aims to replace human employees, any service failures are difficult to recover from. LoweBot can respond to simple requests from shoppers (e.g., “Where can I find 1″ bolts?”), but if the requests are complex or phrased in unconventional ways, it might fail to understand, and thereby induce customer frustration. Thus, autonomous robot deployments demands great caution. The risk of service failure is high, due to both the limitations of current ANI technology and the lack of available human support, and in these direct interactions, such failures would have immediate and costly impacts on the firm-customer relationship. Some initial evidence indicates that customer-facing robots without human assistance may be useful in sales interactions involving embarrassing products (Van Doorn & Holthoewer, 2020). Customers arguably may perceive that robots are less likely to judge them, so when they have a need for sensitive, personal products, they may be more comfortable interacting with a robot. But, generally speaking, as the experience of Softbank with its Pepper robot indicates, it remains risky to implement potentially unreliable robot technology that can induce service failures, sans human support to address such failures. Consistent with this point, Softbank has suspended the Pepper program, while increasing its focus on Whiz.

Research agenda

The robot revolution in retail is just beginning, opening a wide variety of areas for research. Below, we present three areas for future research focus, relating to the interaction of robots and (i) customers, (ii) retail employees, and (iii) retailers; in addition, we also look at public interest issues.

Robots and customers

We briefly discuss four factors relevant to how customers interact with and react to robots. First, understanding how robots influence customers’ evaluations of the retail setting is critical. Rindfleisch et al. (2022, in this issue) posit that customers react positively to Whiz, even though Whiz may not interact with them directly. Any future proposed frameworks must be flexible enough to encompass customers’ evaluations relating to robot deployments across both customer-facing and non–customer-facing roles.

Second, several concerns arise when robots interact with customers. Interactions with a robot may allow the retailer to capture substantial, personal information, e.g., through sensors, which raises ethics and privacy concerns. Should such information even be collected? How may it be used? Furthermore, what are the distinct responsibilities of retailers, customers, and policy makers regarding how such private information is protected, after its collection?

Third, customers might react to and evaluate robots differently, depending on their features. For example, robot evaluations could be contingent on perceived artificiality and perceived intelligence (Guha et al., 2022), and certain features of the robots themselves (e.g., those that evoke anthropomorphism) might have stronger or weaker relative effects on perceived artificiality versus perceived intelligence. Consumers’ anthropomorphism could enhance their evaluations, but at very high levels, it could lead to negative impacts (Castelo et al., 2018; Mende et al., 2019).

Fourth, customers might not respond positively (or neutrally) to robots, as demonstrated by real-world cases where customers and passers-by have caused intentional harm to robots (Wilson, 2017). How should retailers think about such cases, and what protections–if any–should they put in place to limit such damage to the robots they deploy? Should robots have some elements of self-defense capability?

Robots and employees

There is limited research into how retail employees perceive–and react to–robots. We note that Rindfleisch et al. (2022, in this issue) find that employees perceive positively, and react positively, to Whiz. What drives employees’ perceptions of and attitudes toward robots? What downstream outcomes, such as retail store evaluations, stem from such perceptions and attitudes? Can positive employee attitudes towards robots, both non-customer-facing robots and customer-facing robots, influence downstream outcomes, like retail store evaluations, and stock market evaluations?

Robots and retailers

Regarding the retailer (at an organizational level), we suggest four areas for research. First, Guha et al. (2021) investigate which factors drive retailers’ adoption of AI; similarly, it may be worthwhile to examine which factors drive retailers’ adoption of robots. We have identified some likely influences (see Fig. 1), but other factors also could be important. Second, research should specify which types of retailers can benefit most from using robots, as well as whether robot adoption paths or deployment strategies might be contingent on the retailers’ type. Third, robots can take many (physical) forms, such as humanoid forms (e.g., LoweBots), functional forms (e.g., K5, Peanut), or relatively immobile designs (e.g., Café X). Research can determine which factors define the optimal physical form for a robot, contingent on retailers’ type. Fourth, retailers are deploying a host of in-store technologies (Grewal et al., 2020a), so it will be important to study how robots can co-exist and enhance the deployment of those other technologies.

Robots and the public interest

Retailers may use robots in variety of ways. In this paper, we have described how retailers may deploy robots in the store, in both non-customer-facing domains and customer-facing domains. In addition, retailers may deploy robots outside the store, e.g., Domino’s plans to use an autonomous vehicle to deliver pizza to customers (Benveniste, 2021). All of this raises a variety of public policy questions (i) how will robot deployment impact other retail stakeholders e.g., retail employees, delivery companies like Door Dash and Uber Eats, (ii) what public infrastructure may be needed if robots are deployed outside the store e.g., would robots be able to use bicycle lanes (robots cannot co-exist with cars on main roads, as cars are both heavier and move faster), (iii) would the public be accepting of robots, both in-store and outside-of-the-store, noting that robots may be coming close to customers and thence capturing personal customer information as they move around in-store or outside the store while executing their tasks, and (iv) noting that customers sometimes exhibit violent tendencies towards robots (Wilson, 2017), how acceptable would it be, if the robots were allowed some self-defense mechanism (even a relatively non-violent self-defense mechanism, e.g., a high-pitched alarm) towards customers who may wish it harm?

Conclusion

To reflect on how robots are likely to inform the future of retailing, we propose a framework that gives retailers suggestions for how they should deploy robots. We also offer a three-part research agenda, related to the implications of robots for customers, employees, and retailers. Accordingly, we hope this article provides insights for retailers, researchers, and public policy experts.

Although deploying robots promises substantial benefits for retailers, we end on two notes of caution. First, already pressing ethics and privacy concerns are likely to intensify even further, as privacy laws and regulations take full effect. Second, it is important for retailers to develop realistic expectations. According, to previous theorizing, AI (and robots) can provide evolutionary benefits in the immediate term but provide revolutionary benefits in the long-run. We echo this point and anticipate that the benefits of robots might be overestimated in the short term, whereas the cautionary tale of various robots (like Pepper) highlights the need for realism. Still, the long-run potential benefits of robots offer great promise.