From automats to algorithms: the automation of services using artificial intelligence

Chris Meyer (Rensselaer Polytechnic Institute, Troy, New York, USA)
David Cohen (Skidmore College, Saratoga Springs, New York, USA)
Sudhir Nair (University of Victoria, Victoria, Canada)

Journal of Service Management

ISSN: 1757-5818

Article publication date: 31 March 2020

Issue publication date: 24 September 2020

1524

Abstract

Purpose

The paper aims to fill this gap by positing a framework that considers the service automation decision as a matter of knowledge management: a choice between human resident and codified knowledge assets.

Design/methodology/approach

The paper is a conceptual paper, grounded in the knowledge-based view.

Findings

The paper uses the information processing theory, which argues that the level of uncertainty in a process should dictate the type of knowledge deployed, as the contingency for the automation choice, and customer interaction uncertainty as the driver of that contingency. From these ideas, propositions are generated relating customer interaction uncertainty and service automation. Further implications for artificial intelligence (AI) are also explored.

Originality/value

The framework illuminates and informs the strategic choices regarding service automation, including the use of AI in professional services, a timely and highly important topic. It offers a valuable model for practitioners and contributes to the academic literature by pointing the way for future directions for scholarly research.

Keywords

Citation

Meyer, C., Cohen, D. and Nair, S. (2020), "From automats to algorithms: the automation of services using artificial intelligence", Journal of Service Management, Vol. 31 No. 2, pp. 145-161. https://doi.org/10.1108/JOSM-05-2019-0161

Publisher

:

Emerald Publishing Limited

Copyright © 2020, Emerald Publishing Limited


Introduction

Firms have been automating service work for decades. Automats, restaurants in which serving and paying were automated through the use of vending machines, were introduced in the first half of the twentieth century (Strauss, 2017). More recently, the growth in the capability of information technologies has allowed professional services firms to automate a number of aspects of service production. In retail banking, tellers have been both supplemented and supplanted with ATMs and banking apps. The sale and servicing of insurance policies has moved online, and dating services have become algorithm-driven. Artificial intelligence (AI), a package of digital technologies able to replicate cognitively sophisticated tasks, is replacing humans in much more complex knowledge-intensive service work, including medicine, finance, security, education and law (Becerra-Fernandez and Sabherwal, 2014; De Keyser, et al., 2019). This has caused considerable concern among workers, managers and policy makers, who worry that knowledge workers are about to face the same dislocations from cognitive tools as laborers have faced from machine tools (Lee, 2018). One widely read scholar has even suggested that AI's threat to employment is one of the three biggest challenges that humanity faces, trailing only nuclear holocaust and global warming (Harari, 2018).

These concerns are not irrational. AI has the potential to upend our ideas about what tasks are uniquely suited to humans. A clear working definition of AI comes from Amazon, which is investing heavily in service automation: “Artificial Intelligence is the field of computer science dedicated to solving cognitive problems commonly associated with human intelligence, such as learning, problem solving, and pattern recognition” (https://aws..com/learning/ai/, accessed January 8, 2020). It is a “simulation system of collecting knowledge and information and processing intelligence” (Grewal, 2014, p. 13). It takes in data and uses algorithms to make decisions and/or predictions (Agrawal et al., 2017; Lee, 2018). AI competes, as an automation technology, with smart, experienced human professional service providers.

While scholars have studied service automation in a range of arenas, there has been limited work done to lay out a theoretically driven framework to evaluate how firms make strategic decisions with regard to automating service functions. Indeed, it is difficult to find any consideration of service automation that treats it as a strategic choice, rather than a foregone conclusion dictated by competition and/or the mere existence of the technology. The purpose of this paper is to begin to fill this gap. Without such a framework, we cannot predict or evaluate the spread and impact of digital technologies, such as AI. Now that AI is more than simply “software that performs certain repetitive and dreary service tasks previously performed by humans, so that humans can focus on more unstructured and interesting tasks” (Lacity and Willcocks, 2016, p. 41), we urgently need frameworks that help us understand service automation decisions. An appropriate framework will also help evaluate the dissemination of the next pertinent technology to come along.

Deploying automation in services is a strategy-driven contingent decision. Because AI embeds human knowledge and skill within machinery, organizational knowledge scholarship provides the theoretical basis for understanding the implications of this contingent decision, and customer interaction uncertainty (Larsson and Bowen, 1989) provides a strategic rationale. The proposed framework suggests that service firms, including knowledge-intensive professional service firms, should not simply automate whenever the technology becomes (in a narrow sense) cost-effective. Rather, the decision to automate will be driven by the extent to which the new technological solution meshes with the nature of the knowledge involved and the role that knowledge and knowledge deployment plays in the firm's overall customer strategy. For practitioners, the framework's core value is to identify the extent to which implementing automation is contingent and how to resolve any such contingency.

Both managers and scholars need to understand if automation is a best practice or something to be done selectively. To answer this important question, the paper utilizes information processing theory (Galbraith, 1974; Simon, 1978) to link the type of knowledge used to contend with uncertainty. And, given the context of service automation, the model uses a key uncertainty that all service firms face: customer interaction uncertainty. Service firms must allow customers to make choices that affect the production of the service they consume (Wirtz and Zeithaml, 2018). The customer must order a meal before the restaurant can serve it; the doctor and the lawyer are both dependent on the customers' description of their situation and their willingness to follow advice.

The extent to which customers are allowed to introduce uncertainty into the production process is a strategic choice made by the firm (Larsson and Bowen, 1989; Tansik, 1990; Skaggs and Youndt, 2004; Meyer et al., 2015b). The proposed framework, then, links service firm strategic choices regarding customer interaction uncertainty to the resources deployed to suggest a set of decision criteria for the type of knowledge used by the firm: human or organizational capital (e.g. Bontis, 1998). In short, the model uses the knowledge-based view, and specifically, the information processing theory, to suggest where and when to automate service processes, including the use of AI.

This project contributes to the service management literature by proposing a model for service automation based on organizational theories. In so doing, it takes service automation decisions beyond a best practice or a matter of competitive mimicry, grounding them instead on strategic concepts that unite the firm's approach to customers and its deployment of different forms of knowledge resources. This extends extant theory in knowledge management and service strategy, applying both in new ways to address an important gap. The ideas are of use to managers as well, as they suggest where and how to use automation or human service workers based on ideas that are both theoretically sound and operationally practical. The paper also contributes to scholarship and practice regarding the use of AI technologies in service firms by bringing established management theories and concepts to bear on this important and timely topic. Managers, in particular, need such guidance, given the vast enthusiasm for the use of AI that is currently in business press. Like any form of knowledge, AI and all forms of service automation have their place, but managers need good models to know where that place is.

Theoretical background

Whether to implement service automation is often seen as a matter of pure cost-benefit (Ivanov and Webster, 2018). It is, however, a more nuanced question: does automation advance firm strategy and improved performance? On the one hand, humans bring variance into work processes (Meyer et al., 2015a), potentially making quality variable and unpredictable. In addition, wages and benefits might outweigh the cost of automation, and people often cannot work as quickly and efficiently as machines. Further, some clients such as elderly patients might need sustained levels of support, which may not be feasible from a cost perspective (Čaić et al., 2018). Finally, as AI technologies begin to replace even highly sophisticated human service providers (De Keyser, et al., 2019), they can be used to convert human capital (controlled by the employee) into organizational capital assets controlled by the firm.

On the other hand, if poorly implemented or strategically inappropriate, service automation can alienate customers. A confusing phone tree, an online “bot” that cannot address a specific situation, a feeling on the customer's part of being treated as part of an undifferentiated mass, might be much more expensive to the firm than the cost of installing an automated system. Service automation decisions are clearly not simple, but will require useful frameworks to inform them.

In a basic form, automating a service is a matter of moving the knowledge of how to perform the service out of a human and into a machine that can perform the same task, whether cognitive or physical. The conversion of standard physical tasks into machines has become routine in manufacturing, but the conversion of complex knowledge into machinery is not. A model for service automation, then, will need to be based on theories of organizational knowledge that inform decisions about converting and deploying human knowledge as machine-resident knowledge. Given the potential need for human interaction, noted above, such a model would also need to incorporate concepts from the service literature regarding customer interaction. The theoretical foundations of this paper, then, utilize two specific areas of the organizational knowledge literature: intellectual capital (Bontis, 1998) and information processing theory (Galbraith, 1974), along with the concept of customer interaction uncertainty (Larsson and Bowen, 1989). Each is discussed in turn, below.

Knowledge management and intellectual capital assets

The creation, deployment and exploitation of knowledge has come to dominate the management of successful firms as we move toward a post-industrial age characterized by the increasing strategic importance of knowledge and the decreasing importance of physical assets (Meyer et al., 2017b). The core managerial challenge for firms is “devising mechanisms for integrating individuals' specialized knowledge” (Grant, 1996, p. 114) for the purpose of improving organizational performance. One reason that knowledge is so valuable to organizations is that it can reduce uncertainty (Galbraith, 1974; Spender and Scherer, 2007).

Organizational knowledge can be a highly abstract concept, particularly for managers. The concept of intellectual capital, defined as the productive knowledge assets of a firm (Bontis, 1998), is a way to categorize and describe different forms of knowledge that firms use. Scholars generally classify intellectual capital resources as human, organizational and social capital (Youndt and Snell, 2004; Meyer et al., 2015a). Most importantly, these three elements of intellectual capital contribute to performance improvements (Youndt and Snell, 2004). Strategically, they can be rent-generating resources. Through this lens, service automation decisions are a choice between human and organizational capital.

Human capital consists of the knowledge, skill and productive abilities that reside within the individuals that work in a firm. Human capital is owned by the employee, but used by the firm. Historically, human capital is also a critical and central element in knowledge-intensive and professional service firms (Empson, 2001; Von Nordenflycht, 2010).

Organizational capital is the one form of intellectual capital that a firm can be sure of retaining, regardless of who works there. It is recorded or embedded knowledge that can be stored, copied and moved (Youndt and Snell, 2004). It can exist in a number of forms, ranging from stored data to written process instructions, computer code, algorithms, data and simply, “how we do things here.”

Service automation is a form of organizational capital. It is knowledge that has been extracted from humans and placed into a machine. Even in a very simple form, such as in an automat, there is knowledge: when the customer pushes a button, the correct food must emerge. ATMs involve a somewhat more sophisticated form of organizational capital. They include algorithms that check customer balances and passwords prior to dispensing cash. They also have the ability to scan checks and cash electronically to determine amounts and authenticity – tasks involving information, perception and decisions that still are done by human tellers, if that is what the customer prefers. AI processes, which involve enacting complex decision processes based on the review of large amounts of data, are also forms of organizational capital, allowing for the automation of highly sophisticated services. Automating services, from a knowledge perspective, is a matter of choosing between human and organizational capital.

This raises the question of how and where to use these two forms of intellectual capital. First, organizational capital has an advantage over human capital, in that it has properties of a public good (e.g. Samuelson, 1954). A public good is non-rivalrous, in that its use by one actor has no effect on its value or its simultaneous use by another actor. Service automation can be used in multiple instances simultaneously, serially and in distant geographic locations again and again. While the number of employees and their location limit the firm's ability to use human capital, converting human capital to organizational capital can often vastly increase a firm's throughput.

Second, organizational capital, unlike human capital, is actually owned by the firm (Youndt and Snell, 2004), making it more easily controlled and protected than human capital. People can resign and move to (or become) a competitor and even take client-specific human capital with them, but recorded knowledge, including automation technologies, stays in the building, so to speak. Lastly, machines produce a less variable product than humans (Meyer et al., 2015a). While errors do arise from automated services, they are much less likely.

Automation can be beneficial to service firms; it reduces production risk, expands capacity and gives a firm improved protection of its knowledge assets. But, this does not mean (as is sometimes assumed) that firms will or should automate as soon as automation is possible or economical. Instead, the information processing theory (Galbraith, 1974) suggests that automation will be a contingent choice.

Information processing theory and human vs organizational capital

The information processing theory has been applied to a range of organizational phenomena beginning with Galbraith's own (1974) work tying together uncertainty, knowledge, structure, people and strategy. Galbraith gave three mechanisms whereby organizations could deal with uncertainty that have clear corollaries to these two forms of intellectual capital. The first is the use of “rules and programs” (1974, p. 29), which corresponds directly to organizational capital. However, he asserts that rules and programs can only cope with limited amounts of uncertainty. For greater uncertainty, the organization must turn to humans (and human capital), using one or both of Galbraith's other mechanisms.

The second mechanism Galbraith prescribes is hierarchy. In using hierarchy, the organization sends items up the chain that require greater information processing (knowledge) than exists at lower levels. Here, the key information processors are senior, experienced managers with an extensive view of the organization and the environment. The third mechanism is “coordination by targets or goals” (1974:29). In this mode, the organization “reduces the level of information processing in the hierarchy by increasing the discretion at lower levels” (1974, p. 29). Both of these latter two mechanisms require investment in human capital, either in those individuals up in the hierarchy that can resolve uncertainty or in individuals across the organization at lower levels that have had discretion granted to them.

In summary, firms can create and codify plans and guides for decision-making or they can grant individuals the discretion to resolve uncertainty. The former is the form of knowledge used in service automation. Galbraith argues that each of these methods can cope with uncertainty, but that greater levels of uncertainty require the use of human judgment. If you have a particularly thorny situation, you have to put aside the rules and the automation and call in the human experts. This is the primary argument for human capital over organizational capital, particularly in professional service firms (PSFs), that are trying to solve complex customer problems (Von Nordenflycht, 2010). Information processing theory, therefore, explicates the contingency at the heart of service automation decisions – where and when to use which form of knowledge. Organizational capital has several benefits, as noted, but is only able to contend with limited amounts of uncertainty.

This is the basis of our fundamental model of service automation, which applies to all forms of automation, whether it be simple delivery of physical goods or the use of AI to perform sophisticated service tasks. This fundamental model is shown in Figure 1.

Propositions

Knowledge is used to contend with uncertainty (Spender and Scherer, 2007), and for service firms, a key source of uncertainty is the customer. The decision whether to automate particular customer interactions will depend in turn upon strategic choices made by the firm with regard to its customers.

Customer interaction uncertainty

Customer interaction uncertainty (Larsson and Bowen, 1989; Tansik, 1990; Skaggs and Youndt, 2004) is the primary source of uncertainty in service production. Services are performed in conjunction with the customer, unlike manufacturing processes. PSFs, where knowledge deployment is paramount, have particularly deep relationships with customers (Von Nordenflycht, 2010). For service firms, uncertainty is not simply an exogenous result of circumstances, it is inherent in their processes because of the interaction with customers (Meyer et al., 2015b). Scholars have defined this as customer interaction uncertainty, consisting of the diversity of the services offered to the customer and the extent to which the customer interacts with the firm during the production of the service (Larsson and Bowen, 1989). A firm that attempts to satisfy a broad range of customer demands and offers customers close, active participation in the work process is allowing high levels of customer interaction uncertainty. Conversely, customer interaction uncertainty is low where customers must operate in the way the firm wants them to, the output is restricted to a narrow set of offerings and the customer's participation is limited.

For firms that offer relatively straightforward services, such as fast-food, delivery or retailing, this may take the form of setting boundaries around how much individualized work a service employee may do for a client. McDonald's sells what they sell; the postal service delivers when they deliver; Wal-Mart stocks the sizes it stocks. Customer choice exists, but is limited, and thus so is uncertainty. Other firms compete by offering a wide range of choices. Nordstrom's strategy in this regard is famous, with their dictum that there are no rules other than employees using their good judgment in all situations, which allows workers to modify clothing on the fly, for example (Goodall, 1992).

Customer interactions are even more important in PSFs. The function of PSFs is to help clients with difficult problems, such that these firms and their employees often have long-term, close relationships with customers (Lovelock, 1984; Gluckler and Armbruster, 2003). PSF work involves “high interdependence on the part of practitioners and between practitioners and their clients” (McGrath, 2005, p. 551). Uncertainty that stems from their customers and their customers' problems is something that PSFs must deal with on an ongoing basis as a part of their core work.

Even so, PSFs have choice with regard to the breadth of offerings and the extent of the customer interaction. A hospital might offer a range of wellness services in addition to traditional surgical, etc., services, along with extra personal care and attention from health providers. Or, it might offer a limited range of care options with as little interaction as possible. Many emergency rooms, where uncertainty is extremely high, have been closed, as hospitals rethink their strategies.

As these examples illustrate, service firms determine how much uncertainty customers will be allowed bring to their process (Gittell, 2002; Meyer et al., 2015a; Tansik, 1990). A firm can allow a high level of customer interaction uncertainty or allow very little. Importantly, for firms to perform well, this strategic choice must be aligned with the appropriate knowledge resources (Skaggs and Huffman, 2003; Skaggs and Youndt, 2004; Meyer et al., 2015a).

In automats, firms kept customer interaction uncertainty low by cutting interaction itself to a minimum. They also offered a narrow range of services: delivery of a limited set of pre-made items. Their decision to limit customer-specific customization of the food they prepared was closely related to their decision to automate. For service firms that predominantly provide a physical product to customers, such as retailers and delivery services, there is little knowledge involved in most cases, and the automation decision is a matter of customer tastes. For other services that involve more knowledge, where the information asymmetry between customers and service providers becomes a driving force (Nayyar, 1993), the resources and strategic decisions are more involved. For these firms, automation is a matter of whether they choose to deploy knowledge in human form or in the form of a machine.

This is a straightforward model of the key contingency for organizations using knowledge that either lives within or without workers. The form of intellectual capital a firm chooses must fit with its overall strategy and satisfy the customer – services should not just be automated whenever possible. AI is explored in more detail below, but AI does not change this basic contingency; it simply expands the opportunity to use organizational capital with increasingly complex and unpredictable customer interactions.

To summarize, efficiency, control of firm assets and limitation of the uncertainty inherent in using human capital militates in favor of adapting automation in service processes. The information processing theory, however, makes clear that high levels of uncertainty require the use of human capital. This contingency is driven by the firm's choice as to customer interaction uncertainty, the key uncertainty driver for service firms. For a service firm, the desire to automate using highly codified processes requires a strategic tradeoff: it necessitates restricting the firm's opportunity set and customer interaction. The firm's customer strategy and knowledge choices need to be in alignment. Firms allowing high levels of customer interaction uncertainty would utilize relatively higher levels of human capital and lower levels of organizational capital (service automation), and vice versa.

P1.

Firms that choose to allow low (high) levels of customer interaction uncertainty in service processes will be associated with higher (lower) levels of organizational capital usage and automation.

Extending these ideas, an appropriate fit between customer interaction uncertainty, strategy and intellectual capital resources should improve firm performance. Human capital, in particular, can be quite expensive. If human capital is not deployed solely where customer uncertainty requires it, the costs are unlikely to be offset with appropriate revenues. The firm's performance should improve substantially with service automation in a way that would fit with their customer strategy (Wirtz and Zeithaml, 2018). Similarly, if service automation is used in an attempt to counter high levels of customer interaction uncertainty that is better suited to human capital, the firm is likely to underperform, as competitors using the appropriate knowledge resources would be more successful in attracting and retaining customers.

P2.

A fit between customer interaction uncertainty and human and organizational capital (service automation) usage, where high customer interaction uncertainty is combined with low organizational capital and high human capital, and vice versa, will improve performance in service firms.

Knowledge resources, customer interaction uncertainty and artificial intelligence

Metaxiotis et al. (2003) define AI as consisting of three types of functionality. Expert systems are technological systems with substantial domain knowledge that can find solutions “which would normally be achieved by a skilled human when confronted with significant problems” (2003, p. 217). Their second type of AI is artificial neural networks; systems that are “trained” to scour existing data for patterns that can be used to identify or predict. They are the core technology behind pattern recognition in machine learning algorithms. The third form of AI they note is autonomous agents, which are systems that can act on their own within a given environment, such as various bots and robot-applications (Rosenthal-von der Pütten et al., 2018). These elements may combine, too, such that a neural network informs decisions made by an expert systems algorithm (Lee, 2018).

AI is an example of the codification of human capital into organizational capital. It is, however, a more advanced form of such codification, with unique characteristics as discussed, below. Technological advances have allowed for greater storage and management of data, as well as increased computing power, creating platforms on which AI can be made more widely available. Mathematically, AI uses a series of techniques designed to reach optimal decisions. One of the more prominent is deep learning (Dechter, 1986; Hinton, 2007), mathematical functions that have greatly advanced pattern recognition technology.

These algorithms' effectiveness is a function of the data they operate on. AI is useless without the data to inform it, and it is limited by whatever data are used to train it. Even “unsupervised learning,” wherein systems assess data for patterns without guidance in the algorithms (Hinton and Sejnowski, 1999), is completely dependent on the data that feed the system.

Data and algorithms are both forms of organizational capital, though they operate with a different locus. Breaking organizational capital into subcategories, data would be considered declarative capital, i.e. knowledge about something, while algorithms are a form of procedural capital, which is knowledge about how to do something (Ofek and Sarvary, 2001).

Artificial intelligence and human interaction

It is useful to consider AI in light of the two elements of customer interaction uncertainty: interaction with customers and the breadth of services offered (Larsson and Bowen, 1989). Inherent in the idea of customer interaction uncertainty as a strategic choice is the idea that this interaction in-and-of itself provides value for customers (Jaakkola et al., 2015). Morris (2001) notes that even in a consulting context interaction with the client helps the client to feel that value has been created for them. Some customers simply want to participate in the service process, so failing to offer them the chance to do so risks losing their business (Jaakkola et al., 2015). Some contemporary writers see automation as a clear choice driven by the type of service tasks. Lacity and Willcocks, for example, suggest the following as a best practice: “business operations selected structured tasks associated with end-to-end processes for automation and left the tasks requiring judgment and social interaction for humans” (2016, p. 46). Other scholars see anthropomorphized versions of AI, such as in service robots (Wirtz, et al., 2018; Ivanov and Webster, 2019).

Human service interaction must be reconsidered with the emergence of AI technologies. Chatbots, service apps and the proliferation of automated assistants such as Siri and Alexa illustrate that AI can mimic human interaction with reasonable effect, not just offer an awkward imitation of it (Čaić et al., 2018). Robotics research indicates that incorporating human-like nonverbal behaviors into robots can improve their likability (Rosenthal-von der Pütten et al., 2018). Conversely, firms must be aware that any service failures in self-service technologies can lead to long-term concerns with customer loyalty and, consequently, firm performance.

Automation capabilities may still be a poor substitute for a good human conversation, but they are rapidly improving. As AI improves, the customer interaction part of customer interaction uncertainty strategy will change, such that human interaction may not be the competitive draw that it once was. On the other hand, we seem to be approaching a conception of human attention as luxury, as shown by the use of terms such as “artisanal.” As human interaction becomes optional, it becomes a strategic choice in even starker terms. Where service firms compete on the basis of human interaction, they need to ensure that there is sufficient competitive distance between themselves and the bots.

Artificial intelligence and service breadth

AI technologies have an impact with regard to the second segment of customer interaction uncertainty, too: the breadth of the service offering. Here, humans and human capital have historically had a leg up while AI is still limited by its inputs. Using the three forms of AI suggested by Metaxiotis and colleagues (2003), one can explore the impact on using AI where the firm strategically opts to offer a wide range of services to customers. Expert systems are, by definition, expert in a particular domain. They are expert where they have data and in the contexts their algorithms were developed for. Similar constraints exist with neural networks, which rely on a particular data set and must be “trained” to recognize and predict based on new sets of data. AI (or, more precisely, machine learning) systems using neural networks do learn based on the data in those particular networks, but they are of no use outside of the data from those networks or the general strictures of similar problems. Machine learning systems do not learn in the broad, open way that humans do – they only learn from discrete data sets. Autonomous agents are dependent on the data and domain in which they are built to operate in.

When service firms offer a wide array of services, they are apt to offer services in realms beyond what any single AI system can work with and will still need a lot of human capital. Banks, for example, are using AI to evaluate credit and to make investment decisions for clients, but still need humans that can speak to clients about both subjects. Even if a firm used different AI functions for different domains, they would, at least at this point, need human capital to navigate between and across AI systems. Also, in direct consultation with a client, there is obviously no time to pull up a new system, compile the data, train it, etc., to respond to a client's query. The limitations of organizational capital in terms of managing uncertainty that were suggested by Galbraith still apply, even with AI.

P3a.

Service firms engaged in a wide array of domains will be associated with high levels of human capital.

As noted, though, firms do not simply choose automation or human workers to enact services. They can and do use both. As Van Doorn et al. (2017) suggest, the increasing infusion of technology (including automation) in frontline experiences will lead to situations where customers will feel the presence of these automated social entities, even when dealing with humans. And, where firms operate across a range of domains, automation can still provide the efficiency, quality and asset protection benefits that make organizational attractive relative to human capital.

Hybrid models using both forms of intellectual capital instinctively make sense, given the value seen in both forms. Often, firms find sustained success with bundled skills and capabilities (Winter, 2003), and the skillful use of humans and automation (including AI) together is apt to be an example. PSFs, in particular, will need to be able to possess dynamic capabilities, i.e. the ability “to integrate, build, and reconfigure internal and external competencies to address rapidly changing environments” (Teece et al., 1997, p. 516). One would expect that these two forms of organizational knowledge would augment and complement each other. For example, AI can be both efficient and effective in medical diagnostics, but no patient would want to sacrifice a good doctor's bedside manner for a printout that tells them of a prognosis. Physicians and hospitals would want to use both, as would many firms.

P3b.

Human capital and service automation, including AI, will interact to improve performance in service firms engaged in a wide array of domains.

The paper now turns to ideas regarding service automation that speak more directly to the use of AI itself. The model as it relates to the use of AI can be seen in Figure 2.

Another conception of dynamic capabilities is that they are a matter of building second-order capabilities – having the capability to build capabilities (Collis, 1994). Service firms learning to create and build hybrid service delivery models using both humans and AI could constitute such a capability. A key quality that AI possesses that is different from earlier forms of automation is that it is a form of organizational capital that learns itself (Lee, 2018). This will have value for firms that compete on the basis of information asymmetry (e.g. Nayyar, 1993): PSFs.

PSFs exist to solve cognitive problems that their customers cannot solve (McGrath, 2005; Von Nordenflycht, 2010). They compete on the basis of their intellectual capital, and because they are deploying their knowledge with customers on an ongoing basis, would naturally transmit knowledge to customers, whether they want to or not. Yet, maintaining a positive information asymmetry is critical to maintaining customer business (Nayyar, 1993). AI can help with this need for ongoing learning, as it is able to scan and process vast amounts of data in productive ways. AI may learn by being trained on data sets provided by and at the direction of humans, but it learns, nonetheless (Agrawal et al., 2017). For firms that need to continually improve their stock of knowledge, AI will provide a method to build that knowledge. Humans can learn from AI, learn to use AI and learn to refine and improve their use of AI, such that human capital and AI (organizational capital) can build each other up, akin to Nonaka's (1994) “knowledge spiral.” This second-order capability (e.g. Collis, 1994) could be a powerful competitive advantage for PSFs.

Building a knowledge spiral with AI would, however, be a challenging set of bundled resources and capabilities. This is substantially more difficult than simply using an off-the-shelf set of machine learning tools, for example, in a service bot. Not every firm would want, need or be able to have such a complex capability. For PSFs, and particularly those that cater to other PSFs, where information asymmetry is critical, this type of an AI/human capital knowledge spiral would be worth investing in, so as to maintain their knowledge advantage over their clients. Algorithmic trading, for example, can show human traders where they are missing key information, and human traders could then provide additional nuance to the algorithms. This constitutes an additional contingency in the use of AI in service automation.

P4.

Complementary, complex skills wherein human capital and AI build each other to increase the firm's knowledge stock of both forms of intellectual capital will provide sustained competitive advantages to PSFs, particularly those in business-to-business PSF sectors.

Service effectiveness between AI and humans

Organizational capital is a more efficient form of knowledge deployment. This has been said of AI, as well – it reduces cost (Agrawal et al., 2017). It has also been noted that service automation reduces variance in outcomes, as it can be more precise than humans. Even so, AI holds out the promise of improving upon human capital in areas other than efficiency and adherence to standards. It may be more effective in providing solutions as a knowledge and information processing function itself.

This possibility exists because such technologies offer completely rational decision-making based solely upon the rules they have learned. Humans engage in decision-making processes that can be substantially flawed in two ways that AI avoids. First, humans decide using bounded rationality (Simon, 1991). They satisfice, meaning they select the nearest and easiest solution that meets their level of aspiration for the outcome. This is taking the easy way out of a quandary, not finding the best possible solution. AI machines do not care how much effort it takes to explore all possible solutions, and they can avoid satisficing with their ability to analyze far greater numbers of alternatives in far less time than humans can. They will not take the easy path, but instead can consider all of the alternatives provided them.

Second, humans suffer from cognitive biases in decision-making (Kahnemann and Tversky, 1979). Humans make probabilistic decisions in suboptimal ways. This is particularly apt in comparison to AI, as current AI technologies are increasingly using probabilities and Bayesian decision-making (Parkes and Wellman, 2015). They do so with complete objectivity and based on pure mathematics, i.e. without the biases humans use. AI algorithms have no risk preferences unless they are coded in. They do not dislike loss more than they like a commensurate gain unless we tell them to.

AI-based automation that avoids human cognitive pitfalls may provide increased value because of it. This is not to say that machines are preferable overall, but this benefit can add value for clients of PSFs in particular, given the sophistication of their challenges. Examples in finance and medicine abound – two areas of particular growth in AI applications. This is also relevant given the argument that knowledge workers have the ability to learn from algorithms: AI can potentially teach humans to avoid bias.

P5.

Service automation using AI will be favored over human capital in arenas where cognitive human biases are harmful.

Lastly, an important but ironic aspect of AI where it is potentially disadvantageous vs human capital stems from the fact that these systems and algorithms can change quickly and radically when new data, programs, objectives or structures are put in place. This can be done even without human intervention. Today's AI decisions may not resemble yesterday's to an extent that is not present with human decisions.

While this may have benefits for efficient transitions, it also means that trust and reputation are potentially ephemeral (Parkes and Wellman, 2015). When AI acts an agent, that agent can alter its behavior dramatically in short order. Human behavior is much less likely to do so, with implications for trust. The nature of trust and the purpose of reputation is that we can recognize past decisions and behaviors so that future tasks can be handed to an agent with little concern that the agent will not act in our interest (Adler, 2001). Trust is built on history and an implicit assumption that history will repeat itself. That implicit assumption is considerably less valid for an AI algorithm than it is for a human. An AI-based algorithm that informed inventory and supply chain decisions, for example, might react sharply to a new supplier in a way that disrupted long-standing relationship, damaging trust and relationships that could be relied on in difficult times.

Reputation and trust are critical for PSFs, where clients must entrust firms with sensitive information and where the nature of the work makes assessment difficult (Nayyar, 1993; Alvesson, 2000; Gluckler and Armbruster, 2003). PSFs play a role in the spread and perpetuation of institutional norms, too (Nair et al., 2016), another area where trust is important. Where trust, reputation and similar areas are of particular importance to clients, human capital will be an important form of intellectual capital.

P6.

Where trust and reputation are of great importance to clients, human capital will be more likely to be used over service automation technologies such as AI.

Discussion

This study has used the concepts of intellectual capital, customer interaction uncertainty and information processing theory in combination to outline a model of service automation. In addition, the model leads to propositions regarding service automation with AI. While service automation is an old endeavor, many service firms are currently making investments in AI and adopting new technologies to provide services previously provided only by humans. Unfortunately, there has been little academic attention paid to how these firms should make these decisions. The article seeks to fill this important gap in our understanding of the use of AI and service automation in general. In brief, the argument has been that, though organizational capital (service automation) is preferred to human capital in general, firms choosing to allow more uncertainty into their processes from customer interactions will need to use more human capital and less automation, including AI. AI will also work with human capital in firms, with added importance in PSFs, as it helps them to increase their intellectual capital to better perpetuate their information asymmetry advantage vs clients. Hybrid models using both automation (including AI) and human service workers are likely to become increasingly important, too, in all firms where AI is valuable.

In light of this model, the paper has further explored the ways in which AI would impact service firms and PSFs, in particular. AI has unique aspects that are important and especially pertinent to PSFs, given that human capital has traditionally been their core resource. AI also may be better at exercising judgment than some humans, though it may also be problematic in areas where trust is particularly important.

The model has treated the decisions to adopt AI (or not) as morally neutral and driven by considerations of strategy and performance. However, it should not go unremarked upon that the widespread adoption of AI by doctors, architects, engineering firms and lawyers (among others) will work a significant and unpredictable change on society (Davenport and Ronanki, 2018; Parry et al., 2016). While uncertainty for any particular firm might decrease, uncertainty at a societal level might increase. Firms, in the end, are likely to make the decisions they believe will make them more profitable and more sustainable. A larger conversation among academics and professionals about the extent to which this makes sense from a broader point of view would be valuable.

This paper makes several important contributions. First, it has explicated contingencies for using the different forms of intellectual capital in service firms, moving beyond the “more is better” ethos of much of the knowledge-based view literature. It is not always best to have as much automation and/or human capital as possible; a study of 87 countries and territories indicates the discomfort of simply embracing large-scale automation without considering where they would be deployed (Ivanov and Webster, 2019). Prior to this study, little scholarly attention had been paid to when and where to automate. Yet, this paper's arguments illustrate that automation is a choice that is contingent upon a clear strategy based on customer interaction and knowledge resources.

The paper's model better reflects the pertinent costs and investments required to acquire and deploy knowledge resources. It also extends the knowledge-based view literature by better reflecting the different value generating abilities of the two forms of knowledge. Second, the paper has tied knowledge resources to service firm strategy by linking knowledge deployment to customer interaction uncertainty. This extends the PSF literature in particular, which to date has been heavily focused on description and classification or an exploration of human capital and partnerships. The model also suggests how human capital and organizational capital (including AI) can complement each other, as a bundle of resources and capabilities. Lastly, the model extends the organizational knowledge literature by bringing contemporary technology – AI – into the discussion of knowledge management and service automation. Organizational scholarship on this topic to date is scant, and much of it has lacked theoretical basis. By tying AI into these theories and this model, the paper points the way for both research and practice with regard to its use as a strategic knowledge resource.

The model and propositions can also be of use to managers of service firms. First, by describing how customer interaction strategy works with automation decisions, it illustrates how firms should make choices about how and when to automate – automation is not universally the best choice. The paper has discussed the two elements of customer interaction uncertainty as well, and how they impact automation decisions. These elements – the extent of customer interaction and the breadth of the service offering – are well understood by managers of service firms, just as the elements of automation are. Thus, the model can be readily operationalized by managers.

The paper has also suggested to managers how the two primary forms of knowledge – organizational and human capital – can work together to improve performance. Automation and human workers, as noted, can and should be used together, based on careful consideration of the firm's strategy. Lastly, by exploring the unique aspects of AI, the model suggests how and where it may be used most effectively. This should serve as a useful guide for managers who may be bewildered by the suggestions of AI as a potent panacea and/or a mortal threat for any business situation. These arguments are particularly important for managers of PSFs. Again, the academic literature on these firms has focused largely on their use of human capital, but current technologies such as AI are now fundamentally changing the management of these firms. The model in this paper will offer very useful suggestions as to how it should do so.

This paper is conceptual and exploratory, which leaves much room for further research. Empirical testing of the basic model for automation across a variety of service firms would be valuable work. Testing the propositions can be done, as scales have been established for the intellectual capital and customer interaction uncertainty constructs (e.g. Skaggs and Youndt, 2004). In particular, tests of the human/organizational capital contingency based on customer interaction uncertainty would be useful, including a structural model including performance. Ideally, such work would span several service industries and automation technologies.

Empirical testing of the propositions regarding AI would be of great value at this early stage of its deployment. Qualitative work regarding the interaction of human capital with AI technologies (P4) would answer important questions, particularly for managers. Both qualitative and quantitative empirical work regarding AI deployment and firms emphasizing trust and bias (P5 and 6) could be done by testing in different industries and/or different tasks within a given service arena. In all of this work, data from customers on the effectiveness of the service vs the firm's service strategy would provide important insight into the contingent choices the model proposes. With PSFs and knowledge work being such critical parts of the global economy and workforce, empirical study of the model in those areas would be extremely important. AI and similar technologies “threaten employment and will likely displace significant parts of the workforce” (George, et al., 2016, p. 1880), and for that reason alone, we need greater understanding of how to proceed.

In addition, service concepts are also important for manufacturers, such as in linking elements of a supply chain (Grönroos and Helle, 2010; Zhong, et al., 2016), so study of the framework in manufacturers would be productive as well. Supply chain problems are complex and multi-dimensional, so AI technologies hold much promise, but the degree of complexity argues for the need for human judgment as well. Lastly, the discrete categorizations of intellectual capital in the paper are a simplification that can mask how they may interact. This interaction is critical and would be another useful area for further research to explore. Firms clearly must combine human capital with service automation technologies to effectively compete.

Automating services is not a simple decision best made by comparing the costs of a computer program to the cost of a human employee, nor should firms simply eschew automation to protect the jobs of their highly skilled professional employees. Instead, automation is a strategic decision. First, firms must consider the types of knowledge involved. Is this moving organizational capital from one form to another; say embedding an employee handbook into an expert system? Or, is it a way of converting human knowledge (often subtle, nuanced, but subject to exit) into organizational knowledge? Then, what is the overall strategic approach of the firm to customer interaction and the uncertainty it introduces into the production of the service? Is this a firm that has decided to limit customer choice in favor of efficiency? Or, is it a firm that caters to customer choice? Is there value to the customer in efficient, predictable service, or does the customer value the variety that comes with human interaction? Service firms do not use knowledge without regard to strategy. They seek to solve customer problems profitably using knowledge resources, and they use automation to work more efficiently and effectively. This model should help firms and scholars think through the strategic implications of automation decisions.

Figures

Basic service automation model

Figure 1

Basic service automation model

Service automation model using AI

Figure 2

Service automation model using AI

References

Adler, P. (2001), “Market, hierarchy, and trust: the knowledge economy and the future of capitalism”, Organization Science, Vol. 12, pp. 215-234.

Agrawal, A., Gans, J. and Goldfarb, A. (2017), “What to expect from artificial intelligence”, MIT Sloan Management Review, Vol. 58 No. 3, p. 23.

Alvesson, M. (2000), “Social identity and the problem of loyalty in knowledge‐intensive companies”, Journal of Management Studies, Vol. 37, pp. 1101-1124.

Becerra-Fernandez, I. and Sabherwal, R. (2014), Knowledge Management: Systems and Processes, Routledge, New York, NY.

Bontis, N. (1998), “Intellectual capital: an exploratory study that develops measures and models”, Management Decision, Vol. 36, pp. 63-76.

Čaić, M., Odekerken-Schröder, G. and Mahr, D. (2018), “Service robots: value co-creation and co-destruction in elderly care networks”, Journal of Service Management, Vol. 29 No. 2, pp. 178-205.

Collis, D. (1994), “Research note: how valuable are organizational capabilities?”, Strategic Management Journal, Vol. 15, pp. 143-152.

Davenport, T. and Ronanki, R. (2018), “Artificial intelligence for the real world”, Harvard Business Review, Vol. 97 No. 1, pp. 108-116.

De Keyser, A., Köcher, S., Alkire, L., Verbeeck, C. and Kandampully, J. (2019), “Frontline Service Technology infusion: conceptual archetypes and future research directions”, Journal of Service Management, Vol. 30 No. 1, pp. 156-183.

Dechter, R. (1986), “Learning while searching in constraint-satisfaction-problems”, Proceedings of the 5th National Conference on Artificial Intelligence, Philadelphia, PA, August 11-15, 1986, Vol. 1.

Empson, L. (2001), “Introduction: knowledge management in professional service firms”, Human Relations, Vol. 54, pp. 811-817.

Galbraith, J. (1974), “Organization design: an information processing view”, Interfaces, Vol. 4, pp. 28-36.

George, G., Howard-Grenville, J., Joshi, A. and Tihanyi, L. (2016), “Understanding and tackling societal grand challenges through management research”, Academy of Management Journal, Vol. 59 No. 6, pp. 1880-1895.

Gittell, J. (2002), “Coordinating mechanisms in care provider groups: relational coordination as a mediator and input uncertainty as a moderator of performance effects”, Management Science, Vol. 48, pp. 1408-1426.

Gluckler, J. and Armbruster, H. (2003), “Bridging uncertainty in management consulting: the mechanisms of trust and networked reputation”, Organization Studies, Vol. 24, pp. 269-297.

Goodall, H.L. Jr (1992), “Empowerment, culture, and postmodern organizing: deconstructing the Nordstrom employee handbook”, Journal of Organizational Change Management, Vol. 5 No. 2, pp. 25-30.

Grant, R. (1996), “Toward a knowledge-based theory of the firm”, Strategic Management Journal, Vol. 17, pp. 109-122.

Grewal, D. (2014), “A critical conceptual analysis of definitions of artificial intelligence as applicable to computer engineering”, Journal of Computer Engineering, Vol. 16 No. 2, pp. 09-13.

Grönroos, C. and Helle, P. (2010), “Adopting a service logic in manufacturing: conceptual foundation and metrics for mutual value creation”, Journal of Service Management, Vol. 21 No. 5, pp. 564-590.

Harari, Y.N. (2018), Twenty-one Lessons for the 21st Century, Spiegel & Grau, New York.

Hinton, G. (2007), “Learning multiple layers of representation”, Trends in Cognitive Sciences, Vol. 11 No. 10, pp. 428-434.

Hinton, G. and Sejnowski, T. (1999), Unsupervised Learning: Foundations of Neural Computation, MIT Press, Cambridge, MA.

Ivanov, S. and Webster, C. (2018), “Adoption of robots, artificial intelligence and service automation by travel, tourism and hospitality companies – a cost-benefit analysis”, in Marinov, V., Vodenska, M., Assenova, M. and Dogramadjieva, E. (Eds), Traditions and Innovations in Contemporary Tourism, Cambridge Scholars Publishing, Cambridge, pp. 190-203.

Ivanov, S. and Webster, C. (2019), “What should robots do? A comparative analysis of industry professionals, educators and tourists”, Information and Communication Technologies in Tourism 2019, Springer, pp. 249-262.

Jaakkola, E., Helkkula, A. and Aarikka-Stenroos, L. (2015), “Service experience co-creation: conceptualization, implications, and future research directions”, Journal of Service Management, Vol. 26 No. 2, pp. 182-205.

Kahneman, D. and Tversky, A. (1979), “Prospect theory: an analysis of decision under risk”, Econometrica, Vol. 47, pp. 263-291.

Lacity, M.C. and Willcocks, L.P. (2016), “A new approach to automating services”, MIT Sloan Management Review, Vol. 58 No. 1, pp. 41-49.

Larsson, R. and Bowen, D. (1989), “Organization and customer: managing design and coordination of services”, Academy of Management Review, Vol. 14, pp. 213-233.

Lee, K. (2018), AI Superpowers: China, Silicon Valley, and the New World Order, Houghton Mifflin Harcourt Publishing Company, New York.

Lovelock, C. (1984), “Whither services marketing?: in search of a new paradigm and fresh perspectives”, Journal of Service Research, Vol. 7, pp. 20-41.

McGrath, R. (2005), “Thinking differently about knowledge-intensive firms: insights from early medieval Irish monasticism”, Organization, Vol. 12, pp. 549-566.

Metaxiotis, K., Ergazakis, K., Samouilidis, E. and Psarras, J. (2003), “Decision support through knowledge management: the role of the Artificial Intelligence”, Information Management and Computer Security, Vol. 11 No. 5, pp. 216-221.

Meyer, C., Cohen, D. and Nair, S. (2017), “Some have to and some want to: why firms adopt a post-industrial form”, Journal of Management and Governance, Vol. 21 No. 2, pp. 533-559.

Meyer, C., Skaggs, B. and Youndt, M. (2015a), “Developing and deploying organizational capital in services vs Manufacturing”, Journal of Managerial Issues, Vol. 4, pp. 326-344.

Meyer, C., Skaggs, B., Nair, S. and Cohen, D. (2015b), “Customer interaction uncertainty, knowledge, and service firm internationalization strategies”, Journal of International Management, Vol. 3.

Morris, T. (2001), “Asserting property rights: knowledge codification in the professional service firm”, Human Relations, Vol. 54, pp. 819-838.

Nair, S., Cohen, D. and Meyer, C. (2016), “The role of professional service providers during the initial stages of international entrepreneurship: a neo-institutionalist view”, International Journal of Entrepreneurship and Small Business, Vol. 30 No. 4, pp. 526-544.

Nayyar, P. (1993), “Stock market reactions to related diversification moves by service firms seeking benefits from information asymmetry and economies of scope”, Strategic Management Journal, Vol. 14, pp. 569-591.

Nonaka, I. (1994), “A dynamic theory of organizational knowledge creation”, Organization Science, Vol. 5, pp. 14-37.

Ofek, E. and Sarvary, M. (2001), “Leveraging the customer base: creating competitive advantage through knowledge management”, Management Science, Vol. 47, pp. 1441-1456.

Parkes, D.C. and Wellman, M.P. (2015), “Economic reasoning and artificial intelligence”, Science, Vol. 349, pp. 267-272.

Parry, K., Cohen, M. and Bhattacharya, S. (2016), “Rise of the machines: a critical consideration of automated leadership decision making in organizations”, Group and Organization Management, Vol. 41 No. 5, pp. 1-24.

Rosenthal-von der Pütten, A.M., Krämer, N.C. and Herrmann, J. (2018), “The effects of humanlike and robot-specific affective nonverbal behavior on perception, emotion, and behavior”, International Journal of Social Robotics, Vol. 10, pp. 569-582.

Samuelson, P. (1954), “The pure theory of public expenditure”, The Review of Economics and Statistics, Vol. 36, pp. 387-389.

Simon, H.A. (1978), “Information-processing theory of human problem solving”, in Estes, W.K. (Ed.), Handbook of Learning and Cognitive Processes: Human Information Processing, Vol. 5, Psychology Press, New York, pp. 271-295.

Simon, H. (1991), “Bounded rationality and organizational learning”, Organization Science, Vol. 2, pp. 125-134.

Skaggs, B. and Huffman, T. (2003), “A customer interaction approach to strategy and production complexity alignment in service firms”, Academy of Management Journal, Vol. 46, pp. 775-786.

Skaggs, B. and Youndt, M. (2004), “Strategic positioning, human capital, and performance in service organizations: a customer interaction approach”, Strategic Management Journal, Vol. 25, pp. 85-99.

Spender, J.C. and Scherer, A. (2007), “The philosophical foundations of knowledge management: editors' introduction”, Organization, Vol. 14, pp. 5-28.

Strauss, B. (2017), “The Rise and Fall of the Automat”, available at: https://www.thoughtco.com/the-rise-and-fall-of-the-automat-4152992 (accessed 7 January 2019).

Tansik, D.A. (1990), “Balance in service systems design”, Journal of Business Research, Vol. 20, pp. 55-61.

Teece, D.J., Pisano, G. and Shuen, A. (1997), “Dynamic capabilities and strategic management”, Strategic Management Journal, Vol. 18 No. 7, pp. 509-533.

Van Doorn, J., Mende, M., Noble, S.M., Hulland, J., Ostrom, A.L., Grewal, D. and Petersen, J.A. (2017), “Domo arigato Mr Roboto: emergence of automated social presence in organizational frontlines and customers' service experiences”, Journal of Service Research, Vol. 20 No. 1, pp. 43-58.

Von Nordenflycht, A. (2010), “What is a professional service firm? Toward a theory and taxonomy of knowledge-intensive firms”, Academy of Management Review, Vol. 35, pp. 155-174.

Winter, S. (2003), “Understanding dynamic capabilities”, Strategic Management Journal, Vol. 24, pp. 991-995.

Wirtz, J. and Zeithaml, V. (2018) Cost-effective service excellence, Journal of the Academy of Marketing Science, Vol. 46, pp. 59-80.

Wirtz, J., Patterson, P., Kunz, W., Gruber, T., Lu, V.N., Paluch, S. and Martins, A. (2018), “Brave new world: service robots in the frontline”, Journal of Service Management, Vol. 29, pp. 907-931.

Youndt, M. and Snell, S. (2004), “Human resource configurations, intellectual capital, and organizational performance”, Journal of Managerial Issues, Vol. 16, pp. 337-360.

Zhong, R.Y., Newman, S.T., Huang, G.Q. and Lan, S. (2016), “Big Data for supply chain management in the service and manufacturing sectors: challenges, opportunities, and future perspectives”, Computers and Industrial Engineering, Vol. 101, pp. 572-591.

Further reading

Stewart, T.A. (1997), Intellectual Capital, Doubleday-Currency, New York, NY.

Corresponding author

Chris Meyer can be contacted at: meyerc6@rpi.edu

Related articles