Humanizing the Machine: Anthropomorphism and the Uncanny Valley in AI-Mediated Service Recovery

Humanizing the Machine: Anthropomorphism and the Uncanny Valley in AI-Mediated Service Recovery | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Humanizing the Machine: Anthropomorphism and the Uncanny Valley in AI-Mediated Service Recovery Arpandeep Kaur, Jaskirat Singh This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6889879/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract As sophisticated Artificial Intelligence takes on the critical task of service recovery, the design strategy of anthropomorphism presents both a compelling opportunity and a significant risk. An imperfectly human AI can provoke an “uncanny valley” response, turning a recovery attempt into a more alienating experience. Integrating justice theory with social presence and expectancy violation frameworks, our two experimental studies (N = 500; N = 800) dissect this crucial tension. Our results reveal a key fragility: the positive influence of behavioral and emotional anthropomorphism on justice and affinity perceptions is entirely contingent on flawless execution. Minor glitches in highly human-like agents negate these benefits by violating customer expectations of authentic interaction. This research offers a robust theoretical model of the contingent nature of human-like AI design and provides clear, actionable principles for creating service agents that enhance, rather than undermine, customer relationships during recovery. Artificial Intelligence and Machine Learning Marketing Management International Business Other Business Artificial Intelligence Service Recovery Anthropomorphism Uncanny Valley Justice Theory Human-AI Interaction Expectancy Violation Theory Figures Figure 1 1. Introduction The landscape of service interaction is undergoing a seismic transformation, driven by the proliferation of increasingly autonomous and sophisticated Artificial Intelligence (AI). Once relegated to simple automation, AI agents—from chatbots to virtual assistants—are now deployed at the most critical and emotionally charged of customer touchpoints: service recovery. This moment of truth, where a firm’s response to a failure can either salvage or irrevocably shatter a customer relationship, is now frequently entrusted to a machine (Huang and Rust, 2018 ; Xiao and Kumar, 2021 ). In an effort to preserve the relational quality of these encounters, firms increasingly pursue a strategy of anthropomorphism, designing AI agents with human-like appearances, conversational styles, and expressions of empathy (Epley, Waytz, and Cacioppo, 2007 ). The underlying assumption is that a more human-like AI will engender greater social presence, fostering the trust and satisfaction necessary for successful recovery (van Doorn et al., 2017 ). Yet, this pursuit of the perfectly human-like machine walks a razor’s edge, bordering the unsettling phenomenon of the “uncanny valley,” where near-human entities provoke not affinity, but a sense of unease and disquiet (Mori, MacDorman, and Kageki, 2012 ). An AI agent that appears almost—but not quite—human risks violating customer expectations in a manner that can compound the initial service failure, turning a moment of recovery into one of profound alienation. This central paradox—that the very attempt to humanize an AI for service recovery may catastrophically backfire—presents a significant theoretical and practical challenge. Current scholarship offers potentially conflicting guidance. On one hand, a robust body of work extols the virtues of anthropomorphism for elevating the perceived warmth and relational quality of service agents (Belanche, Casaló, Flavián, and Schepers, 2020 ). On the other, human-computer interaction research provides stark warnings about the risks of imperfect human-likeness (Ho and MacDorman, 2017 ). What remains critically unclear is the tipping point: when and how does the laudable pursuit of a human-like design cross the precipice into the uncanny? The core theoretical mechanisms governing this interplay in the high-stakes context of service recovery remain underdeveloped. While the tenets of justice theory are foundational to service recovery—positing that customers seek fairness in outcomes, processes, and interactions (Tax, Brown, and Chandrashekaran, 1998 )—it is a profound theoretical leap to assume they function identically when the interactional partner is a machine. The very essence of "interactional justice" and "emotional affinity" is thrown into question when one party possesses no genuine emotions or consciousness. We must therefore scrutinize the process by which a designed humanness translates—or fails to translate—into these crucial mediating outcomes. This investigation addresses several crucial gaps in the literature. First, we move beyond treating anthropomorphism as a monolithic concept to examine how its distinct dimensions—physical, behavioral, emotional, and cognitive—differentially influence customer perceptions. We integrate this with an analysis of uncanny valley triggers to understand the conditions under which human-like cues become liabilities. Second, we dissect the underlying process, examining the mediating roles of justice perceptions and emotional affinity to explain why these AI design elements affect downstream outcomes like recovery satisfaction, trust, and repatronage intentions. Third, we recognize that customer responses are not uniform. We therefore clarify crucial boundary conditions by investigating how individual differences, namely customers' technological savviness and their prior relationship with the firm, moderate these effects, providing a more complete picture of for whom AI-mediated recovery is most effective. To address these questions, this paper develops and empirically tests a comprehensive framework through two robust experimental studies. Our primary contribution is to illuminate the inherent fragility of anthropomorphism’s benefits in service recovery. We demonstrate that while specific human-like cues can powerfully foster perceptions of justice and build emotional rapport, these gains are precarious and can be completely undone by subtle design flaws that trigger uncanny valley responses. In doing so, this research (1) contributes a novel contingency framework to service theory, revealing how the interplay between anthropomorphism and the uncanny valley reconfigures the established roles of justice and affinity; (2) offers a multidimensional view of AI anthropomorphism, identifying the specific facets most critical for successful recovery; and (3) clarifies key boundary conditions that shape the efficacy of AI service agents. The insights generated have significant implications for both theory and practice, providing a crucial roadmap for designing AI agents that can foster positive experiences when resolving the most challenging service failures. 2. Literature Review and Hypotheses Development 2.1. AI in Service and the Imperative of Relational Recovery Far beyond its origins in simple automation, artificial intelligence now performs sophisticated service functions that can supplement or wholly substitute for human employees (Huang and Rust, 2021). This technological ascent places AI at the forefront of a firm’s most sensitive encounters, particularly the emotionally charged context of service recovery, where consumer acceptance is paramount (Marinova et al., 2017). The bedrock of traditional recovery—a process vital for rebuilding trust and loyalty—has always been the delivery of justice in its three forms: distributive, procedural, and interactional (Tax et al., 1998; Smith, Bolton, and Wagner, 1999). It is precisely this third pillar of interactional justice, rooted in empathy and respect, where AI’s capabilities are most contested. These agents are tasked with emulating a relational warmth they do not inherently possess (De Melo et al., 2014; Xiao and Kumar, 2021), creating the defining challenge for their design: how to reconcile the cold logic of procedural efficiency with the necessary illusion of interpersonal understanding. 2.2. Anthropomorphism and the Cultivation of Social Presence Firms endeavor to close this relational divide through anthropomorphism, a strategy of imbuing agents with human traits, emotions, or intentions (Epley et al., 2007). This is a well-established approach in human-computer interaction, designed to fundamentally alter user perceptions by cultivating social presence—the psychological feeling of "being with" another (Nowak and Biocca, 2003; Biocca, Harms, and Burgoon, 2003). When successful, this enhanced social presence prompts users to interact with the AI not as a mere tool, but as a social counterpart, applying the same norms and heuristics that govern human relationships (Reeves and Nass, 1996). The means of achieving this social presence are, however, varied and distinct. Recent work identifies separate physical, behavioral, emotional, and cognitive dimensions of anthropomorphism (Złotowski et al., 2015). An agent's behavioral and emotional expressiveness, for example, directly speaks to its perceived empathy and thus powerfully shapes judgments of interactional justice. Likewise, its cognitive acuity signals competence, which in turn bolsters evaluations of procedural and distributive fairness. When an AI is perceived as both a caring and capable partner in the recovery process, it is also more likely to engender emotional affinity. From this theoretical foundation, our initial hypotheses emerge: H1: Anthropomorphism (across its physical, behavioral, emotional, and cognitive dimensions) positively influences customers’ justice perceptions following a service failure. H2: Anthropomorphism (across the same dimensions) positively influences customers’ emotional affinity toward the AI following a service failure. 2.3. The Uncanny Valley as an Expectancy Violation While anthropomorphism offers clear benefits, the strategy is inherently perilous. As a humanoid entity approaches but fails to perfectly replicate human appearance and behavior, observers can experience a sharp dip in affinity, a phenomenon famously termed the uncanny valley (Mori et al., 2012). These feelings of eeriness and discomfort arise from minor yet unsettling imperfections—a stilted blink, a disjointed phrase, a subtly unnatural facial expression—that betray the entity’s artifice (Ho and MacDorman, 2017). Expectancy Violation Theory (EVT) provides a robust theoretical lens through which to understand this reaction (Burgoon and Jones, 1976). EVT posits that individuals hold socially ingrained expectations for behavior in any given interaction. When another party’s actions breach these expectations, a violation occurs, which is then appraised for its valence (positive or negative). In the context of a highly anthropomorphic AI, the agent’s human-like features set a strong expectation of a seamless, natural, and authentically human-like interaction. Uncanny valley triggers—such as eeriness , a lack of realism in appearance or movement, or unpredictability in logic or response (Saygin et al., 2012)—constitute a stark negative violation of this expectation. This mismatch between the anticipated human-like coherence and the observed mechanical quirk creates a profoundly negative affective response. In a service recovery context, this adverse reaction can contaminate judgments of the interaction, directly diminishing perceptions of fairness and eroding any nascent emotional affinity. We therefore hypothesize: H3: Uncanny valley triggers (eeriness, lack of realism, unpredictability) negatively influence customers’ justice perceptions following a service failure. H4: Uncanny valley triggers (eeriness, lack of realism, unpredictability) negatively influence customers’ emotional affinity toward the AI following a service failure. Furthermore, we propose that the level of anthropomorphism sets the very conditions for a potential uncanny valley experience. An AI that is obviously a machine creates no expectation of human-like fidelity; its errors are perceived merely as technical failures. In contrast, a highly anthropomorphic AI creates a high-stakes interaction by implicitly promising human-like congruence. When that promise is broken by an imperfect cue, the resulting expectancy violation is magnified. This suggests a perilous interaction where the potential for disappointment escalates with the degree of attempted humanization. This logical tension leads to a cornerstone proposition: H5: Greater levels of anthropomorphism will amplify the negative effect of imperfect cues, leading to a stronger elicitation of uncanny valley responses and consequently diminished customer perceptions. 2.4. The Mediating Pathways to Recovery Outcomes Justice theory posits that fairness perceptions are a primary determinant of customer satisfaction following service recovery (Colquitt, 2001; Smith et al., 1999). When customers perceive the outcome, process, and interaction as fair, their overall evaluation of the recovery experience is enhanced. Similarly, the development of emotional affinity—a positive relational bond with the service agent—has been shown to soften negative reactions and improve recovery evaluations (Van Doorn et al., 2017). We expect these principles to hold firmly in AI-mediated contexts: H6: Justice perceptions positively influence recovery satisfaction in AI-mediated service recovery. H7: Emotional affinity toward the AI positively influences recovery satisfaction in AI-mediated service recovery. The significance of recovery satisfaction extends far beyond the immediate resolution of a complaint, shaping critical downstream attitudes and behaviors (Maxham and Netemeyer, 2002). Specifically, a customer who is satisfied with an AI-mediated interaction is far more inclined to develop trust in the technology’s capabilities for future encounters (Gefen, Karahanna, and Straub, 2003; McKnight, Choudhury, and Kacmar, 2002). This combination of satisfaction and emergent trust is what ultimately cements customer loyalty, manifesting as stronger intentions to repatronize the brand (Zeithaml, Berry, and Parasuraman, 1996). H8: Recovery satisfaction positively influences trust in the AI. H9: Recovery satisfaction positively influences repatronage intentions. 2.5. Boundary Conditions: The Moderating Roles of Customer Characteristics The effects of AI design are unlikely to be uniform across all customers. We propose two critical boundary conditions that shape these relationships: technological savviness and prior relationship strength. Technological Savviness. Individual-level comfort and familiarity with technology can fundamentally shape reactions to AI (Parasuraman, 2000). Tech-savvy users possess more developed and nuanced mental models of AI capabilities and limitations. Consequently, they may be more appreciative of sophisticated anthropomorphic features, leading to a stronger positive effect on justice and affinity perceptions. Conversely, their familiarity may render them more tolerant of minor glitches, which they are more likely to interpret as predictable system errors rather than eerie violations of humanness. Less tech-savvy customers, by contrast, may be more unsettled by near-human AI and more frustrated by its imperfections. H10: Technological savviness positively moderates the effect of anthropomorphism on justice perceptions and emotional affinity, such that the effect is stronger for more tech-savvy customers. H11: Technological savviness negatively moderates the effect of uncanny valley triggers on justice perceptions and emotional affinity, such that tech-savvy customers are less negatively affected by these triggers. Prior Relationship with the Company. A pre-existing bond of loyalty and trust with a firm can serve as a powerful buffer during service failures (De Wulf, Odekerken-Schröder, and Iacobucci, 2001). According to attribution theory (Weiner, 1985), customers with a strong prior relationship are more likely to give the firm the “benefit of the doubt,” attributing failures to temporary or external factors. This relational equity can create a halo effect that extends to the firm’s AI agents. For these customers, a satisfying recovery experience is more likely to translate into reaffirmed trust and loyalty, as it aligns with their existing positive disposition toward the firm. H12: A stronger prior relationship with the company positively moderates the effect of recovery satisfaction on repatronage intentions, such that this link is heightened when the existing relationship is stronger. (INSERT FIGURE I: CONCEPTUAL MODEL HERE) (INSERT TABLE I: HYPOTHESES SUMMARY TABLE HERE) 3. Methodology This research employs a two-study experimental design to establish the causal impact of anthropomorphism and uncanny valley triggers on customer responses in AI-mediated service recovery and to test the full scope of our conceptual model. Both experiments utilize scenario-based vignettes to ensure high internal validity while capturing the essential context of a service failure interaction. In keeping with best practices for experimental research, both studies included rigorous manipulation checks, attention checks, and measures to ensure reliability and validity. Furthermore, this research adheres to open science principles; the anonymized data and analysis scripts for both experiments will be made available on the Open Science Framework (OSF) upon publication. 3.1. Experiment 1: Core Effects of Anthropomorphism and the Uncanny Valley The primary objective of Experiment 1 was to establish the core causal effects of anthropomorphism and uncanny valley triggers on the mediating variables of justice perceptions and emotional affinity. A 2 (anthropomorphism: high vs. low) × 2 (uncanny valley: present vs. absent) between-subjects design was employed. Participants and Procedure. Our participant pool was drawn from Prolific, an online platform recognized for providing diverse and high-quality data samples suitable for academic research (Buhrmester, Kwang, and Gosling, 2011). This demographic is appropriate for this study due to its broad familiarity with online retail and digital service interactions. Five hundred participants were recruited (52% female; M_age_=38.2, SD=10), all of whom were English-speaking adults reporting at least one service failure experience in the preceding six months. This criterion ensured a degree of familiarity with the complaint context. The procedure began with a pre-scenario questionnaire capturing demographics. Participants were then immersed in a scenario where they had received an incorrect item from a fictitious online retailer, “ElectroDeals.” The failure severity was held constant across all conditions. Subsequently, they engaged in a simulated text-based chat with an AI service agent to resolve the issue, where our manipulations were embedded. Immediately following the interaction, they completed a post-scenario questionnaire containing our dependent measures and manipulation checks. Experimental Manipulations. Anthropomorphism (High vs. Low): In the high anthropomorphism condition, the AI agent was designed to maximize social presence. It featured a realistic human avatar, communicated using first-person pronouns (e.g., “I’m truly sorry this happened. I will look into this for you”), and offered personalized solutions. In contrast, the low anthropomorphism condition presented the agent as a functional tool, using a generic icon, formal phrases (e.g., “Your input has been recorded”), and standardized solutions. Uncanny Valley (Present vs. Absent): In the present condition, the interaction included subtle, pre-tested cues designed to elicit an uncanny feeling. These were not mere errors but theoretically grounded violations of human-like expectations, including a momentary, unnatural freezing of the avatar’s facial expression and instances of slight linguistic disfluency. In the absent condition, the AI interaction was seamless. The avatar in the high-anthropomorphism condition was carefully rendered with smooth, naturalistic animations, and all chat responses were coherent. Pre-testing confirmed that the “present” condition cues reliably evoked feelings of eeriness rather than simply being perceived as incompetence. 3.2. Experiment 2: Dimensionality of Anthropomorphism and Moderation The purpose of Experiment 2 was twofold: first, to provide a more fine-grained view of anthropomorphism’s multidimensional impact, and second, to test the proposed moderation effects. The study utilized a 2 (physical anthropomorphism: high vs. low) × 2 (behavioral anthropomorphism: high vs. low) × 2 (emotional anthropomorphism: high vs. low) × 2 (cognitive anthropomorphism: high vs. low) between-subjects factorial design. Participants and Procedure. A separate sample of 800 participants was recruited from Prolific (55% female; M_age_=40.1, SD=9.8). The screening criteria were slightly broadened to an age range of 18–65 to enhance generalizability, while still requiring prior experience with chatbots and service failures. The procedure mirrored that of Experiment 1, with the addition of measures for our moderator variables in the pre-scenario questionnaire. Experimental Manipulations. Each dimension of anthropomorphism was manipulated as a separate factor: Physical (realistic avatar vs. abstract icon), Behavioral (natural language vs. robotic phrasing), Emotional (empathic statements vs. no emotional cues), and Cognitive (adaptive problem-solving vs. scripted solutions). In this more complex design, uncanny valley triggers were measured via items assessing perceived "mismatch" or "creepiness," allowing us to test their effect correlationally. 3.3. Measures All primary constructs were measured using multi-item scales on 7-point Likert scales (1 = strongly disagree, 7 = strongly agree), adapted from established literature to fit the AI service recovery context. A series of Confirmatory Factor Analyses (CFAs) were conducted, and the results indicated that all constructs achieved excellent model fit (e.g., CFI > 0.90, RMSEA < 0.08) and demonstrated strong convergent and discriminant validity, confirming the robustness of our measurement approach. To mitigate potential common method bias, procedural remedies were employed, including the psychological separation of predictors and criteria and the use of attention checks. (INSERT TABLE III: MEASURES HERE) 4. Data Analysis and Results 4.1. Data Cleaning and Manipulation Checks Across both experiments, data were cleaned prior to analysis. In Experiment 1, 16 participants were removed for failing attention checks or for straight-line responding, resulting in a final sample of N = 484. In Experiment 2, 24 participants were removed, yielding a final sample of N = 776. Manipulation checks confirmed the success of our experimental designs. In Experiment 1, participants in the high-anthropomorphism condition perceived the AI as significantly more human-like (t(482) = 22.04, p < .001), and those in the uncanny valley “present” condition reported significantly higher feelings of eeriness (t(482) = 18.10, p < .001). In Experiment 2, manipulation checks for each of the four dimensions of anthropomorphism all yielded significant differences between their respective high and low conditions (all p < .001). (INSERT TABLE II: MANIPULATION CHECKS HERE) 4.2. Experiment 1: Core Effects A 2x2 ANOVA was used to test H1-H5. As predicted, we found a significant main effect of anthropomorphism on both justice perceptions (F(1, 480) = 144.92, p < .001, η² = .23), supporting H1, and emotional affinity (F(1, 480) = 186.44, p < .001, η² = .28), supporting H2. Likewise, the uncanny valley had a significant negative main effect on justice perceptions (F(1, 480) = 97.55, p < .001, η² = .17), supporting H3, and emotional affinity (F(1, 480) = 121.30, p < .001, η² = .20), supporting H4. The omnibus ANOVA test did not reveal a statistically significant interaction effect for H5. However, a planned post-hoc correlational analysis specifically within the high-anthropomorphism condition provided preliminary support for the underlying mechanism: the presence of unsettling glitches was significantly correlated with increased discomfort (r = .31, p < .05), a relationship that was absent in the low-anthropomorphism condition. (INSERT TABLE IV: EXPERIMENT 1 KEY RESULTS HERE) 4.3. Experiment 2: Full Model Test Experiment 2 tested the full conceptual model. A MANOVA assessing the dimensional effects (H1, H2) showed that behavioral and emotional anthropomorphism had the largest impacts on justice and affinity (all p < .001), while physical and cognitive dimensions exerted smaller positive effects (p < .01). Measured "mismatch" cues correlated negatively with both justice (β = -.33, p < .001) and emotional affinity (β = -.26, p < .001), re-affirming H3 and H4. Crucially, a significant interaction emerged, providing stronger support for H5: uncanny feelings were most exacerbated when high physical/emotional anthropomorphism was combined with perceptible glitches. Regression analyses confirmed the full mediation pathway. Justice perceptions (β = .43, p < .001) and emotional affinity (β = .37, p < .001) were strong predictors of recovery satisfaction (H6, H7). Recovery satisfaction, in turn, positively influenced both trust in AI (β = .42, p < .001) and repatronage intentions (β = .54, p < .001) (H8, H9). Finally, moderated regression supported our remaining hypotheses. Technological Savviness amplified the benefits of anthropomorphism (β = .18, p < .01) (H10) and diminished the harm from uncanny triggers (β = -.15, p < .05) (H11). A stronger Prior Relationship strengthened the link between recovery satisfaction and repatronage (β = .22, p < .001), supporting H12. (INSERT TABLE V: EXPERIMENT 2 KEY EFFECTS HERE) 5. General Discussion This research was motivated by a central paradox in the design of AI for service recovery: the simultaneous promise and peril of anthropomorphism. Our findings, derived from two robust experiments, offer a nuanced and comprehensive resolution to this tension, yielding significant contributions to service theory and actionable guidance for practice. We demonstrate that while humanizing AI is a powerful strategy for enhancing customer perceptions of justice and affinity, its benefits are remarkably fragile and highly contingent on both design execution and customer characteristics. 5.1. Theoretical Contributions First and foremost, this study extends service recovery theory into the era of advanced AI, revealing how the interplay of anthropomorphism and the uncanny valley reconfigures foundational tenets. Our research affirms that customers continue to seek fairness in AI-mediated recovery, reinforcing the relevance of justice theory. However, we advance this understanding by showing how AI design directly manipulates these perceptions. The finding that behavioral and emotional dimensions of anthropomorphism are especially critical (Experiment 2) highlights the primacy of interactional justice cues in technologically mediated encounters. This refines our understanding of Colquitt's (2001) framework, suggesting that when an AI agent is the interactional partner, its perceived capacity for empathy and natural conversation disproportionately shapes the entire justice evaluation. Our most significant theoretical insight, however, lies in the explication of H5. By integrating social presence and expectancy violation theories, we move beyond a simple cost-benefit analysis of anthropomorphism. We demonstrate that a highly human-like AI does not simply add benefits; it fundamentally raises the relational stakes. The design creates a strong expectation of human-like authenticity, which, when violated by even subtle imperfections, triggers a powerful uncanny valley response that negates the positive effects. This dynamic presents a novel theoretical tension for service research, suggesting a nonlinear, and at times negative, return on investments in anthropomorphic design. Second, our research contributes a multidimensional and decomposed understanding of AI anthropomorphism in a service context. By operationalizing and testing physical, behavioral, emotional, and cognitive dimensions separately (Złotowski et al., 2015), we provide a more granular and managerially relevant perspective. We show that not all human-like cues are created equal. The pronounced impact of behavioral and emotional cues suggests that for service recovery, what the AI says and how it "feels" is more important than what it looks like or how it thinks . This finding challenges a design ethos that may over-invest in visual avatar fidelity at the expense of developing more sophisticated and contextually aware conversational and emotional intelligence. Third, this paper clarifies critical boundary conditions that delineate the user- and firm-specific contingencies of AI service efficacy. In supporting H10 and H11, we show that technological savviness acts as a powerful lens through which AI interactions are interpreted. For tech-savvy customers, the benefits of anthropomorphism are amplified, and the sting of uncanny triggers is dulled. This suggests their sophisticated mental models of technology allow them to appreciate the design effort while dismissing imperfections as predictable system noise rather than unsettling social violations. Furthermore, by confirming that a strong prior firm relationship buffers recovery outcomes (H12), our findings underscore the enduring power of brand equity. A history of trust creates a powerful halo effect that not only encourages forgiveness but also strengthens the path from a satisfactory recovery to future loyalty, even when that recovery is handled by a machine. 5.2. Managerial Implications The findings of this research offer clear, evidence-based guidance for managers navigating the high-stakes implementation of AI in service recovery. The central takeaway is that designing effective service AI is not merely a technical challenge but a delicate exercise in managing customer psychology. A naive "more human is always better" strategy is not only suboptimal—it is actively dangerous. First, managers must strategically prioritize investments in behavioral and emotional anthropomorphism over purely physical or cognitive dimensions. Our findings show that natural, empathetic language and adaptive, personalized responses yield the highest return on investment in fostering justice perceptions and emotional affinity. Rather than allocating significant resources to developing hyper-realistic but potentially static avatars, firms should focus on advancing the AI's conversational intelligence. This involves designing dialogue systems that can move beyond rigid scripts to utilize natural language, conversational fillers, and personalized acknowledgements of the customer's issue and emotional state. This is where the true sense of interactional fairness is forged. Second, the risk of the uncanny valley must be proactively managed through rigorous, segment-specific pre-testing. The adage that "the last 1% is the hardest" is particularly true for AI design; near-perfect human-likeness makes any small flaw glaring and unsettling. Firms must move beyond standard quality assurance to conduct user experience testing specifically designed to detect "creepiness." This involves testing interactions with target customer segments to identify subtle triggers—be they visual, auditory, or linguistic—that provoke unease. Given that tech-savvy customers are more tolerant of such glitches, firms could consider offering different AI interaction styles. For less savvy segments, a less ambitious but more robustly coherent AI may be safer and more effective than a highly anthropomorphic agent that risks uncanny failure. Finally, managers should leverage existing brand equity as a crucial buffer when deploying service AI. Our findings confirm that a strong prior customer relationship significantly enhances the likelihood that a successful AI recovery will translate into future loyalty. This suggests that initial AI rollouts might be most effective when targeted at loyal, established customers who are more likely to grant the firm—and by extension, its technology—the benefit of the doubt. For all customer segments, transparency about the agent's AI nature is paramount, as disclosure can help manage expectations and frame the interaction appropriately, reducing the risk of a jarring expectancy violation. 5.3. Ethical Considerations and Future Research This research, while robust, has limitations that open compelling avenues for future inquiry. Our use of scenario-based experiments, while excellent for establishing causality, necessarily limits external validity. Future research should seek to replicate and extend these findings using data from live A/I chat logs or through field experiments. Furthermore, our findings are based on a Western, English-speaking sample; future research should test the cross-cultural validity of these effects. The project of "humanizing the machine" is itself laden with profound ethical responsibilities. Our finding that anthropomorphic AI can foster a genuine sense of emotional affinity raises critical questions about manipulation. An AI designed to be "charming" could be deployed not to genuinely solve a problem, but to create a facade of rapport that deflects legitimate complaints. Firms must ensure that the relational warmth projected by an AI is backed by genuine procedural and distributive justice. Furthermore, as AI language models become more sophisticated, they risk perpetuating biases inherited from their training data, leading to discriminatory outcomes that are insidious and difficult to detect. Firms must therefore invest in proactive bias mitigation strategies during model training and deploy transparent audit trails for AI-driven decisions. This necessitates robust governance and continuous auditing for algorithmic bias. The most significant opportunities for future research lie at the intersection of our findings and the rapidly evolving technological frontier. First, the rise of Generative AI (GenAI) fundamentally changes the nature of anthropomorphism. GenAI’s capacity for fluent, contextually rich conversation could dramatically enhance perceived behavioral and emotional humanness, yet its opaque reasoning could also create new, more subtle forms of the uncanny valley—an interaction that is grammatically perfect but emotionally hollow. Future research must investigate how customers perceive the authenticity of GenAI-driven empathy. Second, the emergence of agentic AI —systems capable of autonomous action—presents a new frontier for which no theoretical frameworks yet exist. Finally, future studies should explore critical contingencies, such as how these dynamics differ based on the locus of the service failure (firm vs. customer). By pursuing these questions, the research community can build upon the framework established here to develop a comprehensive understanding of the evolving human-AI service relationship. 6. Conclusion This research offers a timely and comprehensive examination of the complex interplay between anthropomorphism and the uncanny valley in the critical context of AI-mediated service recovery. Our findings consistently demonstrate that while AI agents designed to convey warmth, empathy, and adaptive intelligence can successfully bolster perceptions of justice and foster emotional connection, these benefits are extraordinarily precarious. A human-like design creates high expectations, and even minor imperfections can trigger a powerful uncanny valley response that erodes trust and undermines the entire recovery effort. Behavioral and emotional cues, in particular, emerge as the most potent—and therefore most high-stakes—dimensions of anthropomorphic design. Ultimately, this study expands our theoretical perspective on AI in service research, illuminating the necessity of a carefully calibrated design that avoids unsettling imperfections. It provides clear, actionable guidance for managers seeking to harness the efficiency of AI without sacrificing the essential relational and emotional nuances that underpin successful service and enduring customer trust. Declarations Ethics Approval Statement: This study was reviewed and approved by the Institutional Review Board (IRB) of Chitkara University and was performed in accordance with the ethical standards of the 1964 Helsinki Declaration and its later amendments. Participant Consent Statement: Written informed consent was obtained from all individual participants included in the study Author Profiles: SSRN: https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=7349141 ORCID: https://orcid.org/0000-0003-0337-7885 ResearchGate: https://www.researchgate.net/profile/Jaskirat-Singh-26 References Belanche, D., Casaló, L. V., Flavián, C., & Schepers, J. (2020). Service robot implementation: A theoretical framework and research agenda. Service Industries Journal , 40 (3-4), 203–225. Biocca, F., Harms, C., & Burgoon, J. K. (2003). Toward a more robust theory and measure of social presence: Review and suggested criteria. Presence: Teleoperators & Virtual Environments , 12 (5), 456–480. Blodgett, J. G., Hill, D. J., & Tax, S. S. (1997). The effects of distributive, procedural, and interactional justice on postcomplaint behavior. Journal of Retailing , 73 (2), 185–210. Bolton, R. N., & Mattila, A. S. (2015). How does anthropomorphism affect customer–robot interactions in retail service settings? Journal of Service Research , 18 (4), 427–446. Brislin, R. W. (1970). Back-translation for cross-cultural research. Journal of Cross-Cultural Psychology , 1 (3), 185–216. Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon’s Mechanical Turk: A new source of inexpensive, yet high-quality, data? Perspectives on Psychological Science , 6 (1), 3–5. Burgoon, J. K., & Jones, S. B. (1976). Toward a theory of personal space expectations and their violations. Human Communication Research , 2 (2), 131–146. Colquitt, J. A. (2001). On the dimensionality of organizational justice: A construct validation of a measure. Journal of Applied Psychology , 86 (3), 386–400. De Melo, C. M., Carnevale, P. J., Read, S. J., & Gratch, J. (2014). Reading people’s minds from emotion expressions in interdependent decision making. Journal of Personality and Social Psychology , 106 (1), 73–88. De Wulf, K., Odekerken-Schröder, G., & Iacobucci, D. (2001). Investments in consumer relationships: A cross-country and cross-industry exploration. Journal of Marketing , 65 (4), 33–50. Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review , 114 (4), 864–886. Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2009). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods , 41 (4), 1149–1160. Gefen, D., Karahanna, E., & Straub, D. W. (2003). Trust and TAM in online shopping: An integrated model. MIS Quarterly , 27 (1), 51–90. Go, E., & Carroll, J. M. (2021). Designing for reciprocal relationships with conversational AI. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (pp. 1–14). Hayes, A. F. (2017). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach . Guilford Publications. Ho, C.-C., & MacDorman, K. F. (2010). Measuring the uncanny valley effect. International Journal of Social Robotics , 2 (2), 129–139. Ho, C.-C., & MacDorman, K. F. (2017). Revisiting the uncanny valley theory: Developing and validating an alternative to the Godspeed indices. Computers in Human Behavior , 76 , 150–166. Hofstede, G. (2001). Culture's consequences: Comparing values, behaviors, institutions and organizations across nations . Sage Publications. Huang, M.-H., & Rust, R. T. (2018). Artificial intelligence in service. Journal of Service Research , 21 (2), 155–172. Huang, M.-H., & Rust, R. T. (2021). Engaged to a robot? The role of AI in service. Journal of Service Research , 24 (1), 30–41. Marinova, S., de Ruyter, K., Huang, M.-H., Meuter, M. L., & Challagalla, G. (2017). Getting smart: Learning from technology-empowered frontline interactions. Journal of Service Research , 20 (1), 29–42. Maxham III, J. G., & Netemeyer, R. G. (2002). A longitudinal study of complaining customers’ evaluations of multiple service failures and recovery efforts. Journal of Marketing , 66 (4), 57–71. McKnight, D. H., Choudhury, V., & Kacmar, C. (2002). Developing and validating trust measures for e-commerce: An integrative typology. Information Systems Research , 13 (3), 334–359. Meuter, M. L., Ostrom, A. L., Roundtree, R. I., & Bitner, M. J. (2000). Self-service technologies: Understanding customer satisfaction with technology-based service encounters. Journal of Marketing , 64 (3), 50–64. Mori, M., MacDorman, K. F., & Kageki, N. (2012). The uncanny valley [from the field]. IEEE Robotics & Automation Magazine , 19 (2), 98–100. Nowak, K. L., & Biocca, F. (2003). The effect of the agency and anthropomorphism on users’ sense of telepresence, copresence, and social presence in virtual environments. Presence: Teleoperators and Virtual Environments , 12 (5), 481–494. Parasuraman, A. (2000). Technology Readiness Index (TRI) a multiple-item scale to measure readiness to embrace new technologies. Journal of Service Research , 2 (4), 307–320. Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods , 40 (3), 879–891. Reeves, B., & Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places . Cambridge University Press. Saygin, A. P., Chaminade, T., Ishiguro, H., Driver, J., & Frith, C. (2012). The thing that should not be: Predictive coding and the uncanny valley in perceiving human and humanoid robot actions. Social Cognitive and Affective Neuroscience , 7 (4), 413–422. Smith, A. K., Bolton, R. N., & Wagner, J. (1999). A model of customer satisfaction with service encounters involving failure and recovery. Journal of Marketing Research , 36 (3), 356–372. Sparks, B. A., & McColl-Kennedy, J. R. (2001). Justice strategy options for increased customer satisfaction in a service failure. Journal of Business Research , 54 (3), 209–218. Tax, S. S., Brown, S. W., & Chandrashekaran, M. (1998). Customer evaluations of service complaint experiences: Implications for relationship marketing. Journal of Marketing , 62 (2), 60–76. Van Doorn, J., Mende, M., Noble, S. M., Hulland, J., Ostrom, A. L., Grewal, D., & Petersen, J. A. (2017). Domo arigato Mr. Roboto: Emergence of automated social presence in organizational frontlines and customers’ service experiences. Journal of Service Research , 20 (1), 43–58. Wang, Y., & Siau, K. (2019). Artificial intelligence, machine learning, automation, robotics, future of work and future of humanity: A review and research agenda. Journal of Database Management (JDM) , 30 (1), 61–79. Weiner, B. (1985). An attributional theory of achievement motivation and emotion. Psychological Review , 92 (4), 548–573. Xiao, L., & Kumar, V. (2021). Robotics for customer service: A useful complement or an ultimate substitute? Journal of Service Research , 24 (1), 9–29. Zeithaml, V. A., Berry, L. L., & Parasuraman, A. (1996). The behavioral consequences of service quality. Journal of Marketing , 60 (2), 31–46. Złotowski, J., Proudfoot, D., Yogeeswaran, K., & Bartneck, C. (2015). Anthropomorphism: Opportunities and challenges in human–robot interaction. AI & Society , 30 (3), 347–360. Author Publications (Corresponding Author) and Working Paper References: Singh, J. (2025). Leveraging Financial Innovation and Capability Enhancement for Sustainable Urban Poverty Alleviation in Punjab: Policy Imperatives . SSRN Scholarly Paper 5205435. Social Science Research Network. https://doi.org/10.2139/ssrn.5205435 Singh, J., Batra, G. S., & Chatrath, S. K. (2025a). Blockchain’s Role in Social Welfare, Financial Inclusion, and Public Sector Innovations in India: A Multi-Sector Analysis of Government-Led Initiatives . SSRN Scholarly Paper 5105250. Social Science Research Network. https://doi.org/10.2139/ssrn.5105250 Singh, J., Batra, G. S., & Chatrath, S. K. (2025b). Digital Capabilities for Urban Poverty Alleviation: Integrating E-Payment Awareness and Credit Utilization Patterns in Indian Slums . SSRN Scholarly Paper 5223860. Social Science Research Network. https://doi.org/10.2139/ssrn.5223860 Singh, J., Batra, G. S., & Chatrath, S. K. (2025c). Harnessing Fintech for Poverty Alleviation: Enhancing Credit Utilization and Livelihoods in Urban Slums of North-western India through the Capability Approach and Sustainable Livelihoods Framework . SSRN Scholarly Paper 5207210. Social Science Research Network. https://doi.org/10.2139/ssrn.5207210 Singh, J., & Sharma, D. (2024). Contemporary Challenges of Management Education in India: Review and Assessment. In Interdisciplinary Approaches in Management Education (pp. 149–167). Singh, J., & Singh, M. (2024a). Accelerating Financial Inclusion of the Urban Poor: Role of Innovative e-Payment Systems and JAM Trinity in Alleviating Poverty in India. Global Business Review , 09721509231222609. https://doi.org/10.1177/09721509231222609 Singh, J., & Singh, M. (2024b). Addressing unproductive credit consumption and beneficiary malpractices in social welfare programs for slum-dwellers: A study from India. Cities , 145, 104729. https://doi.org/10.1016/j.cities.2023.104729 Singh, J., & Singh, M. (2024c). Alleviating urban poverty in India: The role of capabilities and entrepreneurship development. International Journal of Social Economics , 51(10), 1314–1335. https://doi.org/10.1108/IJSE-07-2023-0514 Singh, V., & Singh, J. (2024). Quantifying the relationship between e-advertising capabilities and marketing mix cost savings. International Journal of Applied Management Science , 16(1), 44–67. https://doi.org/10.1504/IJAMS.2024.136149 Singh, J., Sharma, D., & Batra, G. S. (2023). Does Credit Utilization Pattern Promote Poverty Alleviation? An Evidence from India. Global Business Review , 24(6), 1227–1250. https://doi.org/10.1177/0972150920918967 Singh, J., & Singh, M. (2023a). Does financial inclusion impact socio-economic stability? A study of social safety net in Indian slums. International Journal of Social Economics , 50(8), 1060–1084. https://doi.org/10.1108/IJSE-04-2022-0261 Singh, J., & Singh, M. (2023b). Fintech applications in social welfare schemes during Covid times: An extension of the classic TAM model in India. International Social Science Journal , 73(250), 979–998. https://doi.org/10.1111/issj.12406 Singh, J., Batra, G. S., Sharma, D., & Singh, V. (2021). Microcredit Usage Pattern and its Impact on Economic Activities of the Urban Deprived: A Study of Punjab State, India . SSRN Scholarly Paper 5206281. Social Science Research Network. https://doi.org/10.2139/ssrn.5206281 Sharma, D., & Singh, J. (2018). Credit expansion programmes for urban poor: A literature review and conceptual framework. International Journal of Advanced in Management, Technology and Engineering Sciences (IJAMTES) , 8(IV), 646–657. Tables Table 1 Hypotheses Summary Hypothesis Description Theoretical Foundation(s) H1 Anthropomorphism → Increased justice perceptions Social Presence Theory (Biocca et al., 2003) H2 Anthropomorphism → Increased emotional affinity CASA Paradigm (Reeves & Nass, 1996) H3 Uncanny valley triggers → Decreased justice perceptions Expectancy Violation Theory (Burgoon & Jones, 1976) H4 Uncanny valley triggers → Decreased emotional affinity Uncanny Valley Theory (Mori et al., 2012) H5 Higher anthropomorphism amplifies uncanny valley effects Expectancy Violation Theory; Uncanny Valley Theory H6 Justice perceptions → Increased recovery satisfaction Justice Theory (Colquitt, 2001; Smith et al., 1999) H7 Emotional affinity → Increased recovery satisfaction Service Recovery Literature (Van Doorn et al., 2017) H8 Recovery satisfaction → Increased trust in AI Trust in Technology (McKnight et al., 2002) H9 Recovery satisfaction → Increased repatronage intentions Relationship Marketing (Zeithaml et al., 1996) H10 Tech. Savviness moderates Anthropomorphism effects (+) Technology Acceptance (Parasuraman, 2000) H11 Tech. Savviness moderates Uncanny Valley effects (-) Individual Differences (Meuter et al., 2000) H12 Prior Relationship moderates Satisfaction → Repatronage Relationship Strength (De Wulf et al., 2001) Table 2 Summary of Experimental Manipulation Checks Construct Manipulation Validation (Check Item & Result) EXPERIMENT 1 Anthropomorphism High: Realistic avatar, empathic language, personal pronouns Low: Generic icon, formal phrasing, impersonal Participants rated perceived human-likeness. High vs. Low differed significantly (p < .001). Uncanny Valley Present: Subtle yet unsettling glitches (frozen avatar, odd wording) Absent: Smooth, coherent AI responses Participants rated “eerie” or “creepy” feelings. Present vs. Absent differed significantly (p < .001). EXPERIMENT 2 Physical High: Realistic human avatar with natural facial proportions Low: Simple or abstract icon (robotic outline) “The AI’s avatar looked human-like.” High vs. Low differed significantly (p < .001). Behavioral High: Casual, conversational style, personal greetings Low: Robotic, formulaic script “The AI used natural conversation.” High vs. Low differed significantly (p < .001). Emotional High: Empathy statements (“I understand how frustrating…”) Low: Minimal or no empathic content “The AI recognized/acknowledged my feelings.” High vs. Low differed significantly (p < .001). Cognitive High: Adaptive problem-solving, personalized solutions Low: Standard responses, no personalization “The AI customized solutions to my problem.” High vs. Low differed significantly (p < .001). Table 3 Construct Measures and Reliability Construct Sample Item Source(s) Reliability Justice Perceptions “The solution I received from the AI was fair.” Colquitt (2001); Tax et al. (1998) α = .89–.93 Emotional Affinity “I felt a personal connection with the AI during the interaction.” Van Doorn et al. (2017) α = .94–.95 Recovery Satisfaction “Overall, I am satisfied with the AI’s handling of my complaint.” Maxham & Netemeyer (2002) α = .92–.96 Trust in AI “I believe this AI agent is competent in resolving issues.” McKnight et al. (2002) α = .90–.92 Repatronage Intentions “I would use this service again in the future.” Zeithaml et al. (1996) α = .92–.94 Technological Savviness “I am comfortable learning new technologies quickly.” Meuter et al. (2000); Parasuraman (2000) α = .85–.88 Prior Relationship “I feel a sense of loyalty to this company.” De Wulf et al. (2001) α = .88–.90 Note : Confirmatory Factor Analyses (CFAs) in each experiment indicated that all constructs achieved good fit indices (CFI > 0.90, RMSEA < 0.08), supporting construct validity. Table 4 Experiment 1: Key ANOVA Results (N = 484) Dependent Variable Source of Variance F-value p-value Effect Size (η²) Justice Perceptions Anthropomorphism (A) 144.92 < .001 .23 Uncanny Valley (UV) 97.55 < .001 .17 Emotional Affinity Anthropomorphism (A) 186.44 < .001 .28 Uncanny Valley (UV) 121.30 .10). Table 5 Experiment 2: Key Path and Moderation Effects (N = 776) Relationship β p-value Supported? Behavioral Anthropomorphism → Justice Perceptions .35 < .001 Yes Emotional Anthropomorphism → Justice Perceptions .38 < .001 Yes Uncanny (Mismatch) → Justice Perceptions -.33 < .001 Yes Justice Perceptions → Recovery Satisfaction .43 < .001 Yes Emotional Affinity → Recovery Satisfaction .37 < .001 Yes Recovery Satisfaction → Trust in AI .42 < .001 Yes Recovery Satisfaction → Repatronage Intentions .54 < .001 Yes Moderation Effects Tech. Savviness × Anthropomorphism → Justice .18 .006 Yes Tech. Savviness × Uncanny → Justice -.15 .037 Yes Prior Rel. × Recovery Satisfaction → Repatronage .22 < .001 Yes Additional Declarations The authors declare no competing interests. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6889879","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":471028787,"identity":"c369e0cc-4f88-4cda-892f-7b448fc7038c","order_by":0,"name":"Arpandeep Kaur","email":"","orcid":"https://orcid.org/0000-0002-2052-6275","institution":"Chitkara University, Punjab, India","correspondingAuthor":false,"prefix":"","firstName":"Arpandeep","middleName":"","lastName":"Kaur","suffix":""},{"id":471028788,"identity":"6d27178a-3a08-4af9-9548-42b7575a5dac","order_by":1,"name":"Jaskirat Singh","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIiWNgGAWjYBACxgYGZgbGfzZy/CBeQgGxWhjY0owlG0BaDIizCKTlcKLBARCbGC3M7ccfG/PwpCUYn1+d+OGBAYM8v9gBAg7ryTFO5pGwyTO78XazBNBhhjNnJxDQ0pDDfJjHIK3Y7MbZDSAtCQa3CWnpf/74ME/C4cTNM85u/kGclhkJQIcdOJy4gb93G5G2zHhjbDi3Ic1Y4gbvNosEAwnCfjHsT38s8bYBGJX9Zzff/FFhI88vTUhLAwMDEw+IJQFWKYFfOQjIgxz3A8TiP0BY9SgYBaNgFIxMAAA7t0YMWJeHfwAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0003-0337-7885","institution":"ICSSR, Ministry of Education, New Delhi, India","correspondingAuthor":true,"prefix":"","firstName":"Jaskirat","middleName":"","lastName":"Singh","suffix":""}],"badges":[],"createdAt":"2025-06-13 16:49:05","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-6889879/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6889879/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":84860258,"identity":"41d1def5-2f08-47ea-8ad5-ad5b979eca7f","added_by":"auto","created_at":"2025-06-18 06:48:46","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":600357,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFIGURE I: CONCEPTUAL MODEL\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-6889879/v1/db3d9e1a24af70ab1a9ff494.png"},{"id":84861308,"identity":"5b0740d3-3704-48c1-be5f-e9f5cb8174e3","added_by":"auto","created_at":"2025-06-18 07:04:47","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2110848,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6889879/v1/7bccf518-8cb4-45aa-88f9-eb657d9720c4.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eHumanizing the Machine: Anthropomorphism and the Uncanny Valley in AI-Mediated Service Recovery\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe landscape of service interaction is undergoing a seismic transformation, driven by the proliferation of increasingly autonomous and sophisticated Artificial Intelligence (AI). Once relegated to simple automation, AI agents\u0026mdash;from chatbots to virtual assistants\u0026mdash;are now deployed at the most critical and emotionally charged of customer touchpoints: service recovery. This moment of truth, where a firm\u0026rsquo;s response to a failure can either salvage or irrevocably shatter a customer relationship, is now frequently entrusted to a machine (Huang and Rust, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Xiao and Kumar, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). In an effort to preserve the relational quality of these encounters, firms increasingly pursue a strategy of anthropomorphism, designing AI agents with human-like appearances, conversational styles, and expressions of empathy (Epley, Waytz, and Cacioppo, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). The underlying assumption is that a more human-like AI will engender greater social presence, fostering the trust and satisfaction necessary for successful recovery (van Doorn et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Yet, this pursuit of the perfectly human-like machine walks a razor\u0026rsquo;s edge, bordering the unsettling phenomenon of the \u0026ldquo;uncanny valley,\u0026rdquo; where near-human entities provoke not affinity, but a sense of unease and disquiet (Mori, MacDorman, and Kageki, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). An AI agent that appears almost\u0026mdash;but not quite\u0026mdash;human risks violating customer expectations in a manner that can compound the initial service failure, turning a moment of recovery into one of profound alienation.\u003c/p\u003e \u003cp\u003eThis central paradox\u0026mdash;that the very attempt to humanize an AI for service recovery may catastrophically backfire\u0026mdash;presents a significant theoretical and practical challenge. Current scholarship offers potentially conflicting guidance. On one hand, a robust body of work extols the virtues of anthropomorphism for elevating the perceived warmth and relational quality of service agents (Belanche, Casal\u0026oacute;, Flavi\u0026aacute;n, and Schepers, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). On the other, human-computer interaction research provides stark warnings about the risks of imperfect human-likeness (Ho and MacDorman, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). What remains critically unclear is the tipping point: \u003cem\u003ewhen\u003c/em\u003e and \u003cem\u003ehow\u003c/em\u003e does the laudable pursuit of a human-like design cross the precipice into the uncanny? The core theoretical mechanisms governing this interplay in the high-stakes context of service recovery remain underdeveloped. While the tenets of justice theory are foundational to service recovery\u0026mdash;positing that customers seek fairness in outcomes, processes, and interactions (Tax, Brown, and Chandrashekaran, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e1998\u003c/span\u003e)\u0026mdash;it is a profound theoretical leap to assume they function identically when the interactional partner is a machine. The very essence of \"interactional justice\" and \"emotional affinity\" is thrown into question when one party possesses no genuine emotions or consciousness. We must therefore scrutinize the process by which a \u003cem\u003edesigned\u003c/em\u003e humanness translates\u0026mdash;or fails to translate\u0026mdash;into these crucial mediating outcomes.\u003c/p\u003e \u003cp\u003eThis investigation addresses several crucial gaps in the literature. First, we move beyond treating anthropomorphism as a monolithic concept to examine how its distinct dimensions\u0026mdash;physical, behavioral, emotional, and cognitive\u0026mdash;differentially influence customer perceptions. We integrate this with an analysis of uncanny valley triggers to understand the conditions under which human-like cues become liabilities. Second, we dissect the underlying process, examining the mediating roles of justice perceptions and emotional affinity to explain \u003cem\u003ewhy\u003c/em\u003e these AI design elements affect downstream outcomes like recovery satisfaction, trust, and repatronage intentions. Third, we recognize that customer responses are not uniform. We therefore clarify crucial boundary conditions by investigating how individual differences, namely customers' technological savviness and their prior relationship with the firm, moderate these effects, providing a more complete picture of \u003cem\u003efor whom\u003c/em\u003e AI-mediated recovery is most effective.\u003c/p\u003e \u003cp\u003eTo address these questions, this paper develops and empirically tests a comprehensive framework through two robust experimental studies. Our primary contribution is to illuminate the inherent fragility of anthropomorphism\u0026rsquo;s benefits in service recovery. We demonstrate that while specific human-like cues can powerfully foster perceptions of justice and build emotional rapport, these gains are precarious and can be completely undone by subtle design flaws that trigger uncanny valley responses. In doing so, this research (1) contributes a novel contingency framework to service theory, revealing how the interplay between anthropomorphism and the uncanny valley reconfigures the established roles of justice and affinity; (2) offers a multidimensional view of AI anthropomorphism, identifying the specific facets most critical for successful recovery; and (3) clarifies key boundary conditions that shape the efficacy of AI service agents. The insights generated have significant implications for both theory and practice, providing a crucial roadmap for designing AI agents that can foster positive experiences when resolving the most challenging service failures.\u003c/p\u003e"},{"header":"2. Literature Review and Hypotheses Development","content":"\u003cp\u003e\u003cstrong\u003e2.1. AI in Service and the Imperative of Relational Recovery\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFar beyond its origins in simple automation, artificial intelligence now performs sophisticated service functions that can supplement or wholly substitute for human employees (Huang and Rust, 2021). This technological ascent places AI at the forefront of a firm\u0026rsquo;s most sensitive encounters, particularly the emotionally charged context of service recovery, where consumer acceptance is paramount (Marinova et al., 2017). The bedrock of traditional recovery\u0026mdash;a process vital for rebuilding trust and loyalty\u0026mdash;has always been the delivery of justice in its three forms: distributive, procedural, and interactional (Tax et al., 1998; Smith, Bolton, and Wagner, 1999). It is precisely this third pillar of interactional justice, rooted in empathy and respect, where AI\u0026rsquo;s capabilities are most contested. These agents are tasked with emulating a relational warmth they do not inherently possess (De Melo et al., 2014; Xiao and Kumar, 2021), creating the defining challenge for their design: how to reconcile the cold logic of procedural efficiency with the necessary illusion of interpersonal understanding.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2. Anthropomorphism and the Cultivation of Social Presence\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirms endeavor to close this relational divide through anthropomorphism, a strategy of imbuing agents with human traits, emotions, or intentions (Epley et al., 2007). This is a well-established approach in human-computer interaction, designed to fundamentally alter user perceptions by cultivating social presence\u0026mdash;the psychological feeling of \u0026quot;being with\u0026quot; another (Nowak and Biocca, 2003; Biocca, Harms, and Burgoon, 2003). When successful, this enhanced social presence prompts users to interact with the AI not as a mere tool, but as a social counterpart, applying the same norms and heuristics that govern human relationships (Reeves and Nass, 1996). The means of achieving this social presence are, however, varied and distinct. Recent work identifies separate physical, behavioral, emotional, and cognitive dimensions of anthropomorphism (Złotowski et al., 2015). An agent\u0026apos;s behavioral and emotional expressiveness, for example, directly speaks to its perceived empathy and thus powerfully shapes judgments of interactional justice. Likewise, its cognitive acuity signals competence, which in turn bolsters evaluations of procedural and distributive fairness. When an AI is perceived as both a caring and capable partner in the recovery process, it is also more likely to engender emotional affinity. From this theoretical foundation, our initial hypotheses emerge:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH1:\u003c/strong\u003e Anthropomorphism (across its physical, behavioral, emotional, and cognitive dimensions) positively influences customers\u0026rsquo; justice perceptions following a service failure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH2:\u003c/strong\u003e Anthropomorphism (across the same dimensions) positively influences customers\u0026rsquo; emotional affinity toward the AI following a service failure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3. The Uncanny Valley as an Expectancy Violation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhile anthropomorphism offers clear benefits, the strategy is inherently perilous. As a humanoid entity approaches but fails to perfectly replicate human appearance and behavior, observers can experience a sharp dip in affinity, a phenomenon famously termed the uncanny valley (Mori et al., 2012). These feelings of eeriness and discomfort arise from minor yet unsettling imperfections\u0026mdash;a stilted blink, a disjointed phrase, a subtly unnatural facial expression\u0026mdash;that betray the entity\u0026rsquo;s artifice (Ho and MacDorman, 2017).\u003c/p\u003e\n\u003cp\u003eExpectancy Violation Theory (EVT) provides a robust theoretical lens through which to understand this reaction (Burgoon and Jones, 1976). EVT posits that individuals hold socially ingrained expectations for behavior in any given interaction. When another party\u0026rsquo;s actions breach these expectations, a violation occurs, which is then appraised for its valence (positive or negative). In the context of a highly anthropomorphic AI, the agent\u0026rsquo;s human-like features set a strong expectation of a seamless, natural, and authentically human-like interaction. Uncanny valley triggers\u0026mdash;such as \u003cstrong\u003eeeriness\u003c/strong\u003e, a \u003cstrong\u003elack of realism\u003c/strong\u003e in appearance or movement, or \u003cstrong\u003eunpredictability\u003c/strong\u003e in logic or response (Saygin et al., 2012)\u0026mdash;constitute a stark negative violation of this expectation. This mismatch between the anticipated human-like coherence and the observed mechanical quirk creates a profoundly negative affective response. In a service recovery context, this adverse reaction can contaminate judgments of the interaction, directly diminishing perceptions of fairness and eroding any nascent emotional affinity. We therefore hypothesize:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH3:\u003c/strong\u003e Uncanny valley triggers (eeriness, lack of realism, unpredictability) negatively influence customers\u0026rsquo; justice perceptions following a service failure.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH4:\u003c/strong\u003e Uncanny valley triggers (eeriness, lack of realism, unpredictability) negatively influence customers\u0026rsquo; emotional affinity toward the AI following a service failure.\u003c/p\u003e\n\u003cp\u003eFurthermore, we propose that the level of anthropomorphism sets the very conditions for a potential uncanny valley experience. An AI that is obviously a machine creates no expectation of human-like fidelity; its errors are perceived merely as technical failures. In contrast, a highly anthropomorphic AI creates a high-stakes interaction by implicitly promising human-like congruence. When that promise is broken by an imperfect cue, the resulting expectancy violation is magnified. This suggests a perilous interaction where the potential for disappointment escalates with the degree of attempted humanization. This logical tension leads to a cornerstone proposition:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH5:\u003c/strong\u003e Greater levels of anthropomorphism will amplify the negative effect of imperfect cues, leading to a stronger elicitation of uncanny valley responses and consequently diminished customer perceptions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4. The Mediating Pathways to Recovery Outcomes\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eJustice theory posits that fairness perceptions are a primary determinant of customer satisfaction following service recovery (Colquitt, 2001; Smith et al., 1999). When customers perceive the outcome, process, and interaction as fair, their overall evaluation of the recovery experience is enhanced. Similarly, the development of emotional affinity\u0026mdash;a positive relational bond with the service agent\u0026mdash;has been shown to soften negative reactions and improve recovery evaluations (Van Doorn et al., 2017). We expect these principles to hold firmly in AI-mediated contexts:\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH6:\u003c/strong\u003e Justice perceptions positively influence recovery satisfaction in AI-mediated service recovery.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH7:\u003c/strong\u003e Emotional affinity toward the AI positively influences recovery satisfaction in AI-mediated service recovery.\u003c/p\u003e\n\u003cp\u003eThe significance of recovery satisfaction extends far beyond the immediate resolution of a complaint, shaping critical downstream attitudes and behaviors (Maxham and Netemeyer, 2002). Specifically, a customer who is satisfied with an AI-mediated interaction is far more inclined to develop trust in the technology\u0026rsquo;s capabilities for future encounters (Gefen, Karahanna, and Straub, 2003; McKnight, Choudhury, and Kacmar, 2002). This combination of satisfaction and emergent trust is what ultimately cements customer loyalty, manifesting as stronger intentions to repatronize the brand (Zeithaml, Berry, and Parasuraman, 1996).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH8:\u003c/strong\u003e Recovery satisfaction positively influences trust in the AI.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH9:\u003c/strong\u003e Recovery satisfaction positively influences repatronage intentions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5. Boundary Conditions: The Moderating Roles of Customer Characteristics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe effects of AI design are unlikely to be uniform across all customers. We propose two critical boundary conditions that shape these relationships: technological savviness and prior relationship strength.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTechnological Savviness.\u003c/strong\u003e Individual-level comfort and familiarity with technology can fundamentally shape reactions to AI (Parasuraman, 2000). Tech-savvy users possess more developed and nuanced mental models of AI capabilities and limitations. Consequently, they may be more appreciative of sophisticated anthropomorphic features, leading to a stronger positive effect on justice and affinity perceptions. Conversely, their familiarity may render them more tolerant of minor glitches, which they are more likely to interpret as predictable system errors rather than eerie violations of humanness. Less tech-savvy customers, by contrast, may be more unsettled by near-human AI and more frustrated by its imperfections.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH10:\u003c/strong\u003e Technological savviness positively moderates the effect of anthropomorphism on justice perceptions and emotional affinity, such that the effect is stronger for more tech-savvy customers.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH11:\u003c/strong\u003e Technological savviness negatively moderates the effect of uncanny valley triggers on justice perceptions and emotional affinity, such that tech-savvy customers are less negatively affected by these triggers.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePrior Relationship with the Company.\u003c/strong\u003e A pre-existing bond of loyalty and trust with a firm can serve as a powerful buffer during service failures (De Wulf, Odekerken-Schr\u0026ouml;der, and Iacobucci, 2001). According to attribution theory (Weiner, 1985), customers with a strong prior relationship are more likely to give the firm the \u0026ldquo;benefit of the doubt,\u0026rdquo; attributing failures to temporary or external factors. This relational equity can create a halo effect that extends to the firm\u0026rsquo;s AI agents. For these customers, a satisfying recovery experience is more likely to translate into reaffirmed trust and loyalty, as it aligns with their existing positive disposition toward the firm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eH12:\u003c/strong\u003e A stronger prior relationship with the company positively moderates the effect of recovery satisfaction on repatronage intentions, such that this link is heightened when the existing relationship is stronger.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT FIGURE I: CONCEPTUAL MODEL HERE)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT TABLE I: HYPOTHESES SUMMARY TABLE HERE)\u003c/strong\u003e\u003c/p\u003e"},{"header":"3. Methodology","content":"\u003cp\u003eThis research employs a two-study experimental design to establish the causal impact of anthropomorphism and uncanny valley triggers on customer responses in AI-mediated service recovery and to test the full scope of our conceptual model. Both experiments utilize scenario-based vignettes to ensure high internal validity while capturing the essential context of a service failure interaction. In keeping with best practices for experimental research, both studies included rigorous manipulation checks, attention checks, and measures to ensure reliability and validity. Furthermore, this research adheres to open science principles; the anonymized data and analysis scripts for both experiments will be made available on the Open Science Framework (OSF) upon publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.1. Experiment 1: Core Effects of Anthropomorphism and the Uncanny Valley\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe primary objective of Experiment 1 was to establish the core causal effects of anthropomorphism and uncanny valley triggers on the mediating variables of justice perceptions and emotional affinity. A 2 (anthropomorphism: high vs. low) \u0026times; 2 (uncanny valley: present vs. absent) between-subjects design was employed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eParticipants and Procedure.\u003c/strong\u003e Our participant pool was drawn from Prolific, an online platform recognized for providing diverse and high-quality data samples suitable for academic research (Buhrmester, Kwang, and Gosling, 2011). This demographic is appropriate for this study due to its broad familiarity with online retail and digital service interactions. Five hundred participants were recruited (52% female; M_age_=38.2, SD=10), all of whom were English-speaking adults reporting at least one service failure experience in the preceding six months. This criterion ensured a degree of familiarity with the complaint context. The procedure began with a pre-scenario questionnaire capturing demographics. Participants were then immersed in a scenario where they had received an incorrect item from a fictitious online retailer, \u0026ldquo;ElectroDeals.\u0026rdquo; The failure severity was held constant across all conditions. Subsequently, they engaged in a simulated text-based chat with an AI service agent to resolve the issue, where our manipulations were embedded. Immediately following the interaction, they completed a post-scenario questionnaire containing our dependent measures and manipulation checks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExperimental Manipulations.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnthropomorphism (High vs. Low):\u003c/strong\u003e In the \u003cem\u003ehigh anthropomorphism\u003c/em\u003e condition, the AI agent was designed to maximize social presence. It featured a realistic human avatar, communicated using first-person pronouns (e.g., \u0026ldquo;I\u0026rsquo;m truly sorry this happened. I will look into this for you\u0026rdquo;), and offered personalized solutions. In contrast, the \u003cem\u003elow anthropomorphism\u003c/em\u003e condition presented the agent as a functional tool, using a generic icon, formal phrases (e.g., \u0026ldquo;Your input has been recorded\u0026rdquo;), and standardized solutions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eUncanny Valley (Present vs. Absent):\u003c/strong\u003e In the \u003cem\u003epresent\u003c/em\u003e condition, the interaction included subtle, pre-tested cues designed to elicit an uncanny feeling. These were not mere errors but theoretically grounded violations of human-like expectations, including a momentary, unnatural freezing of the avatar\u0026rsquo;s facial expression and instances of slight linguistic disfluency. In the \u003cem\u003eabsent\u003c/em\u003e condition, the AI interaction was seamless. The avatar in the high-anthropomorphism condition was carefully rendered with smooth, naturalistic animations, and all chat responses were coherent. Pre-testing confirmed that the \u0026ldquo;present\u0026rdquo; condition cues reliably evoked feelings of eeriness rather than simply being perceived as incompetence.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2. Experiment 2: Dimensionality of Anthropomorphism and Moderation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe purpose of Experiment 2 was twofold: first, to provide a more fine-grained view of anthropomorphism\u0026rsquo;s multidimensional impact, and second, to test the proposed moderation effects. The study utilized a 2 (physical anthropomorphism: high vs. low) \u0026times; 2 (behavioral anthropomorphism: high vs. low) \u0026times; 2 (emotional anthropomorphism: high vs. low) \u0026times; 2 (cognitive anthropomorphism: high vs. low) between-subjects factorial design.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eParticipants and Procedure.\u003c/strong\u003e A separate sample of 800 participants was recruited from Prolific (55% female; M_age_=40.1, SD=9.8). The screening criteria were slightly broadened to an age range of 18\u0026ndash;65 to enhance generalizability, while still requiring prior experience with chatbots and service failures. The procedure mirrored that of Experiment 1, with the addition of measures for our moderator variables in the pre-scenario questionnaire.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eExperimental Manipulations.\u003c/strong\u003e Each dimension of anthropomorphism was manipulated as a separate factor: Physical (realistic avatar vs. abstract icon), Behavioral (natural language vs. robotic phrasing), Emotional (empathic statements vs. no emotional cues), and Cognitive (adaptive problem-solving vs. scripted solutions). In this more complex design, uncanny valley triggers were measured via items assessing perceived \u0026quot;mismatch\u0026quot; or \u0026quot;creepiness,\u0026quot; allowing us to test their effect correlationally.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3. Measures\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll primary constructs were measured using multi-item scales on 7-point Likert scales (1 = strongly disagree, 7 = strongly agree), adapted from established literature to fit the AI service recovery context. A series of Confirmatory Factor Analyses (CFAs) were conducted, and the results indicated that all constructs achieved excellent model fit (e.g., CFI \u0026gt; 0.90, RMSEA \u0026lt; 0.08) and demonstrated strong convergent and discriminant validity, confirming the robustness of our measurement approach. To mitigate potential common method bias, procedural remedies were employed, including the psychological separation of predictors and criteria and the use of attention checks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT TABLE III: MEASURES HERE)\u003c/strong\u003e\u003c/p\u003e"},{"header":"4. Data Analysis and Results","content":"\u003cp\u003e\u003cstrong\u003e4.1. Data Cleaning and Manipulation Checks\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAcross both experiments, data were cleaned prior to analysis. In Experiment 1, 16 participants were removed for failing attention checks or for straight-line responding, resulting in a final sample of N = 484. In Experiment 2, 24 participants were removed, yielding a final sample of N = 776.\u003c/p\u003e\n\u003cp\u003eManipulation checks confirmed the success of our experimental designs. In Experiment 1, participants in the high-anthropomorphism condition perceived the AI as significantly more human-like (t(482) = 22.04, p \u0026lt; .001), and those in the uncanny valley \u0026ldquo;present\u0026rdquo; condition reported significantly higher feelings of eeriness (t(482) = 18.10, p \u0026lt; .001). In Experiment 2, manipulation checks for each of the four dimensions of anthropomorphism all yielded significant differences between their respective high and low conditions (all p \u0026lt; .001).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT TABLE II: MANIPULATION CHECKS HERE)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.2. Experiment 1: Core Effects\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA 2x2 ANOVA was used to test H1-H5. As predicted, we found a significant main effect of anthropomorphism on both justice perceptions (F(1, 480) = 144.92, p \u0026lt; .001, \u0026eta;\u0026sup2; = .23), supporting H1, and emotional affinity (F(1, 480) = 186.44, p \u0026lt; .001, \u0026eta;\u0026sup2; = .28), supporting H2. Likewise, the uncanny valley had a significant negative main effect on justice perceptions (F(1, 480) = 97.55, p \u0026lt; .001, \u0026eta;\u0026sup2; = .17), supporting H3, and emotional affinity (F(1, 480) = 121.30, p \u0026lt; .001, \u0026eta;\u0026sup2; = .20), supporting H4.\u003c/p\u003e\n\u003cp\u003eThe omnibus ANOVA test did not reveal a statistically significant interaction effect for H5. However, a planned post-hoc correlational analysis specifically within the high-anthropomorphism condition provided preliminary support for the underlying mechanism: the presence of unsettling glitches was significantly correlated with increased discomfort (r = .31, p \u0026lt; .05), a relationship that was absent in the low-anthropomorphism condition.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT TABLE IV: EXPERIMENT 1 KEY RESULTS HERE)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e4.3. Experiment 2: Full Model Test\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eExperiment 2 tested the full conceptual model. A MANOVA assessing the dimensional effects (H1, H2) showed that \u003cstrong\u003ebehavioral\u003c/strong\u003e and \u003cstrong\u003eemotional\u003c/strong\u003e anthropomorphism had the largest impacts on justice and affinity (all p \u0026lt; .001), while physical and cognitive dimensions exerted smaller positive effects (p \u0026lt; .01). Measured \u0026quot;mismatch\u0026quot; cues correlated negatively with both justice (\u0026beta; = -.33, p \u0026lt; .001) and emotional affinity (\u0026beta; = -.26, p \u0026lt; .001), re-affirming H3 and H4. Crucially, a significant interaction emerged, providing stronger support for H5: uncanny feelings were most exacerbated when high \u003cstrong\u003ephysical/emotional\u003c/strong\u003e anthropomorphism was combined with perceptible glitches.\u003c/p\u003e\n\u003cp\u003eRegression analyses confirmed the full mediation pathway. Justice perceptions (\u0026beta; = .43, p \u0026lt; .001) and emotional affinity (\u0026beta; = .37, p \u0026lt; .001) were strong predictors of \u003cstrong\u003erecovery satisfaction\u003c/strong\u003e (H6, H7). Recovery satisfaction, in turn, positively influenced both \u003cstrong\u003etrust in AI\u003c/strong\u003e (\u0026beta; = .42, p \u0026lt; .001) and \u003cstrong\u003erepatronage intentions\u003c/strong\u003e (\u0026beta; = .54, p \u0026lt; .001) (H8, H9). Finally, moderated regression supported our remaining hypotheses. \u003cstrong\u003eTechnological Savviness\u003c/strong\u003e amplified the benefits of anthropomorphism (\u0026beta; = .18, p \u0026lt; .01) (H10) and diminished the harm from uncanny triggers (\u0026beta; = -.15, p \u0026lt; .05) (H11). A stronger \u003cstrong\u003ePrior Relationship\u003c/strong\u003e strengthened the link between recovery satisfaction and repatronage (\u0026beta; = .22, p \u0026lt; .001), supporting H12.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(INSERT TABLE V: EXPERIMENT 2 KEY EFFECTS HERE)\u003c/strong\u003e\u003c/p\u003e"},{"header":"5. General Discussion","content":"\u003cp\u003eThis research was motivated by a central paradox in the design of AI for service recovery: the simultaneous promise and peril of anthropomorphism. Our findings, derived from two robust experiments, offer a nuanced and comprehensive resolution to this tension, yielding significant contributions to service theory and actionable guidance for practice. We demonstrate that while humanizing AI is a powerful strategy for enhancing customer perceptions of justice and affinity, its benefits are remarkably fragile and highly contingent on both design execution and customer characteristics.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.1. Theoretical Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFirst and foremost, this study \u003cstrong\u003eextends service recovery theory into the era of advanced AI, revealing how the interplay of anthropomorphism and the uncanny valley reconfigures foundational tenets.\u003c/strong\u003e Our research affirms that customers continue to seek fairness in AI-mediated recovery, reinforcing the relevance of justice theory. However, we advance this understanding by showing \u003cem\u003ehow\u003c/em\u003e AI design directly manipulates these perceptions. The finding that \u003cem\u003ebehavioral\u003c/em\u003e and \u003cem\u003eemotional\u003c/em\u003e dimensions of anthropomorphism are especially critical (Experiment 2) highlights the primacy of interactional justice cues in technologically mediated encounters. This refines our understanding of Colquitt\u0026apos;s (2001) framework, suggesting that when an AI agent is the interactional partner, its perceived capacity for empathy and natural conversation disproportionately shapes the entire justice evaluation. Our most significant theoretical insight, however, lies in the explication of H5. By integrating social presence and expectancy violation theories, we move beyond a simple cost-benefit analysis of anthropomorphism. We demonstrate that a highly human-like AI does not simply add benefits; it fundamentally raises the relational stakes. The design creates a strong expectation of human-like authenticity, which, when violated by even subtle imperfections, triggers a powerful uncanny valley response that negates the positive effects. This dynamic presents a novel theoretical tension for service research, suggesting a nonlinear, and at times negative, return on investments in anthropomorphic design.\u003c/p\u003e\n\u003cp\u003eSecond, our research contributes a \u003cstrong\u003emultidimensional and decomposed understanding of AI anthropomorphism in a service context.\u003c/strong\u003e By operationalizing and testing physical, behavioral, emotional, and cognitive dimensions separately (Złotowski et al., 2015), we provide a more granular and managerially relevant perspective. We show that not all human-like cues are created equal. The pronounced impact of behavioral and emotional cues suggests that for service recovery, what the AI \u003cem\u003esays\u003c/em\u003e and \u003cem\u003ehow it \u0026quot;feels\u0026quot;\u003c/em\u003e is more important than what it \u003cem\u003elooks like\u003c/em\u003e or how it \u003cem\u003ethinks\u003c/em\u003e. This finding challenges a design ethos that may over-invest in visual avatar fidelity at the expense of developing more sophisticated and contextually aware conversational and emotional intelligence.\u003c/p\u003e\n\u003cp\u003eThird, this paper \u003cstrong\u003eclarifies critical boundary conditions that delineate the user- and firm-specific contingencies of AI service efficacy.\u003c/strong\u003e In supporting H10 and H11, we show that technological savviness acts as a powerful lens through which AI interactions are interpreted. For tech-savvy customers, the benefits of anthropomorphism are amplified, and the sting of uncanny triggers is dulled. This suggests their sophisticated mental models of technology allow them to appreciate the design effort while dismissing imperfections as predictable system noise rather than unsettling social violations. Furthermore, by confirming that a strong prior firm relationship buffers recovery outcomes (H12), our findings underscore the enduring power of brand equity. A history of trust creates a powerful halo effect that not only encourages forgiveness but also strengthens the path from a satisfactory recovery to future loyalty, even when that recovery is handled by a machine.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.2. Managerial Implications\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe findings of this research offer clear, evidence-based guidance for managers navigating the high-stakes implementation of AI in service recovery. The central takeaway is that designing effective service AI is not merely a technical challenge but a delicate exercise in managing customer psychology. A naive \u0026quot;more human is always better\u0026quot; strategy is not only suboptimal\u0026mdash;it is actively dangerous.\u003c/p\u003e\n\u003cp\u003eFirst, managers must \u003cstrong\u003estrategically prioritize investments in behavioral and emotional anthropomorphism over purely physical or cognitive dimensions.\u003c/strong\u003e Our findings show that natural, empathetic language and adaptive, personalized responses yield the highest return on investment in fostering justice perceptions and emotional affinity. Rather than allocating significant resources to developing hyper-realistic but potentially static avatars, firms should focus on advancing the AI\u0026apos;s conversational intelligence. This involves designing dialogue systems that can move beyond rigid scripts to utilize natural language, conversational fillers, and personalized acknowledgements of the customer\u0026apos;s issue and emotional state. This is where the true sense of interactional fairness is forged.\u003c/p\u003e\n\u003cp\u003eSecond, the risk of the uncanny valley must be \u003cstrong\u003eproactively managed through rigorous, segment-specific pre-testing.\u003c/strong\u003e The adage that \u0026quot;the last 1% is the hardest\u0026quot; is particularly true for AI design; near-perfect human-likeness makes any small flaw glaring and unsettling. Firms must move beyond standard quality assurance to conduct user experience testing specifically designed to detect \u0026quot;creepiness.\u0026quot; This involves testing interactions with target customer segments to identify subtle triggers\u0026mdash;be they visual, auditory, or linguistic\u0026mdash;that provoke unease. Given that tech-savvy customers are more tolerant of such glitches, firms could consider offering different AI interaction styles. For less savvy segments, a less ambitious but more robustly coherent AI may be safer and more effective than a highly anthropomorphic agent that risks uncanny failure.\u003c/p\u003e\n\u003cp\u003eFinally, managers should \u003cstrong\u003eleverage existing brand equity as a crucial buffer when deploying service AI.\u003c/strong\u003e Our findings confirm that a strong prior customer relationship significantly enhances the likelihood that a successful AI recovery will translate into future loyalty. This suggests that initial AI rollouts might be most effective when targeted at loyal, established customers who are more likely to grant the firm\u0026mdash;and by extension, its technology\u0026mdash;the benefit of the doubt. For all customer segments, transparency about the agent\u0026apos;s AI nature is paramount, as disclosure can help manage expectations and frame the interaction appropriately, reducing the risk of a jarring expectancy violation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.3. Ethical Considerations and Future Research\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research, while robust, has limitations that open compelling avenues for future inquiry. Our use of scenario-based experiments, while excellent for establishing causality, necessarily limits external validity. Future research should seek to replicate and extend these findings using data from live A/I chat logs or through field experiments. Furthermore, our findings are based on a Western, English-speaking sample; future research should test the cross-cultural validity of these effects.\u003c/p\u003e\n\u003cp\u003eThe project of \u0026quot;humanizing the machine\u0026quot; is itself laden with profound ethical responsibilities. Our finding that anthropomorphic AI can foster a genuine sense of emotional affinity raises critical questions about manipulation. An AI designed to be \u0026quot;charming\u0026quot; could be deployed not to genuinely solve a problem, but to create a facade of rapport that deflects legitimate complaints. Firms must ensure that the relational warmth projected by an AI is backed by genuine procedural and distributive justice. Furthermore, as AI language models become more sophisticated, they risk perpetuating biases inherited from their training data, leading to discriminatory outcomes that are insidious and difficult to detect. Firms must therefore invest in proactive bias mitigation strategies during model training and deploy transparent audit trails for AI-driven decisions. This necessitates robust governance and continuous auditing for algorithmic bias.\u003c/p\u003e\n\u003cp\u003eThe most significant opportunities for future research lie at the intersection of our findings and the rapidly evolving technological frontier. First, the rise of \u003cstrong\u003eGenerative AI (GenAI)\u003c/strong\u003e fundamentally changes the nature of anthropomorphism. GenAI\u0026rsquo;s capacity for fluent, contextually rich conversation could dramatically enhance perceived behavioral and emotional humanness, yet its opaque reasoning could also create new, more subtle forms of the uncanny valley\u0026mdash;an interaction that is grammatically perfect but emotionally hollow. Future research must investigate how customers perceive the authenticity of GenAI-driven empathy. Second, the emergence of \u003cstrong\u003eagentic AI\u003c/strong\u003e\u0026mdash;systems capable of autonomous action\u0026mdash;presents a new frontier for which no theoretical frameworks yet exist. Finally, future studies should explore critical contingencies, such as how these dynamics differ based on the locus of the service failure (firm vs. customer). By pursuing these questions, the research community can build upon the framework established here to develop a comprehensive understanding of the evolving human-AI service relationship.\u003c/p\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eThis research offers a timely and comprehensive examination of the complex interplay between anthropomorphism and the uncanny valley in the critical context of AI-mediated service recovery. Our findings consistently demonstrate that while AI agents designed to convey warmth, empathy, and adaptive intelligence can successfully bolster perceptions of justice and foster emotional connection, these benefits are extraordinarily precarious. A human-like design creates high expectations, and even minor imperfections can trigger a powerful uncanny valley response that erodes trust and undermines the entire recovery effort. Behavioral and emotional cues, in particular, emerge as the most potent\u0026mdash;and therefore most high-stakes\u0026mdash;dimensions of anthropomorphic design. Ultimately, this study expands our theoretical perspective on AI in service research, illuminating the necessity of a carefully calibrated design that avoids unsettling imperfections. It provides clear, actionable guidance for managers seeking to harness the efficiency of AI without sacrificing the essential relational and emotional nuances that underpin successful service and enduring customer trust.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cspan\u003eEthics Approval Statement: This study was reviewed and approved by the Institutional Review Board (IRB) of Chitkara University and was performed in accordance with the ethical standards of the 1964 Helsinki Declaration and its later amendments. Participant Consent Statement: Written informed consent was obtained from all individual participants included in the study\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Profiles:\u003c/strong\u003e\u003c/p\u003e\n\u003cul type=\"disc\"\u003e\n \u003cli\u003e\u003cstrong\u003eSSRN:\u003c/strong\u003e https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=7349141\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eORCID:\u003c/strong\u003e https://orcid.org/0000-0003-0337-7885\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eResearchGate:\u003c/strong\u003e https://www.researchgate.net/profile/Jaskirat-Singh-26\u003c/li\u003e\n\u003c/ul\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBelanche, D., Casal\u0026oacute;, L. V., Flavi\u0026aacute;n, C., \u0026amp; Schepers, J. (2020). Service robot implementation: A theoretical framework and research agenda. \u003cem\u003eService Industries Journal\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(3-4), 203\u0026ndash;225.\u003c/li\u003e\n\u003cli\u003eBiocca, F., Harms, C., \u0026amp; Burgoon, J. K. (2003). Toward a more robust theory and measure of social presence: Review and suggested criteria. \u003cem\u003ePresence: Teleoperators \u0026amp; Virtual Environments\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(5), 456\u0026ndash;480.\u003c/li\u003e\n\u003cli\u003eBlodgett, J. G., Hill, D. J., \u0026amp; Tax, S. S. (1997). The effects of distributive, procedural, and interactional justice on postcomplaint behavior. \u003cem\u003eJournal of Retailing\u003c/em\u003e, \u003cem\u003e73\u003c/em\u003e(2), 185\u0026ndash;210.\u003c/li\u003e\n\u003cli\u003eBolton, R. N., \u0026amp; Mattila, A. S. (2015). How does anthropomorphism affect customer\u0026ndash;robot interactions in retail service settings? \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(4), 427\u0026ndash;446.\u003c/li\u003e\n\u003cli\u003eBrislin, R. W. (1970). Back-translation for cross-cultural research. \u003cem\u003eJournal of Cross-Cultural Psychology\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(3), 185\u0026ndash;216.\u003c/li\u003e\n\u003cli\u003eBuhrmester, M., Kwang, T., \u0026amp; Gosling, S. D. (2011). Amazon\u0026rsquo;s Mechanical Turk: A new source of inexpensive, yet high-quality, data? \u003cem\u003ePerspectives on Psychological Science\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(1), 3\u0026ndash;5.\u003c/li\u003e\n\u003cli\u003eBurgoon, J. K., \u0026amp; Jones, S. B. (1976). Toward a theory of personal space expectations and their violations. \u003cem\u003eHuman Communication Research\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(2), 131\u0026ndash;146.\u003c/li\u003e\n\u003cli\u003eColquitt, J. A. (2001). On the dimensionality of organizational justice: A construct validation of a measure. \u003cem\u003eJournal of Applied Psychology\u003c/em\u003e, \u003cem\u003e86\u003c/em\u003e(3), 386\u0026ndash;400.\u003c/li\u003e\n\u003cli\u003eDe Melo, C. M., Carnevale, P. J., Read, S. J., \u0026amp; Gratch, J. (2014). Reading people\u0026rsquo;s minds from emotion expressions in interdependent decision making. \u003cem\u003eJournal of Personality and Social Psychology\u003c/em\u003e, \u003cem\u003e106\u003c/em\u003e(1), 73\u0026ndash;88.\u003c/li\u003e\n\u003cli\u003eDe Wulf, K., Odekerken-Schr\u0026ouml;der, G., \u0026amp; Iacobucci, D. (2001). Investments in consumer relationships: A cross-country and cross-industry exploration. \u003cem\u003eJournal of Marketing\u003c/em\u003e, \u003cem\u003e65\u003c/em\u003e(4), 33\u0026ndash;50.\u003c/li\u003e\n\u003cli\u003eEpley, N., Waytz, A., \u0026amp; Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e114\u003c/em\u003e(4), 864\u0026ndash;886.\u003c/li\u003e\n\u003cli\u003eFaul, F., Erdfelder, E., Lang, A.-G., \u0026amp; Buchner, A. (2009). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. \u003cem\u003eBehavior Research Methods\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e(4), 1149\u0026ndash;1160.\u003c/li\u003e\n\u003cli\u003eGefen, D., Karahanna, E., \u0026amp; Straub, D. W. (2003). Trust and TAM in online shopping: An integrated model. \u003cem\u003eMIS Quarterly\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(1), 51\u0026ndash;90.\u003c/li\u003e\n\u003cli\u003eGo, E., \u0026amp; Carroll, J. M. (2021). Designing for reciprocal relationships with conversational AI. In \u003cem\u003eProceedings of the 2021 CHI Conference on Human Factors in Computing Systems\u003c/em\u003e (pp. 1\u0026ndash;14).\u003c/li\u003e\n\u003cli\u003eHayes, A. F. (2017). \u003cem\u003eIntroduction to mediation, moderation, and conditional process analysis: A regression-based approach\u003c/em\u003e. Guilford Publications.\u003c/li\u003e\n\u003cli\u003eHo, C.-C., \u0026amp; MacDorman, K. F. (2010). Measuring the uncanny valley effect. \u003cem\u003eInternational Journal of Social Robotics\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(2), 129\u0026ndash;139.\u003c/li\u003e\n\u003cli\u003eHo, C.-C., \u0026amp; MacDorman, K. F. (2017). Revisiting the uncanny valley theory: Developing and validating an alternative to the Godspeed indices. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e76\u003c/em\u003e, 150\u0026ndash;166.\u003c/li\u003e\n\u003cli\u003eHofstede, G. (2001). \u003cem\u003eCulture\u0026apos;s consequences: Comparing values, behaviors, institutions and organizations across nations\u003c/em\u003e. Sage Publications.\u003c/li\u003e\n\u003cli\u003eHuang, M.-H., \u0026amp; Rust, R. T. (2018). Artificial intelligence in service. \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(2), 155\u0026ndash;172.\u003c/li\u003e\n\u003cli\u003eHuang, M.-H., \u0026amp; Rust, R. T. (2021). Engaged to a robot? The role of AI in service. \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(1), 30\u0026ndash;41.\u003c/li\u003e\n\u003cli\u003eMarinova, S., de Ruyter, K., Huang, M.-H., Meuter, M. L., \u0026amp; Challagalla, G. (2017). Getting smart: Learning from technology-empowered frontline interactions. \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 29\u0026ndash;42.\u003c/li\u003e\n\u003cli\u003eMaxham III, J. G., \u0026amp; Netemeyer, R. G. (2002). A longitudinal study of complaining customers\u0026rsquo; evaluations of multiple service failures and recovery efforts. \u003cem\u003eJournal of Marketing\u003c/em\u003e, \u003cem\u003e66\u003c/em\u003e(4), 57\u0026ndash;71.\u003c/li\u003e\n\u003cli\u003eMcKnight, D. H., Choudhury, V., \u0026amp; Kacmar, C. (2002). Developing and validating trust measures for e-commerce: An integrative typology. \u003cem\u003eInformation Systems Research\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(3), 334\u0026ndash;359.\u003c/li\u003e\n\u003cli\u003eMeuter, M. L., Ostrom, A. L., Roundtree, R. I., \u0026amp; Bitner, M. J. (2000). Self-service technologies: Understanding customer satisfaction with technology-based service encounters. \u003cem\u003eJournal of Marketing\u003c/em\u003e, \u003cem\u003e64\u003c/em\u003e(3), 50\u0026ndash;64.\u003c/li\u003e\n\u003cli\u003eMori, M., MacDorman, K. F., \u0026amp; Kageki, N. (2012). The uncanny valley [from the field]. \u003cem\u003eIEEE Robotics \u0026amp; Automation Magazine\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(2), 98\u0026ndash;100.\u003c/li\u003e\n\u003cli\u003eNowak, K. L., \u0026amp; Biocca, F. (2003). The effect of the agency and anthropomorphism on users\u0026rsquo; sense of telepresence, copresence, and social presence in virtual environments. \u003cem\u003ePresence: Teleoperators and Virtual Environments\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(5), 481\u0026ndash;494.\u003c/li\u003e\n\u003cli\u003eParasuraman, A. (2000). Technology Readiness Index (TRI) a multiple-item scale to measure readiness to embrace new technologies. \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(4), 307\u0026ndash;320.\u003c/li\u003e\n\u003cli\u003ePreacher, K. J., \u0026amp; Hayes, A. F. (2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. \u003cem\u003eBehavior Research Methods\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(3), 879\u0026ndash;891.\u003c/li\u003e\n\u003cli\u003eReeves, B., \u0026amp; Nass, C. (1996). \u003cem\u003eThe media equation: How people treat computers, television, and new media like real people and places\u003c/em\u003e. Cambridge University Press.\u003c/li\u003e\n\u003cli\u003eSaygin, A. P., Chaminade, T., Ishiguro, H., Driver, J., \u0026amp; Frith, C. (2012). The thing that should not be: Predictive coding and the uncanny valley in perceiving human and humanoid robot actions. \u003cem\u003eSocial Cognitive and Affective Neuroscience\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(4), 413\u0026ndash;422.\u003c/li\u003e\n\u003cli\u003eSmith, A. K., Bolton, R. N., \u0026amp; Wagner, J. (1999). A model of customer satisfaction with service encounters involving failure and recovery. \u003cem\u003eJournal of Marketing Research\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(3), 356\u0026ndash;372.\u003c/li\u003e\n\u003cli\u003eSparks, B. A., \u0026amp; McColl-Kennedy, J. R. (2001). Justice strategy options for increased customer satisfaction in a service failure. \u003cem\u003eJournal of Business Research\u003c/em\u003e, \u003cem\u003e54\u003c/em\u003e(3), 209\u0026ndash;218.\u003c/li\u003e\n\u003cli\u003eTax, S. S., Brown, S. W., \u0026amp; Chandrashekaran, M. (1998). Customer evaluations of service complaint experiences: Implications for relationship marketing. \u003cem\u003eJournal of Marketing\u003c/em\u003e, \u003cem\u003e62\u003c/em\u003e(2), 60\u0026ndash;76.\u003c/li\u003e\n\u003cli\u003eVan Doorn, J., Mende, M., Noble, S. M., Hulland, J., Ostrom, A. L., Grewal, D., \u0026amp; Petersen, J. A. (2017). Domo arigato Mr. Roboto: Emergence of automated social presence in organizational frontlines and customers\u0026rsquo; service experiences. \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 43\u0026ndash;58.\u003c/li\u003e\n\u003cli\u003eWang, Y., \u0026amp; Siau, K. (2019). Artificial intelligence, machine learning, automation, robotics, future of work and future of humanity: A review and research agenda. \u003cem\u003eJournal of Database Management (JDM)\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e(1), 61\u0026ndash;79.\u003c/li\u003e\n\u003cli\u003eWeiner, B. (1985). An attributional theory of achievement motivation and emotion. \u003cem\u003ePsychological Review\u003c/em\u003e, \u003cem\u003e92\u003c/em\u003e(4), 548\u0026ndash;573.\u003c/li\u003e\n\u003cli\u003eXiao, L., \u0026amp; Kumar, V. (2021). Robotics for customer service: A useful complement or an ultimate substitute? \u003cem\u003eJournal of Service Research\u003c/em\u003e, \u003cem\u003e24\u003c/em\u003e(1), 9\u0026ndash;29.\u003c/li\u003e\n\u003cli\u003eZeithaml, V. A., Berry, L. L., \u0026amp; Parasuraman, A. (1996). The behavioral consequences of service quality. \u003cem\u003eJournal of Marketing\u003c/em\u003e, \u003cem\u003e60\u003c/em\u003e(2), 31\u0026ndash;46.\u003c/li\u003e\n\u003cli\u003eZłotowski, J., Proudfoot, D., Yogeeswaran, K., \u0026amp; Bartneck, C. (2015). Anthropomorphism: Opportunities and challenges in human\u0026ndash;robot interaction. \u003cem\u003eAI \u0026amp; Society\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e(3), 347\u0026ndash;360.\u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eAuthor Publications (Corresponding Author) and Working Paper References:\u003c/strong\u003e\u003c/li\u003e\n\u003cli\u003eSingh, J. (2025). \u003cem\u003eLeveraging Financial Innovation and Capability Enhancement for Sustainable Urban Poverty Alleviation in Punjab: Policy Imperatives\u003c/em\u003e. SSRN Scholarly Paper 5205435. Social Science Research Network. https://doi.org/10.2139/ssrn.5205435\u003c/li\u003e\n\u003cli\u003eSingh, J., Batra, G. S., \u0026amp; Chatrath, S. K. (2025a). \u003cem\u003eBlockchain\u0026rsquo;s Role in Social Welfare, Financial Inclusion, and Public Sector Innovations in India: A Multi-Sector Analysis of Government-Led Initiatives\u003c/em\u003e. SSRN Scholarly Paper 5105250. Social Science Research Network. https://doi.org/10.2139/ssrn.5105250\u003c/li\u003e\n\u003cli\u003eSingh, J., Batra, G. S., \u0026amp; Chatrath, S. K. (2025b). \u003cem\u003eDigital Capabilities for Urban Poverty Alleviation: Integrating E-Payment Awareness and Credit Utilization Patterns in Indian Slums\u003c/em\u003e. SSRN Scholarly Paper 5223860. Social Science Research Network. https://doi.org/10.2139/ssrn.5223860\u003c/li\u003e\n\u003cli\u003eSingh, J., Batra, G. S., \u0026amp; Chatrath, S. K. (2025c). \u003cem\u003eHarnessing Fintech for Poverty Alleviation: Enhancing Credit Utilization and Livelihoods in Urban Slums of North-western India through the Capability Approach and Sustainable Livelihoods Framework\u003c/em\u003e. SSRN Scholarly Paper 5207210. Social Science Research Network. https://doi.org/10.2139/ssrn.5207210\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Sharma, D. (2024). Contemporary Challenges of Management Education in India: Review and Assessment. In \u003cem\u003eInterdisciplinary Approaches in Management Education\u003c/em\u003e (pp. 149\u0026ndash;167).\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Singh, M. (2024a). Accelerating Financial Inclusion of the Urban Poor: Role of Innovative e-Payment Systems and JAM Trinity in Alleviating Poverty in India. \u003cem\u003eGlobal Business Review\u003c/em\u003e, 09721509231222609. https://doi.org/10.1177/09721509231222609\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Singh, M. (2024b). Addressing unproductive credit consumption and beneficiary malpractices in social welfare programs for slum-dwellers: A study from India. \u003cem\u003eCities\u003c/em\u003e, 145, 104729. https://doi.org/10.1016/j.cities.2023.104729\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Singh, M. (2024c). Alleviating urban poverty in India: The role of capabilities and entrepreneurship development. \u003cem\u003eInternational Journal of Social Economics\u003c/em\u003e, 51(10), 1314\u0026ndash;1335. https://doi.org/10.1108/IJSE-07-2023-0514\u003c/li\u003e\n\u003cli\u003eSingh, V., \u0026amp; Singh, J. (2024). Quantifying the relationship between e-advertising capabilities and marketing mix cost savings. \u003cem\u003eInternational Journal of Applied Management Science\u003c/em\u003e, 16(1), 44\u0026ndash;67. https://doi.org/10.1504/IJAMS.2024.136149\u003c/li\u003e\n\u003cli\u003eSingh, J., Sharma, D., \u0026amp; Batra, G. S. (2023). Does Credit Utilization Pattern Promote Poverty Alleviation? An Evidence from India. \u003cem\u003eGlobal Business Review\u003c/em\u003e, 24(6), 1227\u0026ndash;1250. https://doi.org/10.1177/0972150920918967\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Singh, M. (2023a). Does financial inclusion impact socio-economic stability? A study of social safety net in Indian slums. \u003cem\u003eInternational Journal of Social Economics\u003c/em\u003e, 50(8), 1060\u0026ndash;1084. https://doi.org/10.1108/IJSE-04-2022-0261\u003c/li\u003e\n\u003cli\u003eSingh, J., \u0026amp; Singh, M. (2023b). Fintech applications in social welfare schemes during Covid times: An extension of the classic TAM model in India. \u003cem\u003eInternational Social Science Journal\u003c/em\u003e, 73(250), 979\u0026ndash;998. https://doi.org/10.1111/issj.12406\u003c/li\u003e\n\u003cli\u003eSingh, J., Batra, G. S., Sharma, D., \u0026amp; Singh, V. (2021). \u003cem\u003eMicrocredit Usage Pattern and its Impact on Economic Activities of the Urban Deprived: A Study of Punjab State, India\u003c/em\u003e. SSRN Scholarly Paper 5206281. Social Science Research Network. https://doi.org/10.2139/ssrn.5206281\u003c/li\u003e\n\u003cli\u003eSharma, D., \u0026amp; Singh, J. (2018). Credit expansion programmes for urban poor: A literature review and conceptual framework. \u003cem\u003eInternational Journal of Advanced in Management, Technology and Engineering Sciences (IJAMTES)\u003c/em\u003e, 8(IV), 646\u0026ndash;657.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u0026nbsp;\u003c/strong\u003e\u003cem\u003eHypotheses Summary\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eHypothesis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDescription\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eTheoretical Foundation(s)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnthropomorphism \u0026rarr; Increased justice perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSocial Presence Theory (Biocca et al., 2003)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnthropomorphism \u0026rarr; Increased emotional affinity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCASA Paradigm (Reeves \u0026amp; Nass, 1996)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny valley triggers \u0026rarr; Decreased justice perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eExpectancy Violation Theory (Burgoon \u0026amp; Jones, 1976)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny valley triggers \u0026rarr; Decreased emotional affinity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny Valley Theory (Mori et al., 2012)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH5\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigher anthropomorphism amplifies uncanny valley effects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eExpectancy Violation Theory; Uncanny Valley Theory\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH6\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eJustice perceptions \u0026rarr; Increased recovery satisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eJustice Theory (Colquitt, 2001; Smith et al., 1999)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH7\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEmotional affinity \u0026rarr; Increased recovery satisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eService Recovery Literature (Van Doorn et al., 2017)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH8\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRecovery satisfaction \u0026rarr; Increased trust in AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTrust in Technology (McKnight et al., 2002)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH9\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRecovery satisfaction \u0026rarr; Increased repatronage intentions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRelationship Marketing (Zeithaml et al., 1996)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH10\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTech. Savviness moderates Anthropomorphism effects (+)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTechnology Acceptance (Parasuraman, 2000)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH11\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTech. Savviness moderates Uncanny Valley effects (-)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eIndividual Differences (Meuter et al., 2000)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eH12\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrior Relationship moderates Satisfaction \u0026rarr; Repatronage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRelationship Strength (De Wulf et al., 2001)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u0026nbsp;\u003c/strong\u003e\u003cem\u003eSummary of Experimental Manipulation Checks\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eConstruct\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eManipulation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation (Check Item \u0026amp; Result)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEXPERIMENT 1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnthropomorphism\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh: Realistic avatar, empathic language, personal pronouns Low: Generic icon, formal phrasing, impersonal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eParticipants rated perceived human-likeness. High vs. Low differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny Valley\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePresent: Subtle yet unsettling glitches (frozen avatar, odd wording) Absent: Smooth, coherent AI responses\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eParticipants rated \u0026ldquo;eerie\u0026rdquo; or \u0026ldquo;creepy\u0026rdquo; feelings. Present vs. Absent differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEXPERIMENT 2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePhysical\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh: Realistic human avatar with natural facial proportions Low: Simple or abstract icon (robotic outline)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ldquo;The AI\u0026rsquo;s avatar looked human-like.\u0026rdquo; High vs. Low differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBehavioral\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh: Casual, conversational style, personal greetings Low: Robotic, formulaic script\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ldquo;The AI used natural conversation.\u0026rdquo; High vs. Low differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEmotional\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh: Empathy statements (\u0026ldquo;I understand how frustrating\u0026hellip;\u0026rdquo;) Low: Minimal or no empathic content\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ldquo;The AI recognized/acknowledged my feelings.\u0026rdquo; High vs. Low differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCognitive\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eHigh: Adaptive problem-solving, personalized solutions Low: Standard responses, no personalization\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026ldquo;The AI customized solutions to my problem.\u0026rdquo; High vs. Low differed significantly (p \u0026lt; .001).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u0026nbsp;\u003c/strong\u003e\u003cem\u003eConstruct Measures and Reliability\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eConstruct\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSample Item\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSource(s)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eReliability\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eJustice Perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;The solution I received from the AI was fair.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eColquitt (2001); Tax et al. (1998)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .89\u0026ndash;.93\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eEmotional Affinity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;I felt a personal connection with the AI during the interaction.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eVan Doorn et al. (2017)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .94\u0026ndash;.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eRecovery Satisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;Overall, I am satisfied with the AI\u0026rsquo;s handling of my complaint.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eMaxham \u0026amp; Netemeyer (2002)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .92\u0026ndash;.96\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eTrust in AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;I believe this AI agent is competent in resolving issues.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eMcKnight et al. (2002)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .90\u0026ndash;.92\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eRepatronage Intentions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;I would use this service again in the future.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eZeithaml et al. (1996)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .92\u0026ndash;.94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003eTechnological Savviness\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;I am comfortable learning new technologies quickly.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eMeuter et al. (2000); Parasuraman (2000)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .85\u0026ndash;.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 170px;\"\u003e\n \u003cp\u003ePrior Relationship\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 189px;\"\u003e\n \u003cp\u003e\u0026ldquo;I feel a sense of loyalty to this company.\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003eDe Wulf et al. (2001)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 101px;\"\u003e\n \u003cp\u003e\u0026alpha; = .88\u0026ndash;.90\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"4\" valign=\"top\" style=\"width: 601px;\"\u003e\n \u003cp\u003e\u003cem\u003eNote\u003c/em\u003e: Confirmatory Factor Analyses (CFAs) in each experiment indicated that all constructs achieved good fit indices (CFI \u0026gt; 0.90, RMSEA \u0026lt; 0.08), supporting construct validity.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u0026nbsp;\u003c/strong\u003e\u003cem\u003eExperiment 1: Key ANOVA Results (N = 484)\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDependent Variable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSource of Variance\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eF-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEffect Size (\u0026eta;\u0026sup2;)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eJustice Perceptions\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnthropomorphism (A)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e144.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.23\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny Valley (UV)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97.55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eEmotional Affinity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAnthropomorphism (A)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e186.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny Valley (UV)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e121.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"5\" valign=\"top\"\u003e\n \u003cp\u003e\u003cem\u003eNote\u003c/em\u003e: The interaction term (A \u0026times; UV) was not statistically significant for either dependent variable (all p \u0026gt; .10).\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 5\u0026nbsp;\u003c/strong\u003e\u003cem\u003eExperiment 2: Key Path and Moderation Effects (N = 776)\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" class=\"fr-table-selection-hover\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eRelationship\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026beta;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eSupported?\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eBehavioral Anthropomorphism \u0026rarr; Justice Perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEmotional Anthropomorphism \u0026rarr; Justice Perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eUncanny (Mismatch) \u0026rarr; Justice Perceptions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-.33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eJustice Perceptions \u0026rarr; Recovery Satisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.43\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eEmotional Affinity \u0026rarr; Recovery Satisfaction\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRecovery Satisfaction \u0026rarr; Trust in AI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eRecovery Satisfaction \u0026rarr; Repatronage Intentions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModeration Effects\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTech. Savviness \u0026times; Anthropomorphism \u0026rarr; Justice\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eTech. Savviness \u0026times; Uncanny \u0026rarr; Justice\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e-.15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.037\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePrior Rel. \u0026times; Recovery Satisfaction \u0026rarr; Repatronage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026lt; .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eYes\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Service Recovery, Anthropomorphism, Uncanny Valley, Justice Theory, Human-AI Interaction, Expectancy Violation Theory","lastPublishedDoi":"10.21203/rs.3.rs-6889879/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6889879/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAs sophisticated Artificial Intelligence takes on the critical task of service recovery, the design strategy of anthropomorphism presents both a compelling opportunity and a significant risk. An imperfectly human AI can provoke an \u0026ldquo;uncanny valley\u0026rdquo; response, turning a recovery attempt into a more alienating experience. Integrating justice theory with social presence and expectancy violation frameworks, our two experimental studies (N\u0026thinsp;=\u0026thinsp;500; N\u0026thinsp;=\u0026thinsp;800) dissect this crucial tension. Our results reveal a key fragility: the positive influence of behavioral and emotional anthropomorphism on justice and affinity perceptions is entirely contingent on flawless execution. Minor glitches in highly human-like agents negate these benefits by violating customer expectations of authentic interaction. This research offers a robust theoretical model of the contingent nature of human-like AI design and provides clear, actionable principles for creating service agents that enhance, rather than undermine, customer relationships during recovery.\u003c/p\u003e","manuscriptTitle":"Humanizing the Machine: Anthropomorphism and the Uncanny Valley in AI-Mediated Service Recovery","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-06-18 06:48:41","doi":"10.21203/rs.3.rs-6889879/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"efba19be-c307-461e-91a9-a7b5ee5ea28c","owner":[],"postedDate":"June 18th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":50025955,"name":"Artificial Intelligence and Machine Learning"},{"id":50025956,"name":"Marketing"},{"id":50025957,"name":"Management"},{"id":50025958,"name":"International Business"},{"id":50025959,"name":"Other Business"}],"tags":[],"updatedAt":"2025-06-18T06:48:41+00:00","versionOfRecord":[],"versionCreatedAt":"2025-06-18 06:48:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6889879","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6889879","identity":"rs-6889879","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}