Driving Service Stickiness in the AI Subscription Economy: The Roles of Algorithmic Curation, Technological Fluidity, and Cognitive Efficiency

Abstract

This study examines the psychological mechanisms underlying service stickiness during the mature phase of the AI subscription economy, with particular attention to the paradox of subscription fatigue. To enhance conceptual clarity, AI-driven stimuli—specifically Algorithmic Curation and Technological Fluidity—are defined as perceived attributes at the individual level. Employing the Stimulus–Organism–Response (S-O-R) framework, the research explores how these perceived stimuli influence consumers’ internal states (Cognitive Efficiency and Serendipity) and subsequent behavioral responses (Service Stickiness). Empirical analysis using partial least squares structural equation modeling (PLS-SEM) on data from U.S. subscription service users yields several theoretical insights. Cognitive Efficiency is identified as the primary driver of stickiness, indicating that, in the context of subscription fatigue, the utilitarian benefit of reduced cognitive effort surpasses hedonic enjoyment. Additionally, the study identifies a “Frictionless Trap,” in which excessive Technological Fluidity negatively affects Serendipity (β = −0.195), suggesting that an entirely seamless experience may create a filter bubble that limits unexpected discovery. As a result, Serendipity does not significantly affect stickiness in the aggregate model. However, post hoc analysis demonstrates that Serendipity remains significant for high-income users, while Cognitive Efficiency is most influential in high-frequency utilitarian contexts, such as food services. These findings indicate that sustainable retention depends on reducing cognitive load while intentionally introducing friction to preserve opportunities for discovery.

1. Introduction

The rapid advancement of artificial intelligence (AI) and big data technologies is fundamentally transforming electronic commerce, positioning the “Subscription Economy” at the forefront of this disruption. In contrast to traditional models focused on recurring delivery, contemporary AI-based subscription services have developed into “Intelligent Curation Platforms.” These platforms utilize advanced algorithms to anticipate consumers’ latent needs and provide highly personalized experiences [1]. This technological progression has shifted consumer values from ownership to access and experience, fueling significant growth across multiple industries [2].

Despite this market expansion, a paradoxical side effect known as ‘Subscription Fatigue’ has emerged [3]. Recent studies indicate that consumers experience cognitive overload and psychological exhaustion when managing a fragmented array of subscription services. This fatigue has become a primary driver of service churn [4]. As the subscription economy matures, the strategic focus for firms has shifted from merely attracting new users to securing “Service Stickiness,” which involves retaining fatigued consumers within the service ecosystem [5].

Previous literature on subscription services has primarily utilized the Technology Acceptance Model (TAM) and the Expectation-Confirmation Model (ECM), emphasizing the roles of perceived usefulness and enjoyment in initial adoption [6,7]. More recently, these frameworks have been extended to examine continuance intentions within the subscription context [8,9]. Some researchers contend that hedonic values, such as the “serendipity” generated by AI recommendations, are central to customer satisfaction [10]. However, this perspective does not fully address the current market environment, which is increasingly defined by subscription fatigue. For consumers facing information overload and decision fatigue, the principal value of AI may lie in delivering “Cognitive Efficiency” by reducing the complexity of decision-making.

To address the limitations of prior research that has primarily relied on rational decision-making models, this study applies the Stimulus–Organism–Response (S-O-R) framework to empirically examine how the technological characteristics of AI-based subscription services influence service stickiness through consumer psychological mechanisms. In contrast to the Technology Acceptance Model (TAM) [6] and the Expectation-Confirmation Model (ECM) [7], which emphasize conscious evaluation and intentional continuance, the S-O-R framework [11] is better suited to capturing the subconscious and reactive aspects of user retention in high-frequency AI environments. Incorporating the S-O-R perspective [12] enables a deeper exploration of the psychological dimensions underlying ‘stickiness’.

This study redefines the technological stimuli of AI as ‘Algorithmic Curation’ (the ability to analyze user preferences) and ‘Technological Fluidity’ (the seamlessness of the user experience). Critically, this study posits that user behavior is influenced not by AI’s objective technical specifications, but by users’ subjective perceptions of these attributes. Therefore, we conceptualize Algorithmic Curation and Technological Fluidity as perceived constructs at the individual level. This approach aligns with experiential outcomes such as Serendipity within the S-O-R framework and addresses the level-of-analysis challenge common in techno-centric research. Additionally, consumers’ internal responses are categorized into ‘Cognitive Efficiency’ (reducing information-processing costs) and ‘Serendipity’ (the joy of unexpected discovery). This model seeks to answer a central question: “What truly makes consumers stick—efficiency or enjoyment?”

This study makes several distinct contributions to the field. First, consistent with recent trends in e-commerce research, traditional concepts of convenience and personalization are redefined as “Technological Fluidity” and “Algorithmic Curation” within the AI context. Second, the prevailing focus on hedonic value is challenged by demonstrating that the utilitarian value of “Cognitive Efficiency” is the primary driver of retention in the context of subscription fatigue. Third, exploratory analyses by service type (e.g., Food versus Non-Food) offer nuanced insights into how consumer subscription characteristics moderate the acceptance of AI-generated value.

To address these research objectives, this study poses three key research questions. First, it examines how AI-driven stimuli, specifically Algorithmic Curation and Technological Fluidity, influence users’ internal organismic states of Cognitive Efficiency and Serendipity. Second, it investigates how these organismic states subsequently drive Service Stickiness within the subscription economy. Finally, it explores whether the type of subscription service moderates the relationship between these cognitive and emotional drivers and stickiness.

The remainder of this paper is structured as follows. Section 2 reviews the theoretical background. Section 3 presents the research model and hypotheses. Section 4 outlines the methodology, and Section 5 reports the results of the data analysis. Section 6 discusses the findings, implications, and limitations.

2. Theoretical Background

2.1. Stimulus–Organism–Response (S-O-R) Framework

The Stimulus–Organism–Response (S-O-R) framework, originating from environmental psychology, posits that external environmental stimuli (Stimulus) influence an individual’s internal state (Organism), which subsequently leads to behavioral responses (Response) [11]. This framework has been widely applied to explain consumer behavior in online shopping and digital service environments [12]. In the context of this study, we operationalize the Stimulus (S) not as objective technological features, but as perceived environmental stimuli. This distinction is critical because environmental stimuli affect organismic states only when they are consciously processed and interpreted by the individual. Accordingly, rather than treating ‘Algorithmic Curation’ and ‘Technological Fluidity’ as system-level attributes, we define them as distinct perceptions formed through users’ interactions with the AI system.

Crucially, while the S-O-R framework is traditionally conceptualized as a linear sequence, we recognize that human-AI interaction is inherently recursive and adaptive. User responses (R) generate feedback data that updates the algorithmic stimuli (S), resulting in a continuous feedback loop of mutual adaptation. However, to empirically validate the specific causal mechanisms underlying stickiness at the maturity stage, this study captures a cross-sectional snapshot of this dynamic. We thus frame the current model as a critical cycle within the broader evolutionary process of user retention.

These perceived factors shape consumers’ cognitive and emotional values—the organism (O)—which, in turn, drive behavioral responses (R) such as continuance intention. Accordingly, this study develops a research model within the context of the AI subscription economy by designating ‘Algorithmic Curation’ and ‘Technological Fluidity’ as stimulus factors, ‘Cognitive Efficiency’ and ‘Serendipity’ as organism factors, and ‘Service Stickiness’ as the response factor.

2.2. AI-Driven Stimuli: Algorithmic Curation and Technological Fluidity

The evolution of the subscription economy parallels the advancement of AI technology. This study defines the core stimulus factors provided by AI as perceived Algorithmic Curation and Technological Fluidity, focusing on the user’s subjective assessment of information-processing capabilities and process optimization. First, Algorithmic Curation is defined as the user’s perception of the system’s ability to accurately analyze past data and behavioral patterns to select optimal alternatives that meet latent needs. This goes beyond simple customization; it involves a perceived active curation process where the system discovers and learns preferences the consumer may not yet be aware of. While existing research posits that personalized services enhance perceived usefulness, this study argues that users’ awareness of AI’s analytical competence serves as a stimulus that dramatically reduces search costs. Second, Technological Fluidity refers to the user’s subjective assessment of a state where the entire process—from search and payment to delivery and post-purchase care—connects seamlessly without intervention. This concept expands upon traditional convenience, emphasizing a perceived ‘unconscious’ service experience achieved through AI prediction and coordination.

2.3. Internal Organism: Cognitive Efficiency and Serendipity

When exposed to these perceived AI stimuli, consumers experience both rational (utilitarian) and emotional (hedonic) reactions. Cognitive Efficiency refers to the extent to which consumers perceive that they have saved mental energy and time during the decision-making process. According to the ‘cognitive miser’ theory [13], humans instinctively seek to minimize cognitive effort. In the context of the ‘paradox of choice,’ AI subscription services provide significant utility by filtering options and making decisions on behalf of users, thereby relieving perceived cognitive load. Conversely, Serendipity denotes the subjective feeling of joy and excitement derived from unexpected discoveries. This represents a form of hedonic value, particularly significant in random-box or content-recommendation services where consumers anticipate novelty. When users perceive recommendations as unexpectedly relevant, it triggers surprise and delight, fostering an emotional attachment to the service.

2.4. Service Stickiness as a Behavioral Response

Service Stickiness is operationalized as a psychological and behavioral state in which a user is deeply embedded within a specific service ecosystem, creating a high barrier to switching. This concept extends beyond simple continuance intention; it encompasses a consolidated state where the service becomes an integral part of the user’s daily routine, often leading to structural lock-in. In the subscription economy, stickiness is the critical determinant of recurring revenue. In this study’s framework, we posit that the organismic states of Cognitive Efficiency and Serendipity function as the primary antecedents that cultivate this enduring stickiness.

2.5. Subscription Fatigue as a Contextual Moderator

Subscription Fatigue, a phenomenon driven by market saturation, is defined as the psychological strain and information overload consumers experience when managing a fragmented array of subscription services. As fatigue levels rise, consumers exhibit a heightened preference for cognitive economy and automation over complex interactivity. Consequently, this study hypothesizes that Subscription Fatigue moderates the S-O-R pathway: specifically, as fatigue increases, the positive influence of AI-driven ‘Cognitive Efficiency’ on Service Stickiness will be significantly amplified. This suggests that under conditions of digital exhaustion, consumers prioritize the utilitarian value of ‘hassle resolution’ over hedonic enjoyment.

3. Research Model and Hypotheses

3.1. Research Model

This study proposes a research model to elucidate the mechanism of service stickiness in the AI-based subscription economy, grounded in the Stimulus–Organism–Response (S-O-R) framework. In constructing the research model, we explicitly distinguish between perceived technological stimuli and user responses to ensure theoretical rigor.

As illustrated in Figure 1, we define the core perceived technological attributes of AI—‘Algorithmic Curation’ and ‘Technological Fluidity’—as external stimuli (Stimulus). Crucially, we position ‘Serendipity’ not as an external stimulus but as an internal organism. While Algorithmic Curation and Technological Fluidity originate from the system’s functional capabilities, in this study, they are operationalized as the user’s subjective perception of those capabilities. Conversely, Serendipity represents the user’s internal affective state derived from the interaction. This classification aligns with the established S-O-R tradition of modeling both cognitive (e.g., Cognitive Efficiency) and affective (e.g., Serendipity) states as distinct organismic responses [12].

Therefore, we posit that these perceived technological stimuli influence consumers’ internal states—categorized into ‘Cognitive Efficiency’ (Cognitive Organism) and ‘Serendipity’ (Affective Organism)—which subsequently drive the behavioral response of ‘Service Stickiness’ (Response). Furthermore, we exploratorily verify the moderating effects of Service Type (Food vs. Non-Food) and User Characteristics (Income Level) on these pathways.

3.2. Hypotheses Development

3.2.1. The Effect of AI Stimuli on Organismic States

Algorithmic Curation is defined in this study as the user’s perception of how accurately the AI system analyzes their data to select optimal alternatives. Recent literature suggests that when users perceive curation as sophisticated, it significantly enhances decision fluency by reducing the cognitive cost of information search [14,15]. Moreover, unexpected recommendations perceived as relevant by the user can lead to serendipity—a pleasant surprise [16]. Therefore, we propose the following hypotheses:

H1: 

Perceived Algorithmic Curation has a positive effect on Cognitive Efficiency.

H2: 

Perceived Algorithmic Curation has a positive effect on Serendipity.

Technological Fluidity refers to the user’s subjective assessment of how seamlessly the service usage process connects without interruption. This provides a perceived ‘frictionless’ experience that minimizes the need for conscious intervention, thereby reducing cognitive load [17]. While previous studies suggest that smoother technological flow generally leads users to experience flow and positive emotions [18], we posit that high fluidity will primarily influence efficiency and potentially hedonic experiences.

H3: 

Perceived Technological Fluidity has a positive effect on Cognitive Efficiency.

H4: 

Perceived Technological Fluidity has a positive effect on Serendipity.

3.2.2. The Effect of Organismic States on Service Stickiness

Cognitive Efficiency represents the user’s internal realization of the mental energy-saving effect provided by the service. From the perspective of the ‘cognitive miser,’ perceived efficiency acts as a powerful switching barrier, preventing consumers from leaving the service [19]. Recent studies [20] also support that the continuance intention of AI services is more strongly influenced by perceived efficiency and usefulness than by playfulness [20].

H5: 

Cognitive Efficiency has a positive effect on Service Stickiness.

Serendipity denotes the subjective feeling of positive surprise experienced during service usage. In the marketing literature, emotional values, such as enjoyment, have been regarded as key factors in enhancing customer loyalty and stickiness [21]. Therefore, we expect that higher levels of perceived serendipity will strengthen the psychological bond with the service.

H6: 

Serendipity has a positive effect on Service Stickiness.

3.2.3. Exploratory Analysis of Moderating Effects

Beyond the baseline S-O-R pathways, this study explores how Service Type (Food vs. Non-Food) and the User’s Income Level influence the stickiness mechanism. To provide concrete practical implications, we compare high-frequency categories such as food services, which demand recurring decision-making, with other services. Furthermore, we analyze how value perception differs across economic demographics to identify boundary conditions for efficiency and serendipity.

4. Methodology

4.1. Data Collection and Sample

To empirically investigate the psychological mechanisms and behavioral responses of consumers in the AI-based subscription economy, we conducted an online survey targeting U.S. consumers. Data collection was conducted in April 2025, focusing on adults who had paid for and used subscription services (e.g., meal kits, OTT platforms, or regular delivery of daily necessities) within the past six months. A multi-stage screening process was employed to exclude insincere responses, resulting in a final dataset of 113 valid responses.

Given the sample size (n = 113), we adopted Partial Least Squares Structural Equation Modeling (PLS-SEM) for data analysis. PLS-SEM is specifically optimized for exploratory research and is widely recognized for its high statistical power and robustness against non-normal data distributions, making it superior to covariance-based SEM for studies with smaller sample sizes [22,23]. To rigorously verify the adequacy of our sample size, we conducted a post hoc power analysis using G*Power 3.1 software [24]. Based on Cohen’s [25] standard criteria for behavioral research—a medium effect size (f²) of 0.15, a significance level (α) of 0.05, a statistical power (1−β) of 0.80, and a conservative estimate of 5 predictors—the analysis indicated that a minimum of 92 participants was required. Our sample of 113 exceeds this threshold, confirming that the study possesses sufficient statistical power to validate the structural model.

Table 1. Demographic Profile of Respondents (n = 113).

Characteristics	Category	Frequency	Percentage (%)
Gender	Male	70	61.9
	Female	43	38.1
Age	20–29	29	25.7
	30–39	49	43.4
	40–49	22	19.5
	50–59	12	10.6
	60 and above	1	0.9
Education	High school or less	12	10.6
	College/Vocational degree	14	12.4
	Bachelor’s degree	71	62.8
	Master’s degree or above	16	14.2
Annual Income	Less than USD 30,000	16	14.2
	USD 30,000– USD 69,999	50	44.2
	USD 70,000– USD 89,999	15	13.3
	Over USD 90,000	32	28.3
Subscription Type	Food & Groceries	58	51.3
	Non-Food (Fashion, Beauty, etc.)	55	48.7
Subscription Duration	Less than 1 year	18	15.9
	1–3 years	71	62.8
	Over 3 years	24	21.2

presents the demographic profile of the respondents. The sample comprised 61.9% males and 38.1% females. In terms of age, individuals in their 30s constituted the largest group (43.4%), followed by those in their 20s (25.7%) and 40s (19.5%), reflecting the trend that digital-native Millennials and Gen Z are the primary consumers in the subscription economy. Regarding education, 77.0% held a bachelor’s degree or higher, indicating a highly educated sample. Critically, to facilitate the analysis of moderating effects, we balanced the sample between the high-involvement ‘Food & Groceries’ category (51.3%, n = 58), which involves daily dietary decisions, and ‘Non-Food’ services (48.7%, n = 55), including fashion, beauty, and hobbies.

Finally, to rule out potential systematic bias arising from the gender imbalance (Male: 61.9% vs. Female: 38.1%), we conducted an independent samples t-test on the key constructs. As presented in

Table 2. Results of Independent Samples t-test by Gender.

Construct	Gender	n	Mean	SD	t-Value	p-Value
Algorithmic Curation	Male	70	4.17	0.59	−0.362	0.718
	Female	43	4.22	0.69
Technological Fluidity	Male	70	4.21	0.62	−0.774	0.441
	Female	43	4.30	0.63
Cognitive Efficiency	Male	70	4.13	0.51	0.816	0.417
	Female	43	4.03	0.66
Serendipity	Male	70	3.76	0.74	0.543	0.589
	Female	43	3.67	0.90
Service Stickiness	Male	70	4.20	0.64	−0.427	0.670
	Female	43	4.26	0.83

, the results showed no statistically significant differences (p > 0.05) between male and female respondents across all major variables. This empirical evidence confirms that gender distribution did not distort the structural relationships proposed in the model.

4.2. Measures

To ensure psychometric rigor, we adapted all measurement items from validated scales in prior literature, contextualizing them to fit the mature phase of the AI-driven subscription economy and the phenomenon of subscription fatigue. All items were measured using a 5-point Likert scale ranging from ‘1 = Strongly Disagree’ to ‘5 = Strongly Agree.’ To ensure construct validity and reliability, we refined the scale through preliminary analysis, retaining only items with robust factor loadings. The final constructs and their corresponding measurement items are presented in

Table 3. Construct Definitions and Measurement Items.

Construct	Item Code	Measurement Item	Source
Algorithmic Curation (AC)	AC1	This service’s AI algorithm accurately analyzes my tastes and preferences.	[26]
	AC2	The more I use this service, the better the recommendation system understands my intent.
	AC3	The personalized recommendations provided by this service feel as sophisticated as if curated by an expert.
Technological Fluidity (TF)	TF1	Thanks to the analytical information provided by AI, I can skip the complex process of comparing alternatives or agonizing over choices.	[27]
	TF2	The subscription renewal or payment process is handled automatically and fluidly by the system without requiring my intervention.
	TF3	All post-purchase processes, such as delivery and exchange, are predicted and perfectly coordinated by the system.
	TF4	The entire process of using this service flows seamlessly as one complete experience from start to finish.
Cognitive Efficiency (CE)	CE1	It helps me make optimal choices without complicated comparison processes.	[28]
	CE2	It makes my daily decision-making process simple and efficient.
Serendipity (SER)	SER1	I find joy in discovering completely unexpected new tastes through this service.	[10]
	SER2	The unexpected recommendations provided by this service give me a pleasant surprise.
	SER3	I look forward to what new products (or content) will be recommended next.
Service Stickiness (STI)	STI1	It is hard to imagine my daily life without this service.	[7]
	STI2	I feel deeply connected to this service, both emotionally and functionally.
	STI3	For the time being, I will not consider any alternatives other than this service.
Subscription Fatigue (FAT) (Measured but used for context)	FAT1	I feel overwhelmed by the number of subscription services I currently manage.	[3]
	FAT2	Checking monthly subscription fees and renewal dates is mentally exhausting.
	FAT3	I feel fatigued by the flood of subscription-based content or product information.
	FAT4	I often feel the urge to go on a digital diet or cancel subscriptions to reduce information overload.

Note: All constructs represent the user’s subjective perceptions of the AI system rather than objective technical specifications.

.

First, regarding the perceived stimulus variables, Algorithmic Curation (AC) was measured using three items adapted from Ramkumar & Woo [26]. These items assess the user’s perception of the AI system’s ability to accurately analyze tastes and identify latent needs. Technological Fluidity (TF) was measured using four items based on the online convenience scale by Jiang et al. [27]. We selected these items to capture the perceived seamless connectivity and frictionless experience throughout the entire service journey.

Second, for the internal organism variables, Cognitive Efficiency (CE) was measured using two core items derived from Overby & Lee [28]. Measurement focused on the user’s perception of reduced mental fatigue and the optimization of choice in the decision-making process. Serendipity (SER) was assessed using three items from Kim et al. [10] that capture the subjective feelings of joy and anticipation derived from unexpected discoveries.

Third, for the behavioral response variable, Service Stickiness (STI) was measured using three items developed by advancing the continuance intention scale of Bhattacherjee [7]. These items extend beyond simple repurchase intention to quantify the psychological bond and the perceived irreplaceability of the service.

Finally, we collected demographic data, including subscription type (Food vs. Non-Food) and Income Level, to serve as moderating variables for the post hoc analysis.

5. Results

5.1. Measurement Model Assessment

We employed SmartPLS 4.0 to evaluate the reliability and validity of the measurement model. To assess internal consistency and convergent validity, we examined Factor Loadings, Cronbach’s Alpha, Composite Reliability (CR), and Average Variance Extracted (AVE). Preliminary analysis ensured measurement rigor by refining the constructs and eliminating items with low factor loadings or significant cross-loadings. s presented in

Table 4. Measurement Model Assessment.

Construct	Item	Factor Loading	Cronbach’s α	Composite Reliability (CR)	Average Variance Extracted (AVE)
Algorithmic Curation (AC)	AC1	0.696	0.749	0.750	0.500
	AC2	0.741
	AC3	0.682
Technological Fluidity (TF)	TF1	0.689	0.814	0.814	0.523
	TF2	0.706
	TF3	0.751
	TF4	0.745
Cognitive Efficiency (CE)	CE1	0.759	0.732	0.731	0.576
	CE2	0.759
Serendipity (SER)	SER1	0.792	0.760	0.777	0.547
	SER2	0.860
	SER3	0.523
Service Stickiness (STI)	STI1	0.814	0.888	0.888	0.726
	STI2	0.883
	STI3	0.858

Note: All factor loadings are significant at p < 0.001.

, all factor loadings for the retained items exceeded the recommended threshold of 0.6 (ranging from 0.682 to 0.888) and were statistically significant at the 0.001 level (p < 0.001). Furthermore, both Cronbach’s α and Composite Reliability (CR) values surpassed the 0.7 threshold for all constructs, confirming high internal consistency. Finally, the Average Variance Extracted (AVE) values ranged from 0.500 to 0.726, exceeding the 0.50 benchmark, thereby demonstrating adequate convergent validity [29].

Subsequently, we assessed discriminant validity using the Fornell-Larcker criterion. As reported in

Table 5. Discriminant Validity (Fornell-Larcker Criterion).

Construct	AC	TF	CE	SER	STI
Algorithmic Curation (AC)	0.707
Technological Fluidity (TF)	0.526	0.723
Cognitive Efficiency (CE)	0.532	0.397	0.759
Serendipity (SER)	0.402	0.110	0.492	0.739
Service Stickiness (STI)	0.608	0.501	0.450	0.291	0.852

Note: The bold values on the diagonal represent the square root of the AVE.

, the square root of the AVE for each latent variable (bold values on the diagonal) exceeded the corresponding inter-construct correlations in the off-diagonal elements. This result confirms that each construct is distinct from the others, thereby establishing the measurement model’s discriminant validity [29].

5.2. Structural Model Assessment

Prior to hypothesis testing, we evaluated the structural model for potential lateral collinearity, as high correlations among constructs can bias path coefficients. We examined the Variance Inflation Factor (VIF) values for all predictor constructs. As presented in

Table 6. Inner VIF Values.

Construct	VIF
Algorithmic Curation	2.044
Technological Fluidity	1.596
Cognitive Efficiency	2.563
Serendipity	1.580
Service Stickiness	-

, all inner VIF values ranged from 1.383 to 2.563, falling well below the conservative threshold of 3.3 recommended by Hair et al. [22]. This confirms that multicollinearity is not a concern and that each construct contributes distinctly to the structural model.

The structural model was subsequently assessed to test the hypothesized relationships and evaluate explanatory power. The model demonstrated satisfactory predictive relevance, accounting for 48.9% of the variance in Cognitive Efficiency (R² = 0.489), 19.0% in Serendipity (R² = 0.190), and 38.0% in Service Stickiness (R² = 0.380). The results of the hypothesis testing are summarized in

Table 7. Structural Model Results.

Hypothesis	Path			Std. Beta (β)	t-Value	p-Value	Result
H1	Algorithmic Curation	→	Cognitive Efficiency	0.462	5.843	0.000 ***	Supported
H2	Algorithmic Curation	→	Serendipity	0.501	5.034	0.000 ***	Supported
H3	Technological Fluidity	→	Cognitive Efficiency	0.341	4.314	0.000 ***	Supported
H4	Technological Fluidity	→	Serendipity	−0.195	−1.959	0.053	Not Supported
H5	Cognitive Efficiency	→	Service Stickiness	0.598	7.230	0.000 ***	Supported
H6	Serendipity	→	Service Stickiness	0.040	0.477	0.634	Not Supported

Significance levels: *** p < 0.001.

.

First, in examining the impact of perceived AI stimuli on organismic states, Algorithmic Curation (AC) exerted a robust positive influence on both Cognitive Efficiency (CE) (β = 0.462, p < 0.001) and Serendipity (SER) (β = 0.501, p < 0.001), providing strong support for H1 and H2. Similarly, Technological Fluidity (TF) significantly enhanced Cognitive Efficiency (β = 0.341, p < 0.001), supporting H3.

However, contrary to expectation, Technological Fluidity did not show a positive effect on Serendipity (β = −0.195, p = 0.053). Instead, it exhibited a marginally significant negative trend, leading to the rejection of H4. This unexpected finding suggests a potential trade-off: while seamless automation optimizes efficiency, it may inadvertently suppress opportunities for serendipitous discovery. This phenomenon is discussed later as the “Frictionless Trap”.

Second, the analysis of the organism-response relationship yielded the study’s most critical insight regarding retention mechanics. Cognitive Efficiency (CE) emerged as the dominant driver of Service Stickiness (STI) (β = 0.598, p < 0.001), providing robust support for H5. In stark contrast, Serendipity (SER) failed to exert a significant influence on stickiness (β = 0.040, p > 0.05), thereby rejecting H6. This statistical evidence underscores that within the mature AI subscription economy, the primary mechanism facilitating consumer lock-in is not ‘hedonic serendipity’ but ‘utilitarian cognitive efficiency’, which directly mitigates decision fatigue.

5.3. Post Hoc Analysis: Moderating Effects

To further investigate boundary conditions, we conducted a post hoc moderation analysis to assess the influence of service typology and user demographics on the stickiness mechanism. The key statistical results are presented in

Table 8. Post hoc Analysis of Moderating Effects.

Path	Moderator	Interaction Coeff. (β)	p-Value	Result	Interpretation
CE → STI	Service Type (1 = Food)	0.154	0.040 *	Significant	Efficiency matters more in high-frequency tasks (e.g., Food).
SER → STI	Service Type (1 = Food)	−0.042	0.658	Not Sig.	Service type does not moderate the effect of enjoyment.
CE → STI	Income Level	−0.071	0.453	Not Sig.	Efficiency is a universal driver regardless of income.
SER → STI	Income Level	0.237	0.038 *	Significant	High-income consumers are more responsive to serendipity.

Note: * p < 0.05. CE = Cognitive Efficiency, SER = Serendipity, STI = Service Stickiness.

.

First, we examined the moderating role of Service Type (Food & Groceries vs. Non-Food). The analysis identified a significant positive interaction effect between Cognitive Efficiency (CE) and Service Stickiness (STI) (β_interaction = 0.154, p < 0.05). As illustrated in Figure 2, the regression slope for the Food group (Blue line) is considerably steeper than that of the Non-Food group (Orange line). This finding indicates that in high-frequency decision contexts, such as food curation, consumer sensitivity to utilitarian efficiency is substantially heightened during the formation of stickiness.

Subsequently, we analyzed the moderating effect of Income Level. The results revealed a significant interaction between Serendipity (SER) and Service Stickiness (STI) (β_interaction = 0.237, p < 0.05). As depicted in Figure 3, the Low/Mid-Income group (Red line) displays a negligible slope, indicating no significant relationship, whereas the High-Income group (Green line) demonstrates a clear positive trajectory. This pattern suggests a hierarchy of needs within the AI subscription economy: efficiency serves as a universal baseline, while the ‘joy of discovery’ (Serendipity) becomes a critical retention driver primarily for affluent consumers who have likely transcended basic utilitarian constraints.

6. Discussion and Conclusions

6.1. Interpretation of Findings

This study aimed to elucidate the underlying psychological mechanisms of service stickiness in the mature phase of the AI subscription economy. Applying the Stimulus–Organism–Response (S-O-R) framework [11], the research empirically examined the effects of perceived Algorithmic Curation (AC) and Technological Fluidity (TF) on Service Stickiness (STI) through the mediators of Cognitive Efficiency (CE) and Serendipity (SER). Hypothesis testing identified a phenomenon referred to as “The Paradox of AI Subscription,” which presents distinct theoretical implications.

First, the analysis established the ‘Dominance of Cognitive Efficiency.’ Consistent with our expectations, Cognitive Efficiency emerged as the most influential driver of perceived service stickiness (β = 0.598, p < 0.001), providing strong support for H5. Additionally, both Algorithmic Curation (H1) and Technological Fluidity (H3) significantly enhanced this efficiency. These results indicate that, within the mature subscription economy, reducing cognitive cost rather than introducing novel features is the primary factor in preventing consumer churn.

Second, the analysis revealed the ‘Frictionless Trap’ and the limitations of Serendipity. Contrary to prior literature that emphasizes the role of unexpected discovery in fostering customer loyalty, the findings indicate that Serendipity did not significantly influence Service Stickiness in the aggregate model (β = 0.040, p > 0.05), thereby rejecting H6. This null result can be interpreted through the theoretical frameworks of ‘Hedonic Adaptation’ and ‘Cognitive Overload.’ Although serendipity may elicit momentary delight, users in the maturity phase rapidly adapt to these positive stimuli, making the emotional impact short-lived. Moreover, under conditions of cognitive overload associated with subscription fatigue, the utilitarian value of efficiency becomes the dominant factor in the hierarchy of needs, thereby diminishing the influence of hedonic enjoyment.

Crucially, regarding the antecedents of serendipity, while Algorithmic Curation successfully triggered Serendipity (H2), the analysis identified a negative relationship between Technological Fluidity and Serendipity (β = −0.195, p = 0.053), thereby rejecting H4. This finding is interpreted not merely as a non-significant result but as a theoretical insight termed the “Frictionless Trap”. Recent studies indicate that although removing friction reduces effort, it also eliminates the “tangential pathways” where serendipitous discoveries typically occur [30,31]. When an AI system becomes excessively fluid by predicting and automating every step seamlessly, it creates a “Filter Bubble of Efficiency” that insulates the user from the experience of unexpected discovery [32,33].

Third, the analysis identified boundary conditions that suggest a ‘Hierarchy of Digital Needs’. Post hoc analysis revealed that the impact of Cognitive Efficiency on stickiness was substantially stronger in high-frequency utilitarian contexts, such as food subscription services, where decision fatigue is prevalent. In contrast, among the High-Income group, the previously insignificant effect of Serendipity became statistically significant. This finding highlights a boundary condition: efficiency drives stickiness in survival-related or routine contexts, while enjoyment and discovery promote retention among users with greater economic freedom. This pattern supports the existence of a “Hierarchy of Digital Needs” within the AI subscription economy, with utility at the base and serendipity at the apex.

6.2. Theoretical Implications and Propositions

This study contributes to the literature by refining the application of the S-O-R framework in the context of human-AI interaction.

First, addressing the need for conceptual precision regarding the level of analysis, we explicitly defined Algorithmic Curation and Technological Fluidity as perceived attributes rather than objective system features. As noted by recent scholarship, user behavior is driven not by the algorithm’s actual code but by the user’s algorithmic awareness and perception of the system’s agency [34,35]. By aligning the stimuli (perceived attributes), organism (psychological states), and response (stickiness) at the individual level, this study resolves the level-of-analysis misalignment often found in techno-centric studies.

Second, we extend the static interpretation of the S-O-R model. While we employed the framework to model the causal flow from AI stimuli to behavioral stickiness, we acknowledge, in agreement with emerging theoretical perspectives, that human-AI interaction is inherently circular [34]. User responses feed back into the system and generate new data that further refines the Algorithmic Curation, creating a continuous feedback loop. Our findings should therefore be interpreted as a cross-sectional snapshot of this “spiral evolutionary process.” We argue that in the early stages of this spiral, efficiency acts as the gravitational force that locks users into the orbit of the service.

Moving beyond empirical validation to theoretical advancement, we articulate three key propositions that logically emerge from our findings to guide future research on human-AI adaptation:

• Proposition 1 (The Threshold of Novelty): In high-frequency algorithmic environments, the behavioral influence of experiential novelty (serendipity) diminishes as the user’s cognitive fatigue increases. This implies a boundary condition where Cognitive Efficiency serves as a prerequisite ‘hygiene factor’ that must be satisfied before hedonic factors can exert any meaningful impact on retention.
• Proposition 2 (The Fluidity-Serendipity Trade-off): As perceived Technological Fluidity approaches a maximum state of being perfectly frictionless, the user’s perception of Serendipity will follow an inverted U-shape or decline due to the loss of exploratory opportunities. This suggests that adaptive learning requires a certain degree of “meaningful friction” to prevent the formation of a filter bubble that inhibits discovery.
• Proposition 3 (Cognitive Dependence on Automation): The strong association between Algorithmic Curation and Cognitive Efficiency suggests that as users’ trust in algorithmic agency deepens, they progressively offload decision-making authority to the system. This leads to a state of ‘Cognitive Dependence,’ where service stickiness transforms from a voluntary choice into a structural reliance on automated decision aids.

6.3. Managerial Implications

For practitioners in the subscription economy, the findings regarding the dominance of efficiency and the insignificance of serendipity offer clear and actionable strategic directives.

First, marketing narratives must undergo a strategic pivot from “Awe of Discovery” to “Economy of Thought.” Since Cognitive Efficiency was identified as the primary retention driver, the core value proposition should not be positioned around the addition of new experiences, but rather the recovery of time and mental energy. As suggested by our proposition on Cognitive Dependence, consumers do not subscribe to AI services to find “more options,” but to liberate themselves from the burden of choice. Therefore, communication strategies should emphasize a “Zero-Effort Lifestyle” where the system minimizes the user’s cognitive load to near-zero, rather than highlighting the complexity or novelty of the algorithm.

Second, platforms should adopt a “Dual-Track Design Strategy” to overcome the Frictionless Trap. Our discovery of the negative relationship between Technological Fluidity and Serendipity implies that excessive automation paradoxically kills the joy of discovery. To address this, managers should distinguish between “Efficiency Mode” (default) and “Discovery Mode” (optional). While the mass market demands seamless efficiency, high-income users or specific segments (as shown in our post hoc analysis) still respond to serendipity. For these premium tiers, services should intentionally engineer “meaningful friction” [36]—such as asking for specific feedback or offering a curated “Choice of the Month”—to invite users to pause and explore, thereby restoring the sense of agency and delight that is often lost in fully automated loops.

Third, context-aware personalization is required based on the service domain. As indicated by our contextual moderation analysis, in utilitarian high-frequency domains like food subscriptions, the strategy should focus ruthlessly on efficiency and predictability. Any attempt to introduce “surprising recommendations” in these sectors may be perceived as unwanted noise rather than serendipity. Conversely, for content or hobby-related services targeting higher economic demographics, algorithms should be tuned to occasionally disrupt the feedback loop with novel suggestions to prevent the service experience from becoming purely mechanical and devoid of emotional engagement.

6.4. Limitations and Future Research

Several limitations identified in this study offer promising directions for future research. First, in terms of research design, the current model provides a cross-sectional view of a fundamentally dynamic interaction. Although the causal flow from stimuli to stickiness was validated, human-AI interaction is inherently recursive [34]. Future research should use longitudinal designs to examine how the balance between efficiency and serendipity evolves as the interaction spiral unfolds.

Second, the sample profile limits the generalizability of the findings. Data were collected solely from U.S. consumers, which may restrict applicability to other cultural contexts with different levels of algorithmic trust and automation tolerance [37]. Future studies should conduct cross-cultural comparisons to evaluate the global relevance of the Frictionless Paradox. The sample also exhibited a gender imbalance (Male: 61.9%). Although this distribution partially reflects the demographics of early adopters in technology-focused sectors, a post hoc independent-samples t-test indicated no significant differences (p > 0.05) in perceptions of key constructs between gender groups. This suggests that the structural relationships remain robust despite the imbalance. Nonetheless, future research should pursue strictly balanced quota sampling.

Third, the operationalization of Service Stickiness requires further development. In this study, stickiness was measured as a perceived psychological state using self-reported data rather than behavioral metrics such as actual renewal rates. Although perceived stickiness is a strong predictor of future behavior [38], discrepancies may arise between intention and action. Future research should incorporate objective log data to triangulate the relationship between psychological stickiness and actual churn.

Finally, future research should investigate the negative aspects of retention. While this study examined positive stickiness, subsequent studies should differentiate between voluntary loyalty and calculative lock-in resulting from Cognitive Dependence (Proposition 3). Examining the impact of AI-induced lock-in on long-term consumer well-being remains an essential research priority.

In conclusion, this research highlights a fundamental paradigm shift: the core of the mature AI subscription economy lies in optimizing the cognitive economy rather than increasing content consumption. The findings indicate that Algorithmic Curation and Technological Fluidity primarily generate value by reducing consumers’ cognitive costs. In an era characterized by subscription fatigue, Cognitive Efficiency emerges as the primary factor sustaining services, demonstrating that sustainable retention depends on a service’s ability to alleviate users’ decision-making burden.