Consumer Valuation of AR-Enabled Indoor Navigation in Shopping Malls: A Choice-Based Conjoint Analysis

Abstract

Augmented reality (AR) indoor navigation can combine wayfinding, information, gamification, and promotions in shopping malls, but how consumers value these features is unclear. In a choice-based conjoint experiment, 295 South Korean adults aged 20–49 from an online panel chose among hypothetical service profiles varying in Navigation Mode, Information Layering, Interactive Character Feature, Reward Type, Reward Acquisition Mode, and monthly fee; a conditional logit model estimated utilities, relative importance, and marginal willingness to pay (MWTP). Within the levels and ranges tested, price showed the highest range-based relative importance (52.2%), followed by Navigation Mode (14.1%) and Information Layering (12.7%). Within the experimental choice sets, conditional MWTP relative to the reference levels was KRW 540 per month (95% CI: 377–702) for Map + AR Guidance and KRW 488 (95% CI: 326–649) for Route + Facility Information + Discount Information. Instant Discount Coupons were preferred to Reward Points and Prize Draw Entry, and Visit-Based acquisition was preferred to Social-Interaction-Based acquisition. The Interactive Character Feature had no significant average effect, but its utility was higher among respondents with greater context-specific gamification-feature affinity. These stated-preference findings suggest a layered design for the proposed service, with guidance and contextual information as the functional core and gamified features as optional.

1. Introduction

The growth of online commerce is fundamentally changing the role of shopping malls. Whereas shopping malls once served as places to search for and purchase products, consumers can now complete much of the shopping process—from product search and price comparison to purchase and delivery—online. Shopping malls can therefore no longer differentiate themselves from online channels merely by selling products and increasingly need to provide experiential value that makes an in-person visit worthwhile. In response, malls are being reconfigured from places of purchase into complex experiential spaces combining leisure, dining, culture, interactive experiences, and social interaction [1,2,3]. The competitiveness of physical retail is thus no longer determined solely by what it sells. How consumers move through the space, what they discover, what information they encounter, and what experiences they accumulate have become important components of customer value.

Paradoxically, however, the shift toward experience-oriented retail creates a new challenge. As malls integrate more stores, amenities, events, promotions, and experiential content into one space, the amount of spatial information and the number of choices users must process increase. In particular, multilevel structures, repetitive circulation patterns, similar store layouts, movement between buildings, multiple entrances, and connecting passageways in large, mixed-use shopping malls can impose a substantial cognitive burden as users determine their current location and the route to a destination [4,5]. Wayfinding is therefore not a secondary activity that merely supports movement but a central process shaping the overall shopping experience. If consumers cannot easily find a desired store or facility or experience confusion and fatigue while moving, the rich experience intended by the mall may instead become a source of inconvenience and stress [4]. A key challenge for experience-oriented physical retail is therefore not only to provide more content but also to help users explore complex spaces efficiently and access the information they need when they need it.

In this context, AR-enabled indoor navigation has attracted attention as a digital interface that can reconfigure the spatial experience of shopping malls. Conventional 2D maps and signs require users to interpret map information and mentally map it onto the physical environment [6]. By contrast, AR navigation overlays route, direction, and destination information in real time onto the physical environment captured by a mobile-device camera, helping users move while maintaining awareness of their surroundings [7]. Prior research has reported that AR guidance can reduce cognitive workload relative to map-based guidance [6] and improve spatial cognition and wayfinding efficiency by providing a high level of situational awareness [8]. These effects, however, are not always consistent. Limited user experience, interface quality, and positioning accuracy can impair navigation performance and usability [9], while AR guidance may reduce navigation demands at the expense of route memory [6,10]. Thus, the value of AR navigation depends not only on the technology itself but also on how it is configured within a service.

In shopping malls, this configuration can extend well beyond route guidance. An AR interface can combine navigation with facility and store information, promotions, characters, and rewards, functioning as a digital touchpoint through which consumers encounter both spatial and commercial information while moving. Yet it remains unclear how consumers value these features relative to one another, particularly when they must trade them off against price. Research on smart retail and immersive technologies has emphasized the potential of AR, gamification, and other experiential elements [11,12], but their availability does not necessarily imply consumer value. Practical features that directly support the shopping task may be valued differently from features that require additional participation.

Existing research provides limited evidence on these trade-offs. AR-wayfinding studies have primarily examined navigation outcomes: Dong et al. [6] compared AR with a 2D electronic map in terms of wayfinding performance, visual attention, and route memory; Xu et al. [8] compared exocentric and AR-enabled egocentric cues in terms of wayfinding efficiency, cognitive load, and spatial awareness; and Qiu et al.’s [7] systematic review of 88 studies similarly organizes the field around navigation technology, user experience, wayfinding performance, and cognition. Choice experiments on mobile apps, such as that of Folkvord et al. [13], have estimated preferences for app features and prices, but not for a service used while moving through a store or mall; retail and e-commerce research has examined customer experience, omnichannel retailing, promotions, and gamification [2,14,15,16]. Taken together, these studies do not establish how consumers trade off AR guidance against contextual information, gamified features, promotional rewards, and price when these attributes are jointly configured within a navigation service embedded in a physical retail environment. Johnson and Barlow [14], for example, conceptualize phygital experiences in terms of the benefits of product experience and the costs of paying, but do not examine the relative value of specific digital features at a phygital touchpoint.

Accordingly, this study conceptualizes AR-enabled indoor navigation as a smart retail service combining functional, experiential, promotional, and economic attributes. The functional attributes comprise Navigation Mode and Information Layering; the experiential attributes comprise an Interactive Character Feature and Reward Acquisition Mode; the promotional attribute is Reward Type; and the economic attribute is monthly price. Using choice-based conjoint (CBC) analysis, this study estimates the relative utility of each attribute level and calculates marginal willingness to pay (MWTP) from the price coefficient. It also evaluates preference heterogeneity associated with context-specific gamification-feature affinity and shopping mall visit frequency. Directional hypotheses for the six attributes (H1–H6) are developed in Section 2.2, Section 2.3, Section 2.4, Section 2.5 and Section 2.6, and two research questions are addressed.

RQ1. Within the tested price range and choice sets, what is consumers’ conditional marginal willingness to pay for each non-price attribute level?

RQ2. How do preferences for each attribute vary with context-specific gamification-feature affinity and shopping mall visit frequency? No directional predictions are made, because prior research provides insufficient basis for consistent directional predictions, and the affinity measure overlaps in content with several attributes.

This study makes three contributions. First, it extends AR-wayfinding research from the performance consequences of AR guidance to its relative consumer value as one component of a configurable retail service. Second, it adds a task-based perspective to phygital retail research [14,17] by interpreting attribute-level preference estimates in relation to the consumer’s focal task. Third, for a service category without an established market benchmark, it provides attribute-level evidence on how consumers trade off functional, experiential, promotional, and economic attributes within a single phygital service configuration, including their conditional MWTP and preference heterogeneity.

2. Literature Review

2.1. AR-Enabled Smart Retail Services as a Phygital Commerce Interface

Within physical retail, AR, mobile applications, location-based services, and promotional systems serve as smart retail technologies that integrate digital information into physical spaces. Smart retail technologies can enhance customer experience by providing more efficient and personalized services [18]. Mobile AR, in particular, represents a key smart retail technology that enables users to interact with digital information within physical stores. By integrating virtual information with the physical environment, mobile AR can create additional value for both consumers and retailers by supporting product and store information search, decision making, and interactions with brands [11,12].

The term phygital, a combination of physical and digital, refers to experiences in which physical and digital elements are integrated within the same physical setting and consumption process. Ref. [17] conceptualized phygital experiences along spatial, temporal, hedonic–sensory, social, and participatory dimensions. This multidimensional perspective suggests that in-store digital technologies can shape not only efficiency and access to information but also users’ spatial movement, sensory experiences, enjoyment, social interaction, and participation. Ref. [14] similarly described phygital marketing as the integration of physical and digital experiences within a single customer journey, with AR, digital payments, and interactive screens as examples of its implementation. Applied to retail environments, this concept is referred to as phygital retail. The key issue in phygital retail is not the presence of individual technologies themselves, but how physical and digital elements are integrated throughout the customer journey. From this perspective, AR-enabled indoor navigation can be viewed as one such touchpoint where physical and digital elements converge.

AR-enabled indoor navigation can therefore be viewed not simply as a technology that provides directions, but as a phygital commerce interface connecting shopping mall environments with digital commerce experiences [14,17]. The AR interface can layer facility locations, discount information, characters, rewards, and other digital elements onto route information, thereby linking users’ spatial movement with consumption decisions. The service attributes specified in this study are conceptually related to multiple dimensions of the phygital experience [17]. Navigation Mode and Information Layering relate to the spatial and temporal dimensions, while AR guidance that overlays a digital route onto the user’s view of the physical environment extends the sensory dimension. Characters and missions relate to hedonic and participatory value during wayfinding, whereas Social-Interaction-Based acquisition relates to the social dimension by enabling interaction with other users.

The consumer value of AR-enabled indoor navigation may vary across its constituent attributes and should therefore be examined at the attribute level. Consumer responses to phygital experiences may vary depending on which physical and digital elements are combined and how they are integrated, while the benefits and costs of a particular technology may also differ across use contexts [14]. AR-enabled indoor navigation may provide functional value through improved wayfinding efficiency and convenient access to information, but excessive information and gamified elements may also cause distraction or impose additional behavioral demands [19]. Understanding the consumer value of an AR-enabled smart retail service therefore requires examining its individual attributes—Navigation Mode, Information Layering, Interactive Character Feature, Reward Acquisition Mode, Reward Type, and Price—rather than evaluating the service as a single technology. The following sections examine the functional, experiential, promotional, and economic value associated with these attributes.

2.2. Functional Value of AR-Enabled Indoor Navigation

2.2.1. Indoor Wayfinding and Cognitive Burden in Complex Retail Spaces

Wayfinding is defined as a process in which people perceive spatial information, determine where to go by selecting a direction and route, and translate that decision into physical movement [20]. It involves both cognitive functions for perceiving and evaluating environmental information and behavioral functions for translating those judgments into actual movement [4]. While moving, users must continually check their location, choose an appropriate route from among alternatives, and repeatedly verify that the selected route leads to the destination.

The cognitive demands of wayfinding are greater in spatially complex settings such as large, mixed-use shopping malls. Their multilevel structures, repetitive circulation patterns, similar store layouts, movement between buildings, numerous entrances, and connecting passageways make it difficult for users to intuitively determine their current location and direction of travel [5]. As the number of available routes and visual cues increases indoors, users must search for relevant information while simultaneously determining their current location, orienting themselves, and selecting an appropriate route. These concurrent demands can place considerable strain on limited working memory and lead to cognitive overload [21,22], potentially resulting in errors in directional judgment or confusion when selecting a route [23].

Prior research reports that wayfinding difficulties in complex indoor spaces can cause frustration, anxiety, and stress and can lead to reduced efficiency and negative experiences in commercial settings [4]. Such difficulties can hinder users from reaching stores or facilities efficiently and undermine the overall shopping mall experience. Because functional convenience, diverse facilities and content, and entertainment experiences all shape the shopping mall experience [24], helping users move smoothly through the space constitutes an important form of functional value. A Navigation Mode that supports spatial information processing and reduces wayfinding burden is therefore a core function of AR-enabled indoor navigation in shopping malls.

2.2.2. Navigation Mode: Map Only Versus Map + AR Guidance

In this study, Navigation Mode refers to how wayfinding information is presented. In the reference condition, route and directional information is displayed on a two-dimensional map; in the AR condition, the same information is presented both on the 2D map and as an overlay on the physical environment. A 2D map requires users to mentally transform map information to align it with the physical environment [6]. This mental transformation requires additional cognitive processing and can impose greater cognitive burden in complex indoor environments. By contrast, AR guidance presents route and directional information from an egocentric perspective by overlaying it directly onto the physical environment, thereby reducing the need for such mental transformation and supporting spatial understanding [8].

The effects of AR navigation have also been examined in prior research. The review in [7] found that AR navigation was generally associated with lower mental workload than traditional wayfinding tools. In the eye-tracking experiment reported in [6], AR users showed shorter fixation durations and smaller mean pupil sizes than 2D map users, indicating relatively lower cognitive load during wayfinding. Ref. [8] found that AR cues could reduce task completion time and support spatial cognition in a multilevel building environment. Taken together, these findings suggest that AR guidance may reduce cognitive load and support faster and more efficient wayfinding than 2D maps in complex indoor environments.

The effectiveness of AR may nevertheless vary depending on usability, visual obstruction, hardware constraints, and how information is presented. Although AR users in [6] showed lower average cognitive load, the study found no significant difference in actual wayfinding performance between the AR and 2D map conditions. Ref. [10] found that AR-based indoor navigation performed better than a paper map in terms of travel time and subjective workload, whereas the paper map performed better in remembering and retaining the route. A GPS-based pedestrian navigation study [25] likewise found that a 2D map performed better on pragmatic outcomes such as task completion time, while AR was rated more highly on hedonic aspects such as immersion and enjoyment through the integration of physical and digital information. Thus, the effects of AR guidance have not been consistent across prior studies. These inconsistencies, however, concern mainly route retention and task-performance outcomes, whereas the more consistent advantage of AR guidance—reduced map-to-environment transformation and lower cognitive load [6,7,8]—bears directly on the focal task of reaching a destination in a complex mall. Consumers evaluating service profiles are therefore expected to assign greater utility to AR guidance.

H1. 

Map + AR Guidance will yield higher choice utility than Map Only.

2.3. Information Layering in AR-Enabled Indoor Navigation

2.3.1. Contextual Information Value in AR Interfaces

An AR interface can overlay information relevant to the user’s current location and movement context onto route guidance in the physical environment. In a shopping mall, users can access information about facilities, nearby stores, and discounts while moving toward a destination. Such contextual information may enrich spatial exploration beyond the functional purpose of wayfinding and influence consumer decision making. Accordingly, Information Layering in this study refers to the progressive addition of contextual information to basic route guidance. The additional information is classified into two types based on its function.

The first type is spatial-functional information, which includes information about elevators, escalators, nursing rooms, and other facilities relevant to movement and facility use. Such information may facilitate route selection and movement efficiency, thereby supporting users’ navigation of the space. Ref. [5] found that stores located in the central area of a shopping mall were visually salient, whereas elevators and some stores along secondary corridors were less visible and therefore more difficult to locate. This finding suggests that route guidance in complex shopping malls should be accompanied by location information for major facilities.

The second type is commercial contextual information, referring to information such as discounts and promotions that is not directly related to wayfinding but can enrich the shopping experience. Mobile location-based retail apps allow users to explore product, discount, coupon, and promotional information and discover new shopping opportunities while moving [26]. Ref. [26] found that perceived interactivity and compatibility were associated with greater affective involvement with mobile LBS retail apps, which in turn was associated with stronger intentions to download and use the apps. Accordingly, integrating location-relevant store and promotional information into AR navigation may create additional value by turning movement through the mall into an opportunity to discover new shopping opportunities.

2.3.2. Information Value Versus Information Overload

More information may increase convenience, but when the amount of information exceeds users’ processing capacity, it can cause distraction and information overload [19]. Information overload refers to a state in which users’ ability to identify relevant information and make decisions is impaired. Commercial information may have both benefits and drawbacks because it adds information that is not directly related to wayfinding and may therefore distract users’ attention and interfere with the processing of essential route information.

Because digital objects are overlaid onto the user’s view of the physical environment in AR, greater amounts of information may interfere with the processing of essential route information. Ref. [7] noted that although AR navigation can generally reduce cognitive load, excessive information or inappropriate interface design may instead impose additional cognitive load. Ref. [27] found that high information density in an AR environment significantly increased the visual search time for text and symbols. Ref. [28] also identified cognitive and motor performance costs when participants simultaneously performed a movement task and an AR visual-discrimination task. These findings suggest that processing additional AR information while moving may divide users’ limited attentional resources.

Information Layering thus entails a trade-off between greater convenience and experiential value on the one hand and increased cognitive burden on the other; providing information beyond an appropriate level may impair decision quality. Within the levels examined in this study, the functional and contextual value of facility and discount information is expected to outweigh the potential burden of processing the additional information. Because the levels are cumulative, H2b compares the full information layer with Route Only and does not test the independent value of discount information.

H2a. 

Route + Facility Information will yield higher choice utility than Route Only.

H2b. 

Route + Facility Information + Discount Information will yield higher choice utility than Route Only.

2.4. Experiential and Gamified Value in AR-Enabled Indoor Navigation

2.4.1. Gamification in Utilitarian Service Contexts

Gamification refers to the application of game-design elements in nongame contexts [29]. It has been widely used in retail and e-commerce to enhance consumer engagement and purchase intention by leveraging affordances such as rewardability, connectiveness, playfulness, and novelty to foster immersive experiences and encourage user behavior [30]. However, the effects of gamification may vary across use contexts and user characteristics [16].

The primary task of AR-enabled indoor navigation is to help users reach their destination efficiently. Gamified elements may add value when they enrich the experience of moving through the space. However, if they function as secondary tasks that must be processed alongside wayfinding, dual-task demands may impose cognitive and behavioral performance costs [28] and thereby reduce service value. Ref. [30] showed that the relationship between gamification affordances and immersive experience varied across user groups in an online shopping context. This suggests that the effects of gamification may not be fully captured by overall average patterns and should also be examined across user groups. The review in [16] similarly noted that the effects of gamification vary by context and user characteristics. Accordingly, the following sections state directional expectations for the average utility of gamified elements in the utilitarian and commercial context of shopping mall wayfinding, while potential variation in these preferences across user groups is examined separately through RQ2.

2.4.2. Interactive Character Feature as an Anthropomorphic Social Cue

The Interactive Character Feature can be regarded as an anthropomorphic cue within a service interface [31], potentially eliciting emotional responses by simulating social interaction with the user. In the experimental condition, a virtual character accompanied users during navigation and could be fed, played with, and leveled up. The attribute therefore bundled visual presence with interaction and progression. When users perceive such a character as an interaction partner, they may experience social presence [32]. This experience is related to the need for relatedness, one of the basic psychological needs identified by self-determination theory [33]. The Interactive Character Feature may therefore introduce social, emotional, and progression-related qualities into the wayfinding experience. These qualities are expected to increase consumers’ choice utility.

H3. 

The presence of the Interactive Character Feature will yield higher choice utility than its absence.

However, the effect of Interactive Character Feature may vary across users with different characteristics. In a context where the utilitarian task of wayfinding takes priority, users with lower gamification-feature affinity may perceive a character as a visual element that interferes with route information processing. Ref. [30] found that the relationship between gamification affordances and immersive experience varied in strength across user groups in an online shopping context. Although that study did not directly examine characters, its findings suggest that responses to gamified elements may vary across users with different characteristics. Such variation is examined separately through RQ2.

2.4.3. Reward Acquisition Mode: Visit-Based, Mission-Based, and Social-Interaction-Based

Reward Acquisition Mode refers to the behaviors and level of participation required to obtain rewards within the service. Visit-Based requires reaching a destination, Mission-Based requires completing an individual task, and Social-Interaction-Based requires interacting with other users, thereby imposing different behavioral and social demands. From the perspective of self-determination theory, task completion and social interaction may support experiences of competence and relatedness, respectively, although their effects may depend on the demands and burdens associated with these activities [33,34].

Under Visit-Based acquisition, a reward is provided automatically when the user arrives at a designated store, requiring no additional behavior beyond wayfinding. Mission-Based acquisition may foster a sense of competence and achievement through completion of an app-based task [34], but it may also increase task burden by requiring additional actions beyond basic wayfinding. Social-Interaction-Based acquisition may support relatedness and social interaction through collaboration with other users but also entails social exposure. Depending on the use context, these demands may contribute to social overload and fatigue [35]. Social-Interaction-Based acquisition adds collaborative action, social exposure, and dependence on other users, none of which is required to reach the destination; it is therefore expected to yield lower utility than Visit-Based acquisition. Mission-Based acquisition, by contrast, adds a brief, individually controlled task whose completion provides immediate performance feedback, which has been associated with greater perceived competence [34]; it is therefore expected to add a sense of achievement and to yield higher utility than Visit-Based acquisition.

H4a. 

Mission-Based acquisition will yield higher choice utility than Visit-Based acquisition.

H4b. 

Social-Interaction-Based acquisition will yield lower choice utility than Visit-Based acquisition.

Self-determination theory is used here to derive the direction of H3 and H4a, not as a mechanism that this study tests. The design manipulated service features rather than need satisfaction: competence and relatedness were neither isolated nor measured, and each participation feature bundled several components. The results can therefore show whether these features increased choice utility, but not whether any effect, or its absence, arose through the satisfaction of a particular need.

2.5. Promotional Value and Reward Type in AR-Enabled Shopping Services

Immediate, Delayed, and Uncertain Rewards

Rewards provided through AR-enabled indoor navigation are not merely gamification devices; they can also function as promotional tools that connect service use with store visits, purchases, and repeat use. While moving toward a destination, users can receive digital benefits such as discount coupons, reward points, or prize draw opportunities through the AR interface. Shopping malls can thereby integrate commercial activities into the process of spatial exploration. This integration reflects a key characteristic of phygital marketing, in which physical and digital experiences are connected within a single customer journey [14]. Digital technologies such as mobile apps and AR also provide users with convenience and novel experiences while serving as touchpoints through which firms can understand user behavior and preferences and deliver promotions [15].

The mere inclusion of a reward does not necessarily generate the same level of promotional value. The effects of rewards may vary not only with their absolute monetary value but also with the timing of benefit realization and the certainty of receiving the actual reward [36]. This study specified three reward forms commonly used in shopping contexts—Instant Discount Coupon, Reward Points, and Prize Draw Entry—as the levels of Reward Type.

An Instant Discount Coupon is a direct and certain reward whose monetary benefit can be recognized and used immediately after service use. Users can realize its economic value without an additional accumulation process or waiting for future use. In low-involvement consumption contexts, immediate rewards have been found to increase the perceived value of loyalty programs more effectively than delayed rewards [37]. This suggests that immediately usable benefits may be more readily evaluated by consumers. In the context of indoor navigation, where fast and efficient movement takes priority, an Instant Discount Coupon may likewise be perceived as a comparatively clear and readily evaluated benefit.

Reward Points are a delayed reward: benefits accumulate through current service use or purchases, but their economic value is realized only after certain conditions are met at a later point. Rather than providing an immediate price reduction, Reward Points are used in loyalty programs to encourage repeat visits and purchases [38]. Perceived reward value may vary depending on the point-accumulation structure, and higher perceived reward value may be associated with greater spending intention [39]. Frequent shopping mall visitors may value the accumulated value of points more highly because they have more opportunities to use them in the future. In contrast, users who visit less frequently may perceive Reward Points as less attractive than an immediate discount coupon. The utility of Reward Points may therefore be related to users’ expectations regarding future opportunities to use and accumulate points.

Prize Draw Entry provides an opportunity to receive a reward, but neither the receipt nor magnitude of the benefit is certain. Unlike a guaranteed discount coupon or reward points, a prize draw may yield no material benefit. Ref. [36] noted that consumers generally value certain gains more highly than uncertain gains. More recent research, however, has reported that optimistic expectations of the best possible outcome can increase participation in uncertain promotions, particularly in intuitive, low-involvement situations in which consumers do not closely evaluate reward probabilities [40]. Prize Draw Entry may also function as a gamification affordance through playfulness [30]. In a context where efficient wayfinding takes priority, however, the lack of a guaranteed benefit may contribute to dispreference. Although the effect of an uncertain reward can depend on its potential value and on how deliberately consumers evaluate reward probabilities [40], choosing a navigation service is a task-oriented decision in which a guaranteed benefit is likely to be more readily evaluated than a probabilistic one [36].

In summary, an Instant Discount Coupon provides direct economic value based on immediacy and certainty, whereas Reward Points provide cumulative value through future benefits and repeated use. Prize Draw Entry, in contrast, does not guarantee an outcome but may generate anticipation and encourage participation through uncertainty. Because the three reward types differ in the timing and certainty of benefits and in the degree of temporal involvement required from users, they may be associated with different levels of choice utility within the same AR-enabled shopping service. Because the three formats also differ in accumulation conditions, redemption effort, and potentially expected monetary value, any utility difference reflects the combined evaluation of these characteristics. Nevertheless, the immediate and certain benefit of an Instant Discount Coupon is expected to yield greater choice utility than the delayed benefit of Reward Points or the uncertain benefit of a Prize Draw Entry.

H5a. 

Reward Points will yield lower choice utility than Instant Discount Coupons.

H5b. 

Prize Draw Entry will yield lower choice utility than Instant Discount Coupons.

2.6. Price Sensitivity and Marginal Willingness to Pay for AR-Enabled Indoor Navigation Services

Phygital retail services create new benefits and experiences by integrating digital functions into physical shopping environments. Consumer choice in phygital settings, however, may depend not only on these benefits but also on the costs and burdens associated with payment. In particular, phygital consumption can be understood in terms of the relationship between gains derived from the product experience and costs associated with the payment experience; as payment costs increase, the psychological pain of paying may also increase, potentially reducing purchase likelihood [14]. Assessing the economic value of phygital retail services therefore requires moving beyond consumers’ attitudes and usage intentions to examine how much they are willing to pay for individual service attributes. The market acceptance of AR-enabled indoor navigation may similarly depend not only on its technological appeal and service benefits but also on the price users must pay.

Consumers’ economic valuation can be assessed using marginal willingness to pay (MWTP). Willingness to pay (WTP) refers to the maximum amount a consumer is willing to pay for an entire alternative, whereas MWTP refers to the additional amount a consumer is willing to pay to obtain a particular attribute level relative to a reference level. Analyzing consumer choices among alternatives that include both price and non-price attributes allows the utility of each attribute level to be estimated and the value of non-price attributes to be expressed in monetary terms based on their relationship with the price coefficient [41]. Because conjoint analysis derives MWTP by relating attribute-level utility coefficients to the price coefficient, the relative economic value of different service attributes can be compared on a common monetary scale. Because the value of a new product or service may not be fully captured by usage intention alone, this approach provides a more concrete assessment of the attribute-level value of a phygital retail service [41].

Accordingly, this study specified three monthly price levels—KRW 990, KRW 1990, and KRW 2990—and used the resulting price coefficient to estimate MWTP for each level of Navigation Mode, Information Layering, Interactive Character Feature, Reward Type, and Reward Acquisition Mode. Given the forced-choice design without an opt-out, free-service, or existing-navigation alternative, these MWTP estimates should be interpreted as conditional, relative monetary valuations within the experimental choice sets rather than as direct estimates of consumers’ actual willingness to subscribe or the prices they would pay in the marketplace. Under this interpretation the estimates enable comparison of the relative economic value of the functional, experiential, and promotional attributes of AR-enabled indoor navigation and can inform feature prioritization and subsequent pricing research. Consistent with the payment-cost perspective described above [14], higher monthly fees are expected to reduce consumers’ choice utility.

H6. 

A higher monthly fee will yield lower choice utility.

2.7. User Heterogeneity in Preferences for AR-Enabled Indoor Navigation Services

AR-enabled indoor navigation can combine functional elements, such as wayfinding support and information provision, with gamified elements, such as characters, missions, and rewards. However, users may not assign the same value to these elements. Evaluations of functional attributes may vary with users’ experience of shopping mall environments, while evaluations of gamified attributes may vary with individual gamification-feature affinity. Average preferences observed in the full sample may therefore not fully capture preference patterns associated with different user characteristics. To explore potential preference heterogeneity, this study focuses on gamification-feature affinity and shopping mall visit frequency. Because prior evidence on the direction of these relationships is limited and several affinity items overlap in content with the conjoint attributes (Section 3.3), both moderators are examined through RQ2 without directional hypotheses.

2.7.1. Context-Specific Gamification-Feature Affinity

Responses to gamified elements may vary with users’ motivations and affinity for particular game-like features. Ref. [42] distinguished utilitarian, hedonic, and social motivations for using gamified services, while Ref. [30] found that the affordances associated with immersive online-shopping experiences varied across latent user segments. These findings suggest that the same gamified elements may be evaluated differently depending on users’ motivations and dispositions. In this study, the moderator is described as context-specific gamification-feature affinity because the researcher-developed items directly referenced game elements relevant to the service concept, including points, missions, collaborative activity, and character nurturing.

From a self-determination theory perspective, game elements may support experiences of competence and relatedness and thereby foster intrinsic motivation [33,34]. However, the effects of gamification may vary depending on the service context and users’ motivations [16]. In a functionally oriented service such as wayfinding, characters and missions may provide enjoyment and participatory experiences for some users but may be perceived by others as additional elements that are not directly related to the primary task. Therefore, in contexts such as AR-enabled indoor navigation, where functional tasks and gamified elements are combined, potential preference heterogeneity associated with gamification-feature affinity warrants examination. Accordingly, whereas H3 and H4a concern the average utility of gamified features, RQ2 examines whether these utilities vary in magnitude or direction with context-specific gamification-feature affinity.

2.7.2. Shopping Mall Visit Frequency

Shopping mall visit frequency is a behavioral characteristic that reflects users’ repeated experience with shopping mall environments. Repeated visits may increase familiarity with spatial structure and facility layouts. In wayfinding research, spatial familiarity has been shown to influence cognitive burden and information-search strategies [5,20]. Ref. [43] found that shopping mall users with greater spatial familiarity tended to rely on internal spatial information stored in memory, whereas less familiar users were more likely to rely on external information sources such as signs and maps. Accordingly, users with higher shopping mall visit frequency may assign relatively less value to the additional support provided by AR guidance and Information Layering because they can draw on accumulated spatial knowledge. Conversely, users with lower visit frequency may place greater value on external guidance and spatial information.

Visit frequency may also be associated with evaluations of gamified elements such as missions and rewards. For frequent visitors who are familiar with the space, Mission-Based rewards may be perceived as an additional procedure or burden during repeated use, whereas less-frequent visitors may experience the same mission as an opportunity to explore the space in a new way. Frequent visitors may also have more opportunities to redeem accumulated points in the future, whereas less-frequent visitors may place greater value on immediately usable rewards. However, because prior research has rarely examined these relationships directly, this study explores preference patterns associated with shopping mall visit frequency without presupposing a particular direction.

2.8. Research Framework and Attribute Specification

Building on the theoretical discussion above, this study identified the key attributes of an AR-enabled indoor navigation service and established the theoretical rationale for each attribute level. AR-enabled indoor navigation is conceptualized not merely as a wayfinding technology, but as a multifaceted service that integrates functional wayfinding value, contextual information, experiential and social cues, gamified forms of engagement, promotional value, and economic cost. Table 1 summarizes the six CBC attributes examined in this study, their levels, and the theoretical rationale underlying each.

Table 1. Descriptive Classification and Specification of the CBC Attributes.

Analytical Value Category	CBC Attribute	Attribute Levels	Theoretical Rationale
Functional value (wayfinding support)	Navigation Mode	Map Only; Map + AR Guidance	AR overlay may reduce map-to-environment mental transformation, lowering wayfinding cognitive load.
Functional value (contextual information)	Information Layering	Route Only; Route + Facility Information; Route + Facility Information + Discount Information	Layered information may enhance exploratory and shopping value while increasing information -processing demands
Experiential value (social cue)	Interactive Character Feature	Interactive Character Absent; Interactive Character Present	Anthropomorphic cue may enhance social presence and relatedness, but may also distract from the utilitarian task
Promotional value	Reward Type	Instant Discount Coupon; Reward Points; Prize Draw Entry	Reward utility may vary with timing and certainty of benefit realization.
Experiential value (gamified participation)	Reward Acquisition Mode	Visit-Based; Mission-Based; Social-Interaction-Based	Reward acquisition may foster achievement and relatedness while imposing behavioral or social demands
Economic value	Price	KRW 990 per month; KRW 1990 per month; KRW 2990 per month	Service choice reflects trade-offs between attribute benefits and monetary cost.

Note. The value categories organize the theoretical roles of the attributes; the CBC model estimates the six attributes separately.

Table 1 groups the attributes according to the primary type of value they are expected to provide: functional, experiential, promotional, and economic. These categories are used as analytical classifications rather than as distinct theoretical constructs, because individual attributes may provide more than one type of value; discount information, for example, is both contextual and promotional. Together, however, they represent different sources of benefit and sacrifice that consumers trade off when choosing among configurations of a phygital navigation service. Functional value concerns support for accomplishing the focal navigation task; experiential value concerns engagement beyond task completion; promotional value concerns incentive-based benefits associated with the shopping visit; and economic value reflects the monetary sacrifice required to obtain these benefits. Consumer choice therefore reflects not the value of any single category in isolation but the trade-offs among these sources of value across alternative service configurations.

A task-based perspective further explains why these forms of value may contribute differently to choice. In AR-enabled indoor navigation, the focal task is wayfinding (Section 2.4.1). Features primarily associated with functional value, such as Navigation Mode and facility information, directly support this task and are therefore closely aligned with its requirements, consistent with the view that a technology is more likely to be used and to improve performance when its capabilities fit the task it supports [44]. Contextual discount information is not required for wayfinding but provides additional information without requiring users to perform a separate activity. By contrast, features that primarily provide experiential value, including the Interactive Character Feature and Mission-Based and Social-Interaction-Based reward acquisition, require attention or action beyond wayfinding. These features may offer experiences of competence and relatedness [33,34], but they may also impose additional demands alongside the focal navigation task [28]. Their value may therefore be more contingent on users’ shopping goals and preferences [45,46]. Promotional value through Reward Type changes the incentive users receive for using the service rather than how they perform the focal task, whereas economic value, represented by the monthly fee, constitutes the monetary sacrifice weighed against these potential benefits. This task-based perspective provides an organizing logic for the attribute-specific hypotheses developed in Section 2.2, Section 2.3, Section 2.4, Section 2.5 and Section 2.6.

3. Materials and Methods

3.1. Conjoint Experimental Design

This study employed choice-based conjoint (CBC) analysis to estimate users’ preferences for six attributes of an AR-enabled indoor navigation service and the marginal willingness to pay (MWTP) associated with changes in attribute levels: Navigation Mode, Information Layering, Interactive Character Feature, Reward Type, Reward Acquisition Mode, and Price. Conjoint analysis is a decompositional method for estimating preferences for multi-attribute products or services by having respondents evaluate combinations of attribute levels [47]. CBC presents respondents with sets of service alternatives and asks them to select one, thereby simulating a structured choice context and enabling the estimation of trade-offs among attributes [48]. Because the gamified AR-enabled indoor navigation service examined in this study was hypothetical and not yet available in the market, actual usage data and A/B testing were unavailable. CBC therefore made it possible to test alternative service profiles experimentally and collect stated-choice data relevant to new service design.

3.1.1. Attribute Levels

The six attributes and their levels (Table 2) were developed through iterative discussions within the research team, informed by the theoretical considerations and research gaps identified in Section 2. Consistent with the study’s focus on AR-enabled indoor navigation as a multifaceted service rather than solely as a wayfinding technology, attribute selection focused on what functions, information, experiences, and rewards the service would provide to users and at what price. Accordingly, technical performance characteristics and other potential service attributes outside this scope were not manipulated in the conjoint design. Respondents were explicitly instructed, through a fixed instruction panel displayed above each choice task, to assume that all features other than those presented in the alternatives were identical across alternatives. This allowed the conjoint experiment to focus on trade-offs among the six attributes examined in this study. No separate formal content-validation procedure was conducted for the final attribute set.

Table 2. Attributes and Levels.

CBC Attribute	Attribute Levels
Navigation Mode	Map Only; Map + AR Guidance
Information Layering	Route Only; Route + Facility Information; Route + Facility Information + Discount Information
Interactive Character Feature	Interactive Character Absent; Interactive Character Present
Reward Type	Instant Discount Coupon; Reward Points; Prize Draw Entry
Reward Acquisition Mode	Visit-Based; Mission-Based; Social-Interaction-Based
Price	KRW 990 per month; KRW 1990 per month; KRW 2990 per month

Navigation Mode (NAV) had two levels: Map Only and Map + AR Guidance. Although the service is referred to as AR-enabled indoor navigation, AR guidance was varied as an attribute: in the Map-Only reference level, route information was displayed on a 2D map, whereas in the Map + AR Guidance level the same information was also overlaid on the physical environment. Including this reference level allows the incremental utility of AR guidance to be estimated relative to conventional map-based navigation. Information Layering (INFO) had three levels: Route Only, Route + Facility Information, and Route + Facility Information + Discount Information. Discount Information displayed discount and promotion information for nearby stores but did not itself issue a coupon or other reward; it therefore differs from Instant Discount Coupon, a level of Reward Type in which a coupon usable at the store was issued to the user as a reward. Interactive Character Feature (CHAR) had two levels: Absent and Present. In the Present condition, a virtual character accompanied the user during navigation, and users could interact with and nurture the character through activities such as feeding, playing, and leveling up. Reward Acquisition Mode (ACQ) had three levels: Visit-Based, in which a reward was automatically provided upon arrival at a designated store; Mission-Based, in which a reward was provided after completion of an app-based mission; and Social-Interaction-Based, in which a reward was provided after a collaborative action with other users. Reward Type (REWARD) also had three levels: Instant Discount Coupon, Reward Points, and Prize Draw Entry. No monetary amount was specified for any reward format; respondents evaluated reward formats rather than rewards of equivalent expected value. Differences among the reward-type coefficients may therefore reflect respondents’ inferences about value as well as differences in immediacy, certainty and redemption effort.

Price was operationalized as a monthly subscription fee at KRW 990, KRW 1990, and KRW 2990 (approximately USD 0.65, 1.30, and 1.96 at the Bank of Korea average exchange rate for June 2026, the month of data collection, of KRW 1527.30 per USD). The monthly subscription format reflected the study’s conceptualization of AR-enabled indoor navigation not as a one-time wayfinding tool for a single visit, but as a service that could provide continued value over repeated visits through navigation, information, experiential, and reward features. The three price levels were selected with reference to lower-priced Korean consumer subscription services, such as Kurly Members and Baemin Club, while the three-level structure was also informed by a prior mobile-app choice experiment [13]. Given the hypothetical nature of the service and the absence of an established market price, the three levels are researcher-defined experimental levels rather than market-derived prices, and the price attribute was intended primarily as a monetary numeraire for estimating trade-offs among the tested service profiles rather than as a validated estimate of the service’s market price. Because the monthly fee was varied over only three levels spanning KRW 2000, the range-based relative importance of Price is inherently conditional on the width of this experimental range. The MWTP estimates are conceptually distinct from range-based relative importance, but they are derived from the price coefficient estimated from trade-offs observed within the tested fee range and should therefore not be assumed to generalize unchanged to substantially different price ranges.

3.1.2. Choice Task Design

An L24 orthogonal plan was generated using SPSS Orthoplan (IBM SPSS Statistics, version 31.0; IBM Corp., Armonk, NY, USA) as a profile-level design for estimating the main effects of the six attributes. This design was used to obtain a manageable set of profiles while preserving the orthogonality and level balance needed for main-effect estimation. The resulting profile-level design was nearly orthogonal (maximum absolute correlation between dummy variables of different attributes = 0.125), and the main-effects design matrix was of full rank. Profile generation and choice-task construction were two separate steps: the plan produced by Orthoplan specified profiles only, and the resulting 24 service profiles were grouped by the research team into eight fixed choice tasks, each containing three unlabeled alternatives. Because this grouping was not generated using a choice-design efficiency criterion, we evaluated the resulting choice sets directly. The resulting multinomial-logit D-error was 0.599 under zero priors and 0.928 under the estimated coefficients. Among 10,000 random reallocations of the same 24 profiles into eight three-alternative choice sets, none produced a lower D-error under zero priors and fewer than 0.1% produced a lower D-error under the estimated coefficients, the latter serving as a post hoc local-efficiency check. The maximum absolute correlation among dummy-coded attribute differences between alternatives within each task was 0.221. Level overlap occurred in all eight tasks for each two-level attribute, as necessarily follows from presenting three alternatives with two levels, and in 0, 2, 3, and 6 of the eight tasks for Information Layering, Reward Type, Reward Acquisition Mode, and Price, respectively. Additional design diagnostics, including a parameter-recovery simulation, and the complete choice-set design are reported in Tables S1–S15 of the Supplementary Materials. The alternatives were identified by the generic labels A, B, and C and were presented in fixed positions and in the same order to all respondents; neither task sequence nor alternative position was randomized. The implications of this fixed presentation for response bias are examined in Section 4.2 and Section 5.7. Across the 24 profiles, each level of every two-level attribute appeared 12 times, and each level of every three-level attribute, including Price, appeared eight times, ensuring overall balance across attribute levels.

In each task, respondents simultaneously compared three service alternatives with different combinations of attribute levels and selected the one they most preferred, assuming that they would actually use the service. The forced-choice format enabled trade-offs among attribute levels to be measured [48]; however, respondents could not decline all three alternatives. The resulting choices therefore indicate relative preferences among the presented paid alternatives conditional on selecting one of them, rather than actual adoption or subscription decisions. Each task was displayed as a table, with the six attributes in rows and alternatives A, B, and C in columns; each cell showed the attribute level assigned to the corresponding alternative. A complete reference table defining all attributes and levels appeared below every task, allowing respondents to consult the definitions while making their choice.

3.2. Pilot Study and Main Survey

Before the main survey, a pilot test was conducted with 100 adults aged 20–69 years to assess the stimuli and attribute explanations, survey completion time, and difficulties in completing the choice tasks. Comprehension was checked using two multiple-choice questions concerning AR guidance and Mission-Based reward acquisition; however, age-group differences in comprehension could not be evaluated because incorrect attempts and response revisions were not recorded. The pilot also included 5-point items measuring familiarity with AR-based apps and enjoyment of trying new IT technologies. Participants aged 20–49 reported slightly higher AR-app familiarity (M = 3.180, SD = 0.764) than those aged 50–69 (M = 3.026, SD = 0.811) and slightly greater enjoyment of trying new IT technologies (M = 3.492, SD = 0.829 vs. M = 3.205, SD = 0.864). These age-group differences were not statistically significant (p = 0.344 and p = 0.104, respectively). Following the pilot, the scope of the main study was restricted to adults aged 20–49. This restriction was informed by the descriptive pilot results but was not based on a statistically established age-group difference. Accordingly, the findings of the present study should be interpreted as reflecting the preferences of adults aged 20–49 rather than those of the broader adult population.

The recruitment plan approved by the Institutional Review Board (IRB) specified a total of 400 respondents: 100 for the pilot test and 300 for the main survey. The pilot data were excluded from the main analysis because they were collected for preliminary assessment and covered a broader age range than the main survey. For the main survey, 300 adults aged 20–49 years were recruited, with equal quotas of 100 participants aged 20–29, 30–39, and 40–49 years to ensure balance across age groups. Five respondents who selected the alternative in the same position across all eight choice tasks were excluded. The final analytic sample consisted of 295 respondents.

The target main-survey sample of 300 respondents was determined based on available study resources rather than a formal prospective power calculation. As a post hoc assessment of sample-size adequacy for aggregate main-effect estimation, we applied the conventional CBC rule of thumb [48]:

$$n\ge {\displaystyle \frac{500c}{t \times a}}$$

where $c$ is the largest number of levels for any attribute, $t$ is the number of choice tasks used for estimation, and $a$is the number of alternatives per task. $c=3$, $t=8$, and $a=3$, the resulting minimum sample-size guideline was 62.5 respondents; thus, the final analytic sample of 295 exceeded this guideline. The data-collection protocol was reviewed and approved by the Sungkyunkwan University Institutional Review Board (IRB; approval no. 2026-05-057-001). The main survey was administered through Embrain (Embrain Co., Ltd., Seoul, Republic of Korea), an online panel provider, between 9 June and 15 June 2026. The questionnaire proceeded in the following sequence: introduction of the service concept and a comprehension check; one holdout choice task (Q1); eight conjoint choice tasks (Q2–Q9); measures of gamification-feature affinity and shopping mall visit frequency; and demographic questions.

Before the attribute explanations and choice tasks, respondents were given textual definitions of AR and indoor AR wayfinding and viewed screenshots illustrating an existing indoor AR wayfinding service provided by Naver (NAVER Corp., Seongnam, Republic of Korea). The screenshots were accompanied by a five-step textual explanation covering route search, environmental scanning, movement between floors, AR guidance, and arrival. Respondents were explicitly informed that the screenshots were provided only as illustrative examples and that the interface and functions of the service evaluated in the survey could differ from those depicted in these examples. Respondents then viewed separate explanation pages for each of the six attributes. Each page combined textual descriptions of the attribute levels with static images, and concrete usage examples were provided for potentially unfamiliar levels, including Mission-Based and Social-Interaction-Based reward acquisition. Comprehension questions were included to assess respondents’ understanding of the key concepts, and a correct response was required to proceed. Respondents who answered incorrectly were directed to review the relevant explanation and attempt the question again. The service description and each attribute explanation were presented on separate pages with a minimum display time of 20 s. The number of attempts was not capped, and the panel provider did not record response revisions or the number of incorrect attempts. The first choice task (Q1) was a holdout task specified in the survey design to assess predictive validity. It consisted of three profiles that were not part of the L24 plan, was presented in the same format as the eight estimation tasks without being identified to respondents as different, and was excluded from utility estimation; only the eight estimation tasks (Q2–Q9) were used to estimate the model.

3.3. Moderator Measures and Analysis Procedures

In addition to the full-sample model, the analysis evaluated preference heterogeneity associated with context-specific gamification-feature affinity and shopping mall visit frequency. Formal moderation was assessed using separate pooled conditional logit models, one for each moderator, containing attribute-by-moderator interaction terms. Descriptive subgroup models were retained to show the direction and magnitude of coefficients within each group. Gender- and age-based interaction models were estimated as additional exploratory checks. For gamification-feature affinity and shopping mall visit frequency, the mean-centered continuous variable was used as the primary moderator. Because the number of visits was right-skewed, the visit-frequency model was also estimated with log-transformed visits as a sensitivity analysis. Median-split subgroup models were used only for descriptive illustration.

Following the choice tasks, context-specific gamification-feature affinity was measured with four researcher-developed items rated on a five-point Likert scale (1 = strongly disagree; 5 = strongly agree): “I enjoy using apps that include game-like elements such as points, missions, and characters”; “I enjoy the process of progressing toward and achieving goals”; “I like completing missions together with other people”; and “I am interested in content that involves nurturing or developing a character.” The items were written for this study to capture affinity for the game-like features included in the service concept; they were not adapted from a previously validated scale and were not subjected to independent expert or consumer content validation. The items showed good internal consistency (Cronbach’s α = 0.802), and their mean was used as the affinity score. The item correlation matrix had one eigenvalue greater than 1 (2.53; the next was 0.60), accounting for 63.2% of the variance, and loadings on a single factor ranged from 0.68 to 0.77. Parallel analysis based on principal components retained one component, whereas the factor-based version retained additional factors with eigenvalues close to zero (Table S16). Three of the four items refer to features that also appear in the conjoint profiles, creating content overlap between the moderator and the conjoint attributes. Thus, an interaction may reflect correspondence between affinity for a specific feature and preference for that feature rather than a broader gamification disposition. The measure is therefore treated as context-specific feature affinity rather than as a general gamification disposition. For descriptive subgroup models, respondents were divided at the sample median of 3.25 into a higher-affinity group (n = 164) and a lower-affinity group (n = 131). The mean-centered continuous score was used in the primary moderation model, and two alternative specifications were examined as sensitivity checks: a three-item score that excluded the character-nurturing item and a single-item score based on the item referring to game-like elements such as points, missions, and characters. Because both alternative specifications retained some content correspondence with the conjoint features, these checks do not eliminate the content-overlap limitation.

Shopping mall visit frequency was measured as the reported number of visits to large, mixed-use shopping malls during the previous three months. For descriptive subgroup models, respondents reporting three or more visits were assigned to the higher-frequency group (n = 164), while those reporting fewer than three visits were assigned to the lower-frequency group (n = 131). Visit frequency is treated as a proxy for repeated exposure to shopping-mall environments rather than as a direct measure of familiarity with any particular mall.

Pooled interaction models included all nine attribute-level coefficients and their interactions with each moderator. Robust joint Wald tests assessed whether each moderator was associated with the overall preference structure. Individual interaction p-values were adjusted within moderator using the Benjamini–Hochberg procedure. Conditional logit models were also estimated independently within each subgroup for descriptive interpretation. Because coefficients from independently estimated logit models may differ in scale, cross-group coefficient magnitudes were not treated as formal evidence of heterogeneity; inferential conclusions were based on the pooled interaction models.

3.4. Data Analysis

Choice data from the choice-based conjoint experiment were analyzed using a conditional logit model (CLM) [49]. The CLM relates the attribute levels defining each alternative to its probability of being chosen within a choice set, assuming that respondents select the alternative that yields the highest utility. The CLM relies on the independence of irrelevant alternatives (IIA) assumption, under which the ratio of the choice probabilities of any two alternatives does not depend on the other alternatives in the choice set, and it assumes that preferences are homogeneous across respondents. Because the alternatives were unlabeled generic profiles, an alternative-removal test of IIA would be difficult to interpret. Respondent-clustered robust standard errors account for dependence among repeated choices by the same respondent but do not relax either assumption. A panel mixed logit model, which allows preferences to vary across respondents and does not impose IIA on the aggregate choice probabilities, was therefore estimated as a robustness check.

U_nij = β₁D_NAV + β₂D_INFO2 + β₃D_INFO3 + β₄D_CHAR + β₅D_REWARD2
+ β₆D_REWARD3 + β₇D_ACQ2 + β₈D_ACQ3 + β₉X_PRICE + ε_nij

Here, D_NAV denotes Navigation Mode; D_INFO2 and D_INFO3 denote Route + Facility Information and Route + Facility Information + Discount Information, respectively; D_CHAR denotes the Interactive Character Feature; D_REWARD2 and D_REWARD3 denote Reward Points and Prize Draw Entry, respectively; and D_ACQ2 and D_ACQ3 denote Mission-Based and Social-Interaction-Based acquisition, respectively. The subscripts n, i, and j denote respondents, choice tasks, and alternatives, and ε_nij is the random error term. Categorical attributes were dummy coded relative to Map Only, Route Only, Interactive Character Absent, Instant Discount Coupon, and Visit-Based. Price was treated as continuous over KRW 990–2990 and scaled in units of KRW 1000. Models were estimated in R (version 4.6.0; R Foundation for Statistical Computing, Vienna, Austria) using survival::clogit (survival package, version 3.8.6), with each respondent-by-task combination defined as a choice set and respondent-clustered robust standard errors. The mixed logit model was estimated with the mlogit package (version 2.0.0) and the dfidx package (version 0.2.0). All nine coefficients, including Price, were specified as independently normally distributed random parameters; parameter correlations were not estimated, and the panel structure of the repeated choices was taken into account. The simulated likelihood was maximized using 1000 Halton draws, and the BFGS algorithm converged after 30 iterations (g′(−H)⁻¹g = 3.99 × 10⁻⁷). Because a normally distributed price coefficient allows positive values for some respondents, the mixed logit model was used only to check the signs and significance of the coefficients and as an indication of preference heterogeneity; MWTP was derived from the CLM.

Based on the estimated utility coefficients, marginal willingness to pay for non-price attribute level k was calculated as MWTP_k = −β_k/β_price. Because price was entered in units of KRW 1000, the ratio was multiplied by 1000 to express MWTP in KRW per month. Standard errors were estimated using the delta method. As a simulation-based check, Krinsky–Robb confidence intervals [50] were also computed from 10,000 draws from the multivariate normal distribution defined by the estimated coefficients and their respondent-clustered robust covariance matrix. Because respondents were required to choose among three paid alternatives and no opt-out, free-service, or existing-navigation option was included, the MWTP estimates represent conditional, relative monetary valuations within the experimental choice sets rather than direct estimates of consumers’ actual willingness to subscribe or market prices. Relative importance (RI) was calculated from each attribute’s utility range, defined as the difference between the maximum and minimum part-worth utilities within that attribute, with the dummy-coded reference level fixed at zero. Each attribute’s utility range was divided by the sum of all attribute ranges and multiplied by 100. The price range was calculated from the utility difference between the lowest and highest monthly fees presented. Accordingly, the RI values are conditional on the attribute levels and ranges included in the experimental design and should not be interpreted as invariant measures of attribute importance.

Predictive validity was assessed with the holdout task. The conditional logit coefficients estimated from Q2–Q9 were used to predict the choice shares of the three Q1 profiles, and the first-choice hit rate and the mean absolute error between predicted and observed shares were computed.

4. Results

4.1. Respondent Characteristics

A final sample of 295 participants was included in the analysis. The sociodemographic characteristics, digital-familiarity measures, and shopping mall visit frequency of the participants are presented in Table 3. Of the participants, 146 (49.5%) were men and 149 (50.5%) were women. By age group, 96 participants (32.5%) were aged 20–29 years, 99 (33.6%) were aged 30–39 years, and 100 (33.9%) were aged 40–49 years. Table 3 additionally reports participants’ education, occupation, monthly personal income, and responses to the three AR-experience and technology-adoption items. The two variables examined as moderators—shopping mall visit frequency and gamification-feature affinity—were also summarized descriptively. The average number of shopping mall visits over the previous three months was 3.50 (median = 3, range = 0–20). Gamification-feature affinity was measured using four items on a 5-point Likert scale. The scale demonstrated adequate internal consistency, with a Cronbach’s α of 0.802, and the median score was 3.25. The association between the two moderators was weak and nonsignificant (Pearson r = 0.080, p = 0.168; Spearman ρ = 0.057, p = 0.329; Table 4). Although both median splits produced groups of 164 and 131 respondents, the two classifications agreed for only 53.2% of respondents (χ²(1) = 0.81, p = 0.367, φ = 0.053).

Table 3. Sociodemographic, Digital-Familiarity, and Mall-Visit Characteristics of Respondents (N = 295).

Characteristic	Category	n (%)	M (SD)
Gender	Men	146 (49.5)	—
	Women	149 (50.5)	—
Age group	20–29 years	96 (32.5)	—
	30–39 years	99 (33.6)	—
	40–49 years	100 (33.9)	—
Visits to a large, mixed-use shopping mall during the preceding three months	0	23 (7.8)	—
	1–2	108 (36.6)	—
	3–5	116 (39.3)	—
	6–10	40 (13.6)	—
	11 or more	8 (2.7)	—
Education	High school or less	24 (8.1)	—
	Currently enrolled in college	34 (11.5)	—
	College graduate	183 (62.0)	—
	Graduate school enrolled or graduate degree	54 (18.3)	—
Occupation	Student	40 (13.6)	—
	Office worker	102 (34.6)	—
	Civil servant	20 (6.8)	—
	Professional	31 (10.5)	—
	Sales or service worker	35 (11.9)	—
	Skilled, agricultural/fishery, or manual worker	13 (4.4)	—
	Homemaker	27 (9.2)	—
	Other	27 (9.2)	—
Monthly personal income	<KRW 1.0 million	46 (15.6)	—
	KRW 1.0 to <2.0 million	30 (10.2)	—
	KRW 2.0 to <3.0 million	72 (24.4)	—
	KRW 3.0 to <4.0 million	55 (18.6)	—
	KRW 4.0 to <5.0 million	32 (10.8)	—
	≥KRW 5.0 million	60 (20.3)	—
AR experience and technology adoption	Familiar with using AR-based applications	—	3.22 (1.01)
	Hesitant to try new IT technologies	—	2.51 (0.98)
	Enjoy trying new IT technologies	—	3.59 (0.91)

Note. The three AR-experience and technology-adoption items were rated on a 5-point scale ranging from 1 (strongly disagree) to 5 (strongly agree). Higher scores on the hesitation item indicate greater hesitation toward new technology. Percentages may not sum to 100 because of rounding. There were no missing values for the variables reported in the table.

Table 4. Cross-Classification of Shopping Mall Visit Frequency and Gamification-Feature Affinity (N = 295).

	Lower Affinity (<3.25)	Higher Affinity (≥3.25)	Total
Lower visit frequency (<3 visits)	62	69	131
Higher visit frequency (≥3 visits)	69	95	164
Total	131	164	295

Note. Groups were formed at each variable’s sample median (three visits in the preceding three months; an affinity score of 3.25), with respondents at the median assigned to the higher group. Pearson r = 0.080 (p = 0.168); Spearman ρ = 0.057 (p = 0.329); χ²(1) = 0.81 (p = 0.367); φ = 0.053; classification agreement = 53.2%.

4.2. Conditional Logit Analysis for the Full Sample

A conditional logit model was estimated using the choice data from the full sample (N = 295; three alternatives × eight choice sets). Because each task required respondents to choose among three paid alternatives without an opt-out option, the estimated utilities represent relative preferences among the presented alternatives conditional on choosing one, rather than actual adoption or subscription decisions. The model log-likelihood was −1934.37, a significant improvement over the null model (log-likelihood = −2592.73; LR χ²(9) = 1316.71, p < 0.001). McFadden’s pseudo-R² was 0.254. Price, scaled in units of KRW 1000, had a significant negative coefficient (β = −0.953, z = −16.05, p < 0.001), indicating that a KRW 1000 increase in monthly fee was associated with an average utility decrease of 0.953.

The estimated part-worth utilities are presented in Table 5. Map + AR Guidance had positive utility relative to Map Only (β = 0.515, z = 6.72, p < 0.001). Both Route + Facility Information (β = 0.320, z = 4.44, p < 0.001) and Route + Facility Information + Discount Information (β = 0.465, z = 5.82, p < 0.001) had positive utility relative to Route Only; in a direct contrast between the two non-reference levels, computed from the robust covariance matrix from the same model, the condition that also included discount information showed higher utility (Δβ = 0.145, robust SE = 0.066, p = 0.028; Benjamini–Hochberg-adjusted p = 0.042 across the three contrasts). Reward Points (β = −0.136, z = −2.01, p = 0.044) and Prize Draw Entry (β = −0.412, z = −6.42, p < 0.001) had negative utility relative to Instant Discount Coupons, and Prize Draw Entry was valued below Reward Points (Δβ = −0.276, robust SE = 0.073, p < 0.001; adjusted p < 0.001). Social-Interaction-Based acquisition had negative utility relative to Visit-Based acquisition (β = −0.286, z = −4.85, p < 0.001), whereas Mission-Based acquisition was not significant at the 5% level (β = −0.134, p = 0.072); Social-Interaction-Based acquisition was valued below Mission-Based acquisition (Δβ = −0.152, robust SE = 0.077, p = 0.048; adjusted p = 0.048). The Interactive Character Feature was not significant in the full sample (β = 0.070, p = 0.177).

Table 5. Conditional Logit Coefficient Estimates for the Full Sample.

Attribute Levels	β	Robust SE	z	95% CI	p
Map + AR Guidance	0.515 ***	0.077	6.72	[0.365, 0.665]	<0.001
Route + Facility Information	0.320 ***	0.072	4.44	[0.179, 0.461]	<0.001
Route + Facility Information + Discount Information	0.465 ***	0.080	5.82	[0.308, 0.621]	<0.001
Interactive Character Present	0.070	0.052	1.35	[−0.032, 0.173]	0.177
Mission-Based	−0.134	0.074	−1.80	[−0.280, 0.012]	0.072
Social-Interaction-Based	−0.286 ***	0.059	−4.85	[−0.401, −0.170]	<0.001
Reward Points	−0.136 *	0.068	−2.01	[−0.269, −0.003]	0.044
Prize Draw Entry	−0.412 ***	0.064	−6.42	[−0.538, −0.286]	<0.001
Price (monthly fee; per KRW 1000)	−0.953 ***	0.059	−16.05	[−1.070, −0.837]	<0.001

Note. Reference levels were Map Only, Route Only, Interactive Character Absent, Visit-Based, and Instant Discount Coupon. Standard errors are clustered by respondent. Log-likelihood = −1934.37; McFadden’s pseudo-R² = 0.254. * p < 0.05; *** p < 0.001.

As a robustness check, the panel mixed logit model fitted the data significantly better than the conditional logit model (log-likelihood = −1708.03 versus −1934.37; LR χ²(9) = 452.67, p < 0.001; Table 6). All nine mean coefficients had the same signs as the conditional logit coefficients. Significance differed for two levels: the mean coefficient for Reward Points was not significant (β = −0.142, p = 0.172), whereas that for Mission-Based acquisition was negative and significant (β = −0.242, p = 0.021). The standard deviations were significant for seven of the nine coefficients, with the largest for Price (SD = 1.819, p < 0.001), but not for Route + Facility Information (SD = 0.383, p = 0.091) or Social-Interaction-Based acquisition (SD = 0.048, p = 0.911). Significant random-parameter standard deviations may reflect both preference heterogeneity and differences in choice consistency across respondents and therefore should not be interpreted as evidence of preference heterogeneity alone.

Table 6. Conditional Logit and Mixed Logit Estimates and Hypothesis Tests for the Full Sample.

Attribute Levels	Hypothesis	Conditional Logit β (Robust SE)	Mixed Logit Mean (SE)	Mixed Logit SD (SE)	Result
Map + AR Guidance	H1 (+)	0.515 *** (0.077)	0.947 *** (0.112)	1.483 *** (0.174)	Supported
Route + Facility Information	H2a (+)	0.320 *** (0.072)	0.520 *** (0.119)	0.383 † (0.226)	Supported
Route + Facility Information + Discount Information	H2b (+)	0.465 *** (0.080)	0.803 *** (0.121)	1.168 *** (0.154)	Supported
Interactive Character Present	H3 (+)	0.070 (0.052)	0.122 (0.087)	0.720 *** (0.168)	Not supported
Mission-Based	H4a (+)	−0.134 † (0.074)	−0.242 * (0.105)	0.718 *** (0.186)	Not supported (opposite sign)
Social-Interaction-Based	H4b (−)	−0.286 *** (0.059)	−0.708 *** (0.125)	0.048 (0.435)	Supported
Reward Points	H5a (−)	−0.136 * (0.068)	−0.142 (0.104)	0.936 *** (0.179)	Supported (CLM only)
Prize Draw Entry	H5b (−)	−0.412 *** (0.064)	−0.791 *** (0.112)	0.751 *** (0.193)	Supported
Price (monthly fee; per KRW 1000)	H6 (−)	−0.953 *** (0.059)	−2.039 *** (0.159)	1.819 *** (0.165)	Supported
Log-likelihood		−1934.37	−1708.03

Note. Reference levels were Map Only, Route Only, Interactive Character Absent, Instant Discount Coupon, and Visit-Based. Price is expressed per KRW 1000. The mixed logit model specified all nine coefficients as independent normally distributed random parameters (panel specification; 1000 Halton draws). Because mixed logit and conditional logit coefficients are not directly comparable in magnitude, the mixed logit results are used here primarily to assess the robustness of coefficient direction and statistical significance. SDs are reported as absolute values. † p < 0.10; * p < 0.05; *** p < 0.001. Results refer to the conditional logit model; “CLM only” indicates that the mean coefficient was not significant in the mixed logit model.

Table 6 summarizes the hypothesis tests. In the conditional logit model, H1, H2a, H2b, H4b, H5a, H5b, and H6 were supported, whereas H3 and H4a were not (Table 6). The mixed logit results were generally consistent with these conclusions, except that the mean coefficient for Reward Points (H5a) was not statistically significant. Mission-Based acquisition, contrary to H4a, had a negative coefficient relative to Visit-Based acquisition, which reached statistical significance in the mixed logit model.

Alternatives in the first, second, and third positions were chosen 841, 977, and 542 times across the eight estimation tasks, which departs from a uniform distribution (χ²(2) = 125.90, p < 0.001). Because the profiles shown in each position differed, this comparison does not separate position from profile attractiveness. Because alternative position was fixed rather than randomized, the model was re-estimated with indicators for the second and third positions (Table S15). The third-position indicator was negative (β = −0.438, robust SE = 0.087, p < 0.001; joint Wald χ²(2) = 28.55, p < 0.001). All nine attribute-level coefficients retained their signs and significance status. However, two of the three pairwise contrasts were sensitive to position adjustment; unadjusted p-values are reported for these sensitivity analyses. The contrast between Route + Facility Information + Discount Information and Route + Facility Information changed sign and was no longer statistically significant (Δβ = −0.125, p = 0.108), and the contrast between Prize Draw Entry and Reward Points was also no longer significant (Δβ = −0.033, p = 0.702). The contrast between Social-Interaction-Based and Mission-Based acquisition remained significant (Δβ = −0.187, p = 0.039). Because alternative position was associated with Information Layering in the fixed design (Cramér’s V = 0.515), position and Information Layering effects cannot be fully disentangled. This specification is therefore treated as a sensitivity analysis rather than as an adjusted primary model.

4.3. Marginal Willingness to Pay and Relative Importance by Attribute

The conditional MWTP estimates relative to the reference levels are presented in Table 7. Within the experimental choice sets, MWTP was highest for Map + AR Guidance at approximately KRW 540, followed by Route + Facility Information + Discount Information at KRW 488 and Route + Facility Information at KRW 336. In contrast, Prize Draw Entry and Social-Interaction-Based acquisition showed negative MWTP values relative to Instant Discount Coupon and Visit-Based rewards, respectively (KRW −432 and KRW −300). Negative MWTP indicates a negative relative monetary valuation compared with the corresponding reference level within the experimental choice sets. For example, the KRW −432 estimate for Prize Draw Entry indicates that its utility difference relative to Instant Discount Coupon corresponds to approximately KRW 432 in monthly-fee terms within the experimental choice sets. Because these estimates are obtained as the negative ratio of each attribute coefficient to the price coefficient, which was estimated from trade-offs within the tested fee range of KRW 990–2990, they should not be assumed to apply unchanged to substantially different price ranges.

Table 7. Marginal Willingness to Pay Estimates by Attribute Level.

Attribute	Attribute Level	MWTP (KRW Per Month)	Delta-Method 95% CI	Krinsky–Robb 95% CI
Navigation Mode	Map Only	Reference
	Map + AR Guidance	+540	[377, 702]	[380, 710]
Information Layering	Route Only	Reference
	Route + Facility Information	+336	[187, 484]	[189, 495]
	Route + Facility Information + Discount Information	+488	[326, 649]	[328, 654]
Interactive Character Feature	Interactive Character Absent	Reference
	Interactive Character Present	+74	[−33, 181]	[−34, 181]
Reward Type	Instant Discount Coupon	Reference
	Reward Points	−143	[−281, −5]	[−282, −5]
	Prize Draw Entry	−432	[−561, −303]	[−563, −306]
Reward Acquisition Mode	Visit-Based	Reference
	Mission-Based	−140	[−299, 18]	[−310, 12]
	Social-Interaction-Based	−300	[−416, −183]	[−417, −183]

Note. Reference levels: Map Only, Route Only, Interactive Character Absent, Instant Discount Coupon, and Visit-Based. MWTP estimates are reported in KRW per month. They are conditional, relative monetary valuations within the experimental choice sets rather than estimates of market willingness to pay. Delta-method 95% confidence intervals were computed from the respondent-clustered robust covariance matrix; Krinsky–Robb intervals are based on 10,000 draws from the estimated coefficient distribution. At the Bank of Korea average exchange rate for June 2026 (KRW 1527.30 per USD), KRW 1000 corresponds to approximately USD 0.65.

The Krinsky–Robb intervals were closely aligned with the delta-method intervals. For Map + AR Guidance, the delta-method interval was [KRW 377, KRW 702] and the Krinsky–Robb interval was [KRW 380, KRW 710]; the corresponding intervals were [KRW −561, KRW −303] and [KRW −563, KRW −306] for Prize Draw Entry and [KRW −416, KRW −183] and [KRW −417, KRW −183] for Social-Interaction-Based acquisition. The choice of method did not change the inference for any estimate.

The relative importance of each attribute was calculated based on the range of its part-worth utilities. The results showed the following order: Price (52.2%) > Navigation Mode (14.1%) > Information Layering (12.7%) > Reward Type (11.3%) > Reward Acquisition Mode (7.8%) > Interactive Character Feature (1.9%) (see Table 8). Range-based relative importance depends on the levels selected for each attribute and on the width of each attribute’s range, and Price was the only attribute modeled as a continuous variable using three researcher-defined experimental levels. These percentages therefore apply only to the levels and ranges tested in this experiment and should not be read as stable or general measures of attribute importance; for example, applying the same estimated price coefficient to a hypothetical price range half as wide (KRW 990–1990) would reduce the range-based share attributed to Price from 52.2% to approximately 35%. Within the tested ranges, Price accounted for the largest share, followed by Navigation Mode and Information Layering. In contrast, Interactive Character Feature had the lowest relative importance at 1.9%, but its full-sample coefficient was not statistically distinguishable from zero (β = 0.070, p = 0.177), so this value should not be interpreted as evidence of a small but established effect.

Table 8. Utility Ranges and Relative Importance of the Attributes.

Attribute	Utility Range	RI (%)
Price	1.906	52.2
Navigation Mode	0.515	14.1
Information Layering	0.465	12.7
Reward Type	0.412	11.3
Reward Acquisition Mode	0.286	7.8
Interactive Character Feature	0.070	1.9

Note. RI was calculated from the range of the point-estimated part-worth utilities within each attribute and therefore depends on the levels and ranges included in the design; the price range corresponds to the tested fee range of KRW 990–2990. The RI for Interactive Character Feature was derived from a coefficient that was not statistically significant in the full sample (β = 0.070, p = 0.177); its 95% Krinsky–Robb interval was [0.1%, 4.7%]. Because RI is calculated from non-negative utility ranges, this interval is bounded at zero by construction and therefore does not provide evidence of a non-zero character effect.

Among the non-price attributes, Navigation Mode and Information Layering showed the highest range-based relative importance and positive MWTP. These RI values depend on the attribute levels and ranges tested, while the MWTP estimates rely on the price coefficient estimated within the KRW 990–2990 fee range. Neither should therefore be read as applying unchanged outside the tested attribute levels and fee range.

4.4. Moderation and Subgroup Analyses

Separate pooled interaction models, each with one mean-centered continuous moderator, formally tested whether the overall preference structure varied by shopping mall visit frequency and context-specific gamification-feature affinity. Descriptive subgroup estimates are reported alongside the interaction tests to aid interpretation. Gender and age were examined in additional exploratory interaction models.

4.4.1. Visit Frequency and Gamification-Feature Affinity

The overall preference structure varied with shopping mall visit frequency. In the primary model using the mean-centered number of visits (joint robust Wald χ²(9) = 27.49, p = 0.001), only the Monthly Fee × Visit Frequency interaction remained significant after Benjamini–Hochberg correction (γ = 0.051, robust SE = 0.015, adjusted p = 0.010), indicating lower price sensitivity among respondents who visited more often (Table S18). Because the number of visits was right-skewed, the model was re-estimated with log-transformed visit frequency; the monthly-fee interaction remained significant (γ = 0.367, adjusted p < 0.001), and a negative Reward Points interaction also survived correction (γ = −0.332, adjusted p = 0.008; Table S19).

The median-split estimates in Table 9 describe coefficient patterns within the higher- and lower-frequency groups. Because separately estimated logit models may differ in scale and the moderator was dichotomized, differences in the significance or magnitude of coefficients across these subgroups are descriptive. A negative Mission-Based interaction appeared in the median-split interaction model (γ = −0.582, adjusted p < 0.001) but not in the continuous or log-transformed specifications (adjusted p = 0.616 and 0.126, respectively; Table S23), indicating that this pattern depends on how visit frequency is coded.

The overall preference structure also differed by context-specific gamification-feature affinity. In the primary model using the mean-centered continuous score (joint robust Wald χ²(9) = 34.20, p < 0.001), two of the nine interactions remained significant after Benjamini–Hochberg correction: Interactive Character × Affinity (γ = 0.239, robust SE = 0.075, adjusted p = 0.009) and Monthly Fee × Affinity (γ = 0.241, robust SE = 0.077, adjusted p = 0.009). No other interaction survived correction (Table S17). The character interaction remained significant when the character item was excluded from the score (γ = 0.207, adjusted p = 0.014) and when only the item referring to game-like elements such as points, missions, and characters was used (γ = 0.184, adjusted p = 0.002). However, because these alternative specifications still capture affinity toward game-like features, content correspondence cannot be ruled out. The median-split estimates in Table 9 are presented only to illustrate coefficient patterns within the higher- and lower-affinity groups.

Table 9. Conditional Logit Coefficient Estimates by Visit Frequency and Gamification-Feature Affinity.

Attribute Level	Visit Frequency Higher (n = 164)	Visit Frequency Lower (n = 131)	Gamification-Feature Affinity Higher (n = 164)	Gamification-Feature Affinity Lower (n = 131)
	β (Robust SE), p	β (Robust SE), p	β (Robust SE), p	β (Robust SE), p
Map + AR Guidance	+0.441 (0.092), <0.001	+0.716 (0.138), <0.001	+0.441 (0.093), <0.001	+0.657 (0.138), <0.001
Route + Facility Information	+0.253 (0.088), 0.004	+0.456 (0.130), <0.001	+0.189 (0.091), 0.037	+0.580 (0.123), <0.001
Route + Facility Information + Discount Information	+0.387 (0.100), <0.001	+0.615 (0.134), <0.001	+0.372 (0.096), <0.001	+0.694 (0.146), <0.001
Interactive Character Present	+0.088 (0.067), 0.187	+0.050 (0.086), 0.564	+0.199 (0.062), 0.001	−0.173 (0.097), 0.074
Mission-Based	−0.351 (0.092), <0.001	+0.231 (0.124), 0.062	−0.174 (0.095), 0.068	−0.034 (0.122), 0.783
Social-Interaction-Based	−0.276 (0.072), <0.001	−0.306 (0.106), 0.004	−0.268 (0.073), <0.001	−0.346 (0.105), 0.001
Reward Points	−0.190 (0.084), 0.024	−0.036 (0.124), 0.775	−0.024 (0.091), 0.793	−0.323 (0.105), 0.002
Prize Draw Entry	−0.402 (0.085), <0.001	−0.508 (0.093), <0.001	−0.325 (0.088), <0.001	−0.562 (0.093), <0.001
Monthly fee (per KRW 1000)	−0.770 (0.072), <0.001	−1.269 (0.100), <0.001	−0.812 (0.073), <0.001	−1.175 (0.099), <0.001

Note. Reference levels were Map Only, Route Only, Interactive Character Absent, Visit-Based, and Instant Discount Coupon. Cells show coefficients from separately estimated conditional logit models, with respondent-clustered robust standard errors in parentheses, followed by p-values. Subgroups were formed by median splits; these estimates are descriptive and differences between groups are not tested here. Inferential conclusions are based on the continuous interaction models (Table 10).

Table 9. Conditional Logit Coefficient Estimates by Visit Frequency and Gamification-Feature Affinity.

Attribute Level	Visit Frequency Higher (n = 164)	Visit Frequency Lower (n = 131)	Gamification-Feature Affinity Higher (n = 164)	Gamification-Feature Affinity Lower (n = 131)
	β (Robust SE), p	β (Robust SE), p	β (Robust SE), p	β (Robust SE), p
Map + AR Guidance	+0.441 (0.092), <0.001	+0.716 (0.138), <0.001	+0.441 (0.093), <0.001	+0.657 (0.138), <0.001
Route + Facility Information	+0.253 (0.088), 0.004	+0.456 (0.130), <0.001	+0.189 (0.091), 0.037	+0.580 (0.123), <0.001
Route + Facility Information + Discount Information	+0.387 (0.100), <0.001	+0.615 (0.134), <0.001	+0.372 (0.096), <0.001	+0.694 (0.146), <0.001
Interactive Character Present	+0.088 (0.067), 0.187	+0.050 (0.086), 0.564	+0.199 (0.062), 0.001	−0.173 (0.097), 0.074
Mission-Based	−0.351 (0.092), <0.001	+0.231 (0.124), 0.062	−0.174 (0.095), 0.068	−0.034 (0.122), 0.783
Social-Interaction-Based	−0.276 (0.072), <0.001	−0.306 (0.106), 0.004	−0.268 (0.073), <0.001	−0.346 (0.105), 0.001
Reward Points	−0.190 (0.084), 0.024	−0.036 (0.124), 0.775	−0.024 (0.091), 0.793	−0.323 (0.105), 0.002
Prize Draw Entry	−0.402 (0.085), <0.001	−0.508 (0.093), <0.001	−0.325 (0.088), <0.001	−0.562 (0.093), <0.001
Monthly fee (per KRW 1000)	−0.770 (0.072), <0.001	−1.269 (0.100), <0.001	−0.812 (0.073), <0.001	−1.175 (0.099), <0.001

Note. Reference levels were Map Only, Route Only, Interactive Character Absent, Visit-Based, and Instant Discount Coupon. Cells show coefficients from separately estimated conditional logit models, with respondent-clustered robust standard errors in parentheses, followed by p-values. Subgroups were formed by median splits; these estimates are descriptive and differences between groups are not tested here. Inferential conclusions are based on the continuous interaction models (Table 10).

Table 10. Pooled Interaction Tests for Preference Heterogeneity.

Moderator	Interaction	Estimate	Robust SE	p	BH-Adjusted p
Visit frequency (continuous)	Monthly fee × visits	0.051	0.015	0.001	0.010
Feature affinity (continuous)	Interactive character × affinity	0.239	0.075	0.002	0.009
Feature affinity (continuous)	Monthly fee × affinity	0.241	0.077	0.002	0.009

Note. Both moderators were entered as mean-centered continuous variables. Joint robust Wald tests were significant for visit frequency, χ²(9) = 27.49, p = 0.001, and gamification-feature affinity, χ²(9) = 34.20, p < 0.001. Only interactions surviving Benjamini–Hochberg correction are shown. With log-transformed visit frequency, the monthly fee (γ = 0.367) and Reward Points (γ = −0.332) interactions survived correction (Table S19).

Table 10. Pooled Interaction Tests for Preference Heterogeneity.

Moderator	Interaction	Estimate	Robust SE	p	BH-Adjusted p
Visit frequency (continuous)	Monthly fee × visits	0.051	0.015	0.001	0.010
Feature affinity (continuous)	Interactive character × affinity	0.239	0.075	0.002	0.009
Feature affinity (continuous)	Monthly fee × affinity	0.241	0.077	0.002	0.009

Note. Both moderators were entered as mean-centered continuous variables. Joint robust Wald tests were significant for visit frequency, χ²(9) = 27.49, p = 0.001, and gamification-feature affinity, χ²(9) = 34.20, p < 0.001. Only interactions surviving Benjamini–Hochberg correction are shown. With log-transformed visit frequency, the monthly fee (γ = 0.367) and Reward Points (γ = −0.332) interactions survived correction (Table S19).

4.4.2. Additional Exploratory Moderation Analyses by Gender and Age

Gender and age were not prespecified in RQ2 and were examined as additional exploratory moderators. Joint interaction tests were not significant for gender (Wald χ²(9) = 13.2, p = 0.154) or age group (Wald χ²(18) = 21.4, p = 0.260), and no individual interaction survived Benjamini–Hochberg correction.

4.5. Holdout Predictive Validity

The conditional logit model estimated from the eight estimation tasks was used to predict choices in the holdout task (Q1), whose three profiles did not appear in the estimation design (Table 11). The predicted choice shares for Alternatives A, B, and C were 7.1%, 73.0%, and 19.9%, whereas the observed shares were 18.3% (n = 54), 64.7% (n = 191), and 16.9% (n = 50). Because the model assigns the same predicted probabilities to all respondents, it predicts Alternative B for every respondent, and the first-choice hit rate therefore equals the observed share of Alternative B: 191 of 295 respondents (64.7%), compared with the 33.3% expected by chance (binomial p < 0.001). This rate lies within the range of task-specific hit rates in the estimation tasks (49.2–84.4%; mean = 67.7%). Aggregate share prediction was less accurate: the mean absolute error across the three shares was 7.46 percentage points, with the largest error for Alternative A (predicted 7.1% versus observed 18.3%), which combined AR guidance with the highest fee and Prize Draw Entry. Because the holdout task was presented first, its fit may also reflect warm-up effects. The holdout results therefore provide limited support for the model’s ability to identify the most frequently chosen alternative outside the estimation design, but not for precise prediction of aggregate choice shares.

Table 11. Holdout Task (Q1): Profiles and Predicted and Observed Choice Shares (N = 295).

Alternative	Profile	Predicted Share (%)	Observed Share (%)
A	Map + AR Guidance; Route Only; Interactive Character Absent; Prize Draw Entry; Mission-Based; KRW 2990	7.1	18.3 (n = 54)
B	Map Only; Route + Facility Information; Interactive Character Present; Instant Discount Coupon; Visit-Based; KRW 990	73.0	64.7 (n = 191)
C	Map Only; Route + Facility Information + Discount Information; Interactive Character Absent; Reward Points; Social-Interaction-Based; KRW 1990	19.9	16.9 (n = 50)

Note. Predicted shares were computed from the conditional logit model estimated on Q2–Q9 (Table 5). First-choice hit rate = 64.7%; mean absolute error of the choice shares = 7.46 percentage points.

5. Discussion

5.1. Functional Value of the AR-Enabled Phygital Interface

This study conceptualized AR-enabled indoor navigation in large, mixed-use shopping malls as a phygital commerce interface and quantified the choice utility associated with Navigation Mode, Information Layering, an Interactive Character Feature, Reward Acquisition Mode, Reward Type, and monthly price, as well as the conditional MWTP for non-price attribute levels. The conditional logit analysis used 2360 choices from 295 South Korean adults aged 20–49. These estimates should be interpreted as relative stated preferences among the paid service alternatives presented in the experiment, in which no opt-out or free alternative was available, rather than as evidence of actual adoption or subscription decisions. Viewed through the framework developed in Section 2.8, the result for Navigation Mode indicates positive utility for a feature that primarily provides functional value by supporting the focal wayfinding task.

Compared with Map-Only conditions, Map + AR Guidance showed significant positive utility (p < 0.001). Wayfinding is a complex cognitive process that involves perceiving spatial information, determining the direction of travel, and translating a selected route into actual movement [20]. While 2D maps require users to mentally transform map information from an external perspective to align it with their current position and orientation in the physical environment [6], AR guidance can reduce the burden of mapping this information onto the real world by directly overlaying route and directional cues onto the physical environment [7,8]. In addition, Ref. [51] showed that AR guideposts may reduce cognitive burden and support the formation of spatial cognitive maps by providing route knowledge directly to users.

Accordingly, the positive utility and relatively high conditional MWTP associated with AR guidance within the experimental choice sets suggest that users may have evaluated AR route guidance not merely as an additional visual feature, but as a function that facilitates wayfinding. However, because this study did not directly measure cognitive load or actual wayfinding performance, it is difficult to conclude that AR guidance actually reduced cognitive burden. These findings suggest that one source of value in a phygital interface may lie in how digital features support users in addressing tasks encountered in physical space.

5.2. Contextual Value of Information Layering

Information Layering also showed positive utility relative to Route Only when facility information, or facility information together with discount information, was added to route guidance. Both Route + Facility Information (β = 0.320) and Route + Facility Information + Discount Information (β = 0.465) yielded higher choice utility than Route Only (p < 0.001). In a direct contrast, the condition that also included discount information showed higher utility than Route + Facility Information alone (Δβ = 0.145, BH-adjusted p = 0.042), although this difference was not robust to adjustment for alternative position (Table S15). Conditional MWTP within the experimental choice sets was estimated at KRW 336 and KRW 488, respectively. Importantly, the KRW 488 estimate is relative to Route Only and does not represent the independent MWTP of discount information alone.

Facility information provides spatial information that can be directly used during wayfinding, enabling users to plan their routes around the locations of elevators, escalators, nursing rooms, and other facilities. Previous research has noted that restrooms, elevators, and stores located along secondary corridors may be visually obscured and therefore difficult to locate [5]. In this context, the positive utility of Route + Facility Information suggests that users may have perceived such information as useful for navigating and exploring the shopping mall.

Discount information is not directly related to wayfinding but is closely tied to the purpose of visiting a shopping mall. Ref. [26] also reported that the contextual fit of location-based retail applications influences consumer engagement and intention to use. Accordingly, the positive utility of the condition combining route, facility, and discount information is consistent with the possibility that AR-enabled indoor navigation can integrate information relevant not only to movement but also to the broader shopping context. However, because facility and discount information were introduced cumulatively across the attribute levels, this finding should not be interpreted as the independent effect of discount information alone.

Adding more information does not necessarily lead to greater utility. When the amount of information exceeds users’ processing capacity, identifying relevant information becomes more difficult and decision quality may decline [19]. In the primary model, both information-rich conditions showed positive utility relative to Route Only, although the incremental contrast between the two information-rich levels was sensitive to alternative-position adjustment. Because cognitive load was not directly measured, these findings do not establish whether the tested levels of Information Layering increased or reduced users’ processing burden.

5.3. Experiential Value and Costs of Gamification

The attribute-level results for the Interactive Character Feature and Reward Acquisition Mode differed from those observed for Navigation Mode and Information Layering. Viewed through the framework in Section 2.8, these patterns can be interpreted in relation to the potential experiential benefits and additional attention or action associated with these features beyond wayfinding; however, the study did not directly compare functional and experiential value as constructs.

5.3.1. Conditional Value of the Interactive Character Feature

The Interactive Character Feature was not statistically significant in the full-sample model (β = 0.070, 95% CI [−0.032, 0.173], p = 0.177). The confidence interval includes zero and both negative and positive values, indicating uncertainty about the direction of the average effect within the tested configuration. This nonsignificant result should not be interpreted as evidence that the feature has no consumer value. However, its interaction with context-specific gamification-feature affinity was positive and remained significant after multiplicity correction (γ = 0.239, adjusted p = 0.009). The interaction also remained significant when the character item was excluded from the affinity score. The utility of the bundled interactive-character feature was therefore higher among respondents with greater affinity for the game-like features represented in the measure.

This pattern may reflect responses to the social, interactive, and progression-related elements bundled within the character feature. People may apply interpersonal social rules to media characters [52], and perceiving a character as a social entity may foster social presence [32] and relatedness [33,34]. Because these components were neither separated nor measured, however, social presence and relatedness remain possible explanations rather than identified mechanisms. At the same time, the moderator included an item about character progression, creating some content overlap between the affinity measure and the conjoint attribute. Although the interaction remained significant when this item was excluded, the affinity measure still captured preferences for game-like features in this specific service context. The finding should therefore not be generalized to a broader or stable disposition toward gamification.

5.3.2. Preferences of Reward Acquisition Modes

For Reward Acquisition Mode, Social-Interaction-Based yielded significant negative utility relative to Visit-Based (β = −0.286, conditional MWTP = KRW − 300, p < 0.001). Mission-Based also had a negative coefficient, but its difference from Visit-Based was not statistically significant (β = −0.134, conditional MWTP = KRW − 140, p = 0.072). This indicates lower utility for Social-Interaction-Based relative to Visit-Based, whereas the evidence for Mission-Based was weaker.

Unlike Mission-Based, which requires only individual task performance, Social-Interaction-Based involves cooperation with others and social exposure, potentially introducing additional social costs. Prior research has shown that gamification elements requiring social interaction can generate social overload and user fatigue [35], which may have reduced the appeal of this format in a wayfinding context. Social-Interaction-Based acquisition, however, also required additional effort and depended on the cooperation of other users, so its lower utility reflects these characteristics jointly and cannot be attributed to social overload alone.

In contrast, previous research has found that social influence was positively associated with participation in an exercise gamification service [53]. The findings of the present study suggest that the value of social elements may depend on the purpose and context of service use. In exercise gamification services, interaction with other users can be an integral part of the service experience, whereas in wayfinding, social participation is not directly related to the core task. This contextual difference suggests that the same social feature may be evaluated differently depending on its relevance to the task at hand. A direct contrast also indicated lower utility for Social-Interaction-Based than Mission-Based acquisition (Δβ = −0.152, BH-adjusted p = 0.048), and this contrast remained significant in the alternative-position sensitivity analysis (Table S15). More broadly, these attribute-level patterns are consistent with the possibility that the value of features involving participation beyond wayfinding may vary with feature design. This interpretation remains a proposition derived from the task-based framework rather than a direct test of task–feature fit.

5.4. Preferences Across Reward Types

Compared with Instant Discount Coupon, Reward Points yielded significant negative utility (β = −0.136, conditional MWTP = KRW − 143, p = 0.044), and Prize Draw Entry also yielded significant negative utility (β = −0.412, conditional MWTP = KRW − 432, p < 0.001). Overall, respondents preferred Instant Discount Coupon to both Reward Points and Prize Draw Entry.

Instant Discount Coupon provides a benefit that can be realized immediately after service use, whereas Reward Points require users to accumulate a sufficient balance or return later before realizing their economic value. Prior research has shown that, under low product-category involvement, immediate rewards may be more effective than delayed rewards in fostering loyalty [37]. The delayed realization of Reward Points may have contributed to their lower utility relative to Instant Discount Coupon. Previous research has conceptualized loyalty points as future-oriented rewards and found that consumers with stronger temporal discounting tendencies perceive delayed rewards as less valuable [39]. Accordingly, the relatively lower utility of Reward Points may be related to the time required to realize their value and the additional usage conditions needed to do so; however, because time preference was not measured and no monetary amount was specified for any reward format, this difference may also reflect differences in certainty, redemption effort, or the value respondents inferred for each format.

Prize Draw Entry provides an uncertain benefit because receipt of the actual reward is not guaranteed. Although prior research has shown that uncertain promotions can heighten anticipation [40], Prize Draw Entry yielded lower utility than Instant Discount Coupon in this study. Uncertainty about receiving the reward provides one possible explanation for this lower utility, but Prize Draw Entry also differed from Instant Discount Coupon in its inferred value and in the effort required to obtain a benefit.

Rewards serve a promotional function in phygital services by linking digital benefits with visits to and purchases in physical stores. The findings suggest that reward formats differing in immediacy, certainty, redemption requirements, and inferred value can be evaluated differently by users. A direct contrast showed that Prize Draw Entry was valued below Reward Points (Δβ = −0.276, p < 0.001), although this difference was not robust to adjustment for alternative position (Table S15). Because the two formats differ in several respects at once, however, this difference does not establish that uncertainty is a stronger source of dispreference than delayed reward realization.

5.5. Preference Heterogeneity

Formal interaction tests showed that the overall preference structure varied by both visit frequency and context-specific gamification-feature affinity. The descriptive subgroup coefficients remain useful for interpreting direction, but the pooled interactions provide the inferential basis for heterogeneity claims.

Visit frequency was associated mainly with price sensitivity: respondents who visited more often were less sensitive to the monthly fee in both the continuous (γ = 0.051, adjusted p = 0.010) and log-transformed specifications. In the log-transformed model, more frequent visitors also evaluated Reward Points less favorably; however, because this interaction did not survive correction in the continuous specification (adjusted p = 0.073), it should be regarded as tentative. The other visit-frequency differences, including the Mission-Based pattern in the median-split model and the larger functional-attribute coefficients in the lower-frequency subgroup, were not supported across specifications. Visit frequency should therefore be interpreted as a proxy for repeated mall exposure rather than direct evidence of spatial familiarity.

In the primary continuous model, context-specific gamification-feature affinity was associated with two preference differences: higher utility for the Interactive Character Feature and lower sensitivity to the monthly fee. None of the other seven interactions survived multiplicity correction, so the information-layering pattern in the median-split models is treated as descriptive. Interpretation of the character interaction remains limited by the content overlap between the affinity measure and the character attribute.

Gender and age did not show significant joint moderation, and no individual interaction survived multiplicity correction. The observed heterogeneity associated with visit frequency and feature affinity suggests that these variables may warrant further investigation in segmentation research; however, the present study did not compare the predictive performance of alternative segmentation strategies.

5.6. Theoretical and Managerial Implications

5.6.1. Theoretical Implications

These findings have theoretical, empirical, methodological, and contextual implications. Theoretically, the study clarifies how the value of a phygital touchpoint should be understood. Existing phygital frameworks describe the dimensions of integrated physical and digital experiences and conceptualize such integration as a source of customer value [14,17]. Viewed through the four value categories and the task-based logic developed in Section 2.8, the present results suggest that digital augmentation does not contribute to consumer utility uniformly.

At the attribute level, Map + AR Guidance and both information-rich conditions showed positive utility relative to their reference levels, whereas the Interactive Character Feature showed no significant average effect and Social-Interaction-Based reward acquisition showed lower utility than Visit-Based acquisition. Interpreted through the task-based perspective, this pattern is consistent with the possibility that how a digital feature relates to the focal consumer task may matter for its valuation. However, the study did not manipulate functional versus experiential value as constructs or directly test task–feature fit as a mechanism. This task–feature relationship, which is central to the task–technology fit perspective [44], is less explicit in existing phygital frameworks and therefore provides a proposition for extending the interpretation of consumer value in phygital retail rather than a mechanism established by the present experiment. Because the present evidence is based on stated preferences for a proposed service rather than on observed use, this proposition should be tested in actual phygital environments.

Empirically, the study provides attribute-level evidence on how consumers trade off navigation, information, gamified, promotional, and price features within complete AR-enabled indoor-navigation service configurations. Rather than treating digital augmentation as a single service characteristic, the results show that different forms of augmentation can contribute differently to stated choice utility.

From a methodological perspective, the conjoint approach complements studies that evaluate digital features individually by estimating attribute-level trade-offs within complete service configurations.

Contextually, the study extends evidence on phygital retail preferences to AR-enabled indoor navigation in large, multi-level mixed-use malls within a market characterized by highly developed mobile payment and loyalty programs. This setting provides evidence for a service category for which an established market benchmark is not yet available. At the same time, the evidence is limited to a single country, a single online panel, and adults aged 20–49 and should therefore not be generalized beyond this context without further validation.

5.6.2. Managerial Implications

For shopping mall operators and AR service developers, the observed preference estimates suggest a staged approach to feature development and testing. AR guidance and functional spatial information represent candidate priorities for initial development and validation. Navigation Mode showed the highest range-based relative importance among the non-price attributes, and Map + AR Guidance yielded the highest conditional MWTP among the non-price attribute levels. Route + Facility Information also showed positive utility relative to Route Only (conditional MWTP = KRW 336). Adding discount information further increased utility in the primary model, but this incremental contrast was sensitive to position adjustment. Additional contextual or commercial information could therefore be tested incrementally to determine whether it adds value without interfering with the wayfinding task.

Once the navigation and information functions have been validated, gamified elements could be tested as optional additions rather than assumed to be default components of the service. The Interactive Character Feature showed no statistically significant average effect but was valued more positively by respondents with higher context-specific gamification-feature affinity. A practical implementation could therefore allow users to enable or disable the interactive character. Social-Interaction-Based acquisition should likewise be optional rather than a default requirement, and whether brief Mission-Based tasks can be offered without interfering with wayfinding is a question for usability testing.

Reward design should consider immediacy and certainty. Instant Discount Coupon was preferred over both Reward Points and Prize Draw Entry. Accordingly, services may place greater emphasis on rewards that users can recognize and use immediately. When Reward Points are offered, accumulation requirements, redemption conditions, and expiration periods should be clearly communicated. For Reward Points and Prize Draw Entry, future service tests should separately manipulate reward value, timing, certainty, and redemption requirements to determine which characteristics drive user preferences.

Within the tested fee range, Price accounted for the largest share of range-based relative importance (RI = 52.2%), with choice utility decreasing by 0.953 for each KRW 1000 increase in the monthly fee (p < 0.001). Because this share depends on the width of the tested fee range, it should not be treated as a general measure of price sensitivity. The MWTP estimates reported in this study represent conditional, relative monetary valuations across attribute levels within the experimental choice sets rather than direct estimates of consumers’ actual willingness to subscribe or market prices. Therefore, specific pricing levels and subscription decisions should be evaluated through subsequent field validation and pricing experiments that include opt-out and free-service alternatives.

5.7. Limitations and Future Research

This study has several limitations, which we group into four areas.

First, the findings are stated preferences for a proposed service. Participants evaluated written and visual descriptions of service profiles in an online survey rather than using AR navigation in a shopping mall, so the study does not observe the interaction between physical movement and digital information that defines a phygital experience. Choices were made among three paid alternatives, with no opt-out, free-service, or existing-navigation option, and the realism of the price levels was not separately validated; the MWTP estimates are therefore conditional, relative valuations within the experimental choice sets rather than estimates of actual willingness to subscribe or of market prices. Stated values also tend to exceed those observed when payment is real [54], and alternative revenue models, such as freemium or retailer-funded provision, were not compared. Accordingly, the observed price sensitivity may partly reflect the absence of free or alternatively funded options in the choice sets, rather than solely respondents’ valuation of differences within the tested fee range. Field experiments with behavioral measures, such as movement paths, task completion times, and promotion use, and pricing studies incorporating actual payment decisions and free alternatives are needed to establish whether these preferences translate into actual use, adoption, promotional engagement, and payment.

Second, the experimental design constrains some inferences. The L24 plan estimated main effects only, so theoretically relevant interactions among attributes, such as Interactive Character Feature × Reward Type, were not modeled, and the choice sets were formed from a profile-level orthogonal plan rather than generated with a choice-design efficiency criterion. In addition, task sequence and alternative position were not randomized, so the design cannot exclude response biases arising from task order, such as learning or fatigue, or from alternative position. In an exploratory analysis, the task-specific first-choice hit rate was not significantly associated with task order across the eight estimation tasks (Spearman ρ = −0.071, p = 0.867; Section B.4 of the Supplementary Materials). Because all respondents received the tasks in the same order, however, this check cannot rule out learning or fatigue effects. A significant third-position effect was also observed, and two pairwise contrasts were sensitive to position adjustment (Table S15). Because alternative position was partly associated with Information Layering in the fixed design, position and attribute effects cannot be fully disentangled. These design features therefore represent a threat to internal validity, particularly for the affected pairwise comparisons. Several attributes were also implemented as bundles: the character condition combined presence, interaction, nurturing, and progression; the reward formats differed in timing, certainty, redemption requirements, and inferred value; and the reward-acquisition modes differed in effort, social exposure, and dependence on others. Their coefficients therefore reflect preferences for bundled formats rather than isolated mechanisms. Finally, the attribute set was not subjected to a formal content-validation procedure, and the comprehension checks allowed unlimited attempts that were not recorded. Future choice experiments should randomize both task sequence and alternative placement and use efficient designs that permit theoretically relevant interactions to be estimated. Studies seeking to isolate specific feature mechanisms should vary individual design components while controlling other reward characteristics, including expected monetary value, timing, certainty, and redemption requirements. Future studies should also cap and record comprehension-check attempts.

Third, the sample limits generalization. Respondents were adults aged 20–49 recruited from a single online panel in one country, so the findings should not be generalized to adults aged 50 and above, to consumers with lower digital familiarity, or to markets with different mobile payment and loyalty-program norms. The study therefore cannot establish whether the observed preference patterns also hold among consumers aged 50 and above; the nonsignificant age-group moderation within the sampled 20–49 age range does not establish that these patterns apply to older consumers. Likewise, the single-country sample does not establish whether the relative valuations observed in South Korea hold in other national retail markets. No prospective power analysis was conducted for the moderation analyses, and the smallest descriptive subgroup contained 131 respondents. Replication across age groups, countries, and recruitment modes is needed.

Fourth, the analysis of preference heterogeneity is exploratory. The gamification-feature-affinity measure was researcher-developed and includes items that overlap in content with the conjoint attributes. Although the character interaction remained significant when the character-nurturing item was excluded from the measure, the remaining items still assess affinity for game-like features in this specific service context. The interaction should therefore not be interpreted as evidence of a general disposition toward gamification. The mixed logit results suggest heterogeneity in several coefficients, but part of this variation may reflect differences in choice consistency across respondents, which the present specification cannot separate. Future research should use independently validated gamification-preference measures and preregistered moderation models, and should examine preference and scale heterogeneity using approaches such as latent-class models for discrete preference heterogeneity and scale-adjusted models for differences in choice consistency.

6. Conclusions

Within the tested attribute levels and fee range, price accounted for the largest share of range-based relative importance, while AR guidance and information-rich navigation configurations showed positive stated-choice utility among respondents aged 20–49. The Interactive Character Feature showed no significant average effect, while preferences varied across reward and reward-acquisition formats.

These results support a layered design approach in which navigation and information features serve as candidate core components, while gamified features are treated as optional elements for further testing. Because the monetary estimates are conditional valuations from a forced-choice experiment rather than market prices, specific price points should be evaluated through field studies and pricing experiments that include opt-out and free-service alternatives.