Measuring Customer Experience in AI Contexts: A Scale Development

Abstract

With the advent of the digital intelligence era and the rapid evolution of emerging technologies, Artificial Intelligence (AI) is fundamentally transforming the way consumers and businesses interact, gradually becoming one of the primary tools for companies to continuously improve customer experience and maintain competitiveness. However, existing research on customer experience largely overlooked the disruptive changes brought by the widely applied AI technologies. Therefore, this paper focuses on customer AI experience in the new context, using a mixed research method combining qualitative and quantitative approaches to explore the connotation, measurement, formation mechanism, and related action mechanisms of this construct. This study finds the following: (1) the customer AI experience is an intrinsic and subjective response generated by customers after interacting with AI capabilities, mediated by AI. It specifically includes five dimensions: social experience, intellectual experience, classification experience, exploitation experience, and service experience; (2) its formation and development is a cyclical model comprising three stages: expectation, realization, and reflection, corresponding to the mechanisms of contact, interaction, and comparison; (3) the perceived innovative characteristics of AI technology help customers to have a better AI experience, thereby stimulating customer engagement behavior. This provides certain guidance and reference for enterprises to better understand and utilize AI’s innovative characteristics to improve the customer experience, promote customer engagement, seize opportunities in AI technology development, and maintain a competitive advantage.

1. Introduction

The advancement of digital technology has shifted the traditional business model into an interactive digital ecosystem [1,2]. Enterprises and customers increasingly rely on digital interaction platforms—comprising related artifacts (both physical and digital, including data in the form of text, images, audio, and video). The interactive marketing becomes a new normal in the digital era [2] so that people (including customers, employees, partners, and stakeholders), processes (enhanced by an increasing number of enabled software such as algorithms), and interfaces (both physical and digital) create value through interaction within an evolving digital network system [3,4,5]. The rapid development and widespread adoption of artificial intelligence have led to the emergence of B2C commerce platforms (such as Taobao and JD.com) that combine AI digital technology with digital interaction platforms [6,7]. This evolution not only enhances the efficiency of e-commerce operations but also continuously transforms customer interaction experiences [8].

AI technology is particularly effective in providing personalized services and recommend tailored products by analyzing customers’ past purchase experiences and preferences [9]. For instance, Huawei’s fitness bands, NetEase Cloud’s personalized FM (the function of intelligently recommending songs based on users’ listening habits), and Toutiao’s slogan “What you care about is the headline” demonstrate how brands can effectively generate personalized styles and product suggestions based on different needs and preferences. Many brands focused on customer experience strategically deploy various AI technologies at key customer touchpoints. As stimuli related to brands, AI can evoke different experiential dimensions, thereby creating new experiential value [10]. Current research on customer experience in the AI context often draws from the connotations of brand experience, categorizing it into traditional experiential dimensions such as cognitive, behavioral, sensory, and hedonic [11]. Although Puntoni et al. [12] identified customer AI experience as the experience generated after customers interact with AI functionalities in online shopping contexts, specifically including the data capture experience, classification experience, delegation experience, and social experience, the study also highlighted the possibility of “unknown experiences”.

In summary, although previous studies have preliminarily explored the connotation and structure of the customer AI experience, there are still three main research gaps: (1) the dimensions of the customer AI experience remain ambiguous and uncertain; this indirectly leads to (2) an unclear process in the formation of the customer AI experience; and limits the exploration of (3) the factors influencing the customer AI experience and their related outcomes. These gaps significantly hinder the further development of the customer experience in the context of emerging technologies, leading to an incomplete and inaccurate understanding of the complex phenomenon of the customer AI experience (CAE). This lack of clarity makes it difficult to provide targeted and effective recommendations for improving the customer AI experience to digital interaction platform owners, thus impeding the integration and development of AI technologies and digital interaction platforms.

Therefore, this paper will focus on B2C e-commerce platforms within digital interaction platforms, selecting two service scenarios where customers most frequently engage and are most perceptive of AI services: AI-based personalized recommendations and AI-powered intelligent customer service. This study aims to clarify the structure and measurement tools of the customer AI experience (Study 1), explore the dynamic mechanisms underlying its formation (Study 2), and conduct preliminary empirical tests of its rule network through the construction of a theoretical model (Study 3). The goal is to enrich the theoretical research on the customer AI experience, provide service improvement recommendations for AI applications on B2C platforms, and ultimately deliver a better AI experience for customers on digital interaction platforms.

2. Literature Review and Theoretical Background

2.1. Customer AI Experience

2.1.1. The Connotation and Measurement of the Customer AI Experience

The customer AI experience stems from the customer experience, which refers to the multidimensional response of customers to a company’s services, encompassing five dimensions: cognitive, emotional, sensory, behavioral, and relational [13]. Research on customer experience phenomena can be broadly classified into two categories: the customer experience as a response to the consumption process and the customer experience as a response to managerial stimuli [14]. One approach describes the interaction between customers and companies as a journey of customer experience touchpoints, while the other distinguishes between different types of customer reactions [15]. With the integration of AI technology, companies can enhance the customer experience by gaining a deeper understanding of customers and predicting their needs, analyzing customer emotions and feedback with scale, precision, and speed that are beyond human capabilities [16]. The intelligent customer experience, therefore, refers to a customer experience mediated by technology, specifically manifested in customers’ perception of AI’s advantages and the extent of interaction [17].

The customer AI experience refers to the reactions customers have after interacting with AI functionalities, encompassing four main capabilities: listening, predicting, generating, and interacting [12]. The listening capability enables AI systems to gather data about consumers and their environments. AI also predicts customer behavior based on algorithms applied to collected data. Generating refers to AI’s ability to perform tasks that customers could typically handle themselves, given authorization. Interacting denotes AI’s capacity to engage in interactive communication. The customer AI experience specifically includes the data capture experience, classification experience, delegation experience, and social experience [12]. The listening capability allows AI systems to collect data about customers and their environments, resulting in a “data capture experience”. In the “classification experience”, customers perceive AI-supported personalized predictions as outcomes classified for specific consumer types. The “delegation experience” involves customers using AI solutions during production to perform tasks they could normally handle themselves. Lastly, the “social experience” encompasses the interaction and communication between customers and AI. Therefore, the customer AI experience refers to the intrinsic, subjective responses customers have after interacting with AI functionalities (such as intelligent chatbots) as intermediaries [18]. Previous research concepts and dimensions related to the customer AI experience are summarized in

Table 1. Review and dimensional analysis of prior research on customer AI experience.

Core Concept	Specific Dimensions	Source
AI-supported customer experience	Hedonic experience (memorable, entertaining, thrilling, and comfort) and cognitive experience (respectable, popular, secure, and aesthetic)	Ameen et al. (2020)
Intelligent experience co-creation	Hedonic experience, cognitive experience, social and personal experience, and pragmatic economic experience	Roy et al. (2019)
Technology service experience	Reliability, assurance, empathy, and responsiveness	Prentice and Nguyen (2020)
Online customer experience	Informativeness (cognitive), entertainment (emotional), sociality, and sensory appeal	Bleier et al. (2019)
Customer AI experience	Data capture experience, classification experience, delegation experience, and social experience	Puntoni et al. (2021)
	Data capture experience, classification experience, authorization experience, social experience, and anthropomorphic experience	Wang et al. (2024)
Intelligent customer experience	Relative advantage and perceived interactivity	Ren Lina (2021)

.

Regarding the measurement scale of the customer AI experience, most current studies directly adopt traditional brand experience scales, which primarily include five dimensions: cognitive experience, hedonic experience, sensory experience, behavioral experience, and social experience, or employee service experience scales. Some studies have also considered the customer experience in the AI context, as shown in Table 1. However, with the continuous development of AI technology, the customer AI experience has evolved into new forms. In particular, the customer AI experience based on digital interaction platforms, which integrates the characteristics of the online customer experience, requires further in-depth exploration.

2.1.2. Formation Mechanism of Customer AI Experience

Current research on the formation mechanisms of the customer experience, both domestically and internationally, primarily focuses on aspects such as the customer decision-making process, customer expectations, and the cycle of expectation, realization, and reflection. Schmitt (1999) argued that the formation of the customer experience follows a series of stages, including “initial awareness, understanding, attitude formation, and final purchase” [13]. Meanwhile, the model of user experience formation for e-commerce shopping websites consists of four steps: characteristic perception, characteristic construction, evaluation, and outcome formation [19]. De Keyser et al. [20] emphasized that the formation of the customer experience should consider the concepts of customer immersion and customer value, proposing that the process is cyclical and includes three stages: the expectation stage, the realization stage, and the reflection stage. The expectation stage involves the anticipated customer experience and value estimation; the realization stage includes customer immersion and the generation of specific experiences; and the reflection stage focuses on customer value judgments, where the experience value, influenced by memory and encoding (explicit and implicit learning), impacts future experiences. Schallehn et al. [21] further refined this cyclical process, suggesting that the customer experience could include past, present, or future real or imagined experiences and could occur during interactions between customers, service providers, other customers, and/or other participants, with an emphasis on the resulting relational outcomes. Additionally, some scholars highlight the presence of both the expectation and interaction stages, where customers form expectations before interacting with the service or product, and during the interaction, they make judgments and compare these judgments to their expectations, thus completing the customer experience [22].

However, the formation mechanisms of the customer AI experience have not been deeply explored in existing research, with only a few scholars attempting to address this issue. For example, Wang Shuting [23] found that consumer experience with AI products is influenced by factors related to the AI products themselves, consumer characteristics, and the interaction between the product and the consumer. However, this study did not fully investigate the dynamic process of customer AI experience formation. Therefore, from a comprehensive perspective, this paper argues that the study of the formation mechanisms of the customer AI experience should address multiple stages and adopt a dynamic perspective to explore the repetitive and evolving nature of the process, thereby rethinking the formation process and mechanisms of the customer AI experience.

2.2. Technology Acceptance Model and Innovation Diffusion Theory

2.2.1. Technology Acceptance Model

The Technology Acceptance Model (TAM), proposed by Davis in 1986, applies the Theory of Reasoned Action (TRA) from social psychology to explain and predict people’s acceptance of information technology. It incorporates factors such as intrinsic beliefs, subjective attitudes, behavioral intentions, and external variables. Initially, the model was used to explain employees’ acceptance behaviors and their willingness to use information systems technology in traditional organizational contexts [24]. In the classic model, Davis introduced the concepts of perceived usefulness and perceived ease of use and subsequently developed corresponding measurement scales. Since the introduction of TAM, scholars have primarily focused on identifying additional key factors, beyond perceived usefulness and perceived ease of use, to enhance the model’s adaptability [25]. For example, Rese et al. (2020) [26] extended the classic model by incorporating entertainment as a factor, which effectively predicted customers’ acceptance intentions toward chatbots. This is particularly relevant considering that, in the context of online shopping, hedonic factors (such as enjoyment) may exert a stronger influence than information technology factors [27]. This implies that the classic TAM model is insufficient in explaining customer behavior in online shopping contexts [28]. Therefore, this paper argues that the TAM model should incorporate entertainment as an additional factor and consider it as a precursor that influences customer AI experience.

2.2.2. Innovation Diffusion Theory

Innovation Diffusion Theory (IDT) posits that innovation diffusion is the process by which an innovation is communicated through certain channels within a social system and adopted at varying speeds by its members. Rogers and Shoemaker (1971) [29] found that the characteristics of an innovation itself influence the rate at which people adopt new technologies. These characteristics include five key attributes: relative advantage, complexity, compatibility, observability, and trialability. Subsequently, Moore and Benbasat (1991) [30] improved upon this framework by developing a structure for studying individual technology acceptance, introducing additional attributes such as visibility and result demonstrability to complement the original innovation characteristics. Given that this paper focuses on customers as individual subjects, it selects compatibility, visibility, and result demonstrability as specific perceived characteristics of AI innovations to measure, incorporating them as external influencing variables into the final research model within the Technology Acceptance Model (TAM). This approach helps to assess how customers perceive AI innovation features, further contributing to understanding AI adoption in customer experiences.

2.3. Customer Engagement

Customer engagement, as the “twin brother” of customer experience, has been extensively studied recently in the interactive marketing literature [31,32,33,34]. For instance, the relationship between customer engagement and customer value forms a natural alliance within customer experience, collectively constituting the trinity of marketing [20]. Digital customer engagement further increases consumer participation in various social media platforms with a significant impact on marketing strategic implications [35,36]. Various conceptualizations and measurements of customer engagement can be categorized into four main dimensions: psychological states, customer behaviors, company value-add mechanisms, and intrinsic motivations. Customer engagement behavior refers to actions taken by customers beyond the purchase, driven by motivation towards a brand or company, such as word-of-mouth activities, recommendations, assisting other customers, blogging, writing reviews, and engaging in legal actions. A positive artificial intelligence (AI) service experience motivates customers to integrate more physically, mentally, socially, and emotionally with the company [37]. Previous research has shown that customer engagement behaviors can bring significant benefits to companies, such as improved company performance, support for new product development, promotion of service innovation, and intangible benefits like attracting more customers [38,39]. This paper incorporates both customer AI experience and customer engagement behaviors into the Technology Acceptance Model because, in the context of e-commerce platforms, the customer AI experience serves as a response to the use of AI services and can effectively measure usage attitude, which is a key component in the classic TAM model.

3. Study 1: Development and Validation of Customer AI Experience Scale

The main objective of Study 1 is to clarify the dimensions of the customer AI experience and provide corresponding measurement tools. First, a qualitative research approach using grounded theory was adopted. Through semi-structured interviews with 17 e-commerce customers and systematic coding of the interview content, 34 key concepts of the customer AI experience were identified. Second, based on these key concepts, the development of a customer AI experience measurement scale was carried out in three steps: constructing an initial pool of measurement items, purifying the scale, and conducting further validity and reliability testing. This process resulted in the development of a customer AI experience measurement scale, consisting of five dimensions and 14 items, ensuring an accurate assessment of customer AI experience.

3.1. Initial Construction of Customer AI Experience Scale

3.1.1. Interview Study

The initial items for the customer AI experience (CAE) were primarily derived through a literature review and engagement with customer interview methods. This study conducted semi-structured interviews with 17 e-commerce customers based on a formulated interview outline. The specific interview content is detailed in

Table 2. Interview outline.

Interview Topic	Questions	Purpose
Introduction via Shopping Apps	Do you usually use shopping apps (such as Taobao, JD.com, Pinduoduo, VIPshop, etc.) for online shopping? How often do you use them per week? What is your age as a user?	To inquire about the use of shopping apps by the interviewee
Customer AI Experience Dimension (Personalized Recommendations Context)	2. Besides actively searching for products, do you browse through the app’s smart recommended products? How often per week and how much time on average do you spend on this? Examples of smart recommendations include: Featured For You, Recommended for You, Discover Great Deals, JD Flash Sales, Free Shipping Thresholds, etc. 3. How do you find the service experience of its smart recommended products? What do you think it brings to you, and what is your overall evaluation? Can you provide 1–2 examples to illustrate your experience with these smart recommended products?	To understand the customer AI experience in the context of personalized recommendations: whether there is and what it entails
Customer AI Experience Dimension (Intelligent Customer Service Context)	4. Have you encountered intelligent customer service while using shopping apps for online shopping? Can you distinguish between human customer service and AI-powered customer service? How was your experience with them? Please provide examples. 5. How do you find the experience of using intelligent customer service? What do you think it brings to you, and what is your overall evaluation? 6. Can you give a few examples to elaborate on your experience with intelligent customer service?	To understand the customer AI experience in the context of intelligent customer service: whether it exists and what it involves
Basic Information	7. Age, marital status, education level, occupation, annual household income	Gather basic demographic information about the interviewee

.

To ensure sample diversity (including different age groups and professions) and to capture the background of AI usage among respondents (including usage frequency and duration), one-on-one semi-structured in-depth interviews were conducted with 9 industry professionals (Samples 1–9) and 8 academic experts (Samples 10–17). This approach effectively gathered data from a total of 17 interview samples. The research reserved Samples 8, 9, and 17 for saturation testing, and no new content emerged from these samples, indicating that the theoretical saturation principle had been met. As a result, sampling was concluded.

The interviews were primarily conducted through WeChat messaging or face-to-face interviews. The interviewees ranged in age from 19 to 48 years old, with 11 female and 6 male participants. They represented various fields including students (undergraduate, master’s, and doctoral students), farmers, teachers, regular employees, and company executives. Upon obtaining consent from the interviewees, screenshots of the chat records or recordings were taken. Immediately after the interviews, the raw data were transcribed into text format, resulting in nearly 20,000 words of textual data.

3.1.2. Coding Process

(1)

Open Coding

The researcher employs line-by-line, sentence-by-sentence, and paragraph-by-paragraph coding to extract relevant constructs from the interview content. During the open coding process, the research aims to restore the interviewees’ original expressions as much as possible, using their words and phrases for coding. This maximizes the representation of digital leadership characteristics and ensures consistency in the meaning of each coding item. Through the open coding process, this study extracted a total of 198 initial statements from the sampled data.

(2)

Selective Coding

Selective coding involves purposefully choosing items to extract subcategories and main categories. In this study, the concepts extracted during open coding were selected, merged, and classified, resulting in 12 subcategories highly relevant to consumer AI experiences, which were ultimately grouped into five main categories, as shown in

Table 3. Selective coding results.

Main Categories	Subcategories	Examples of Corresponding Initial Open Coding Concepts
Social Experience	Problem Solving	8-4 Able to quickly propose solutions to problems 14-6 Helps consumers quickly resolve issues
	Communication	1-9 Smart customer service is polite, patient, and effective in communication 11-11 Responses are timely
	Humanized Communication	1-1 The experience with smart recommended products is user-friendly 11-6 Recommendations are really user-friendly
Service Experience	Accurate Prediction of Preferences	1-2 Products recommended after big data analysis match personal needs 11-2 Meets preferences and focus points
	Convenience and Time-Saving	1-3 It brings great convenience 1-4 Saves more time 10-2 Saves search time
	Provides Multiple Choices	5-11 Recommendations are diverse 5-3 Allows comparison and more choices
Intellectual Experience	Enhancing Consumer Capability	2-6 Use personalized recommendation patterns to buy more cost-effective products 6-5 Commission smart customer service for product prediction
	Triggering Exploration and Thinking	2-7 Consciously stay on the recommendation page 4-4 Search for other items on a whim
Exploitation Experience	Using Consumer Information	3-8 Collect consumer data for recommendations 12-4 Collect personal information through purchase records
	Intruding Consumer Privacy	5-7 Exposing consumer privacy 12-3 Feeling life is being monitored 12-6 Feeling eavesdropped
Classification Experience	Accurate Prediction of Consumption Ability	10-7 Recommended products are similar to what I usually buy 11-5 Recommended products are reasonably priced
	Recommended Products Align with Self-Identity	1-6 The types of recommended products match my preferences 10-8 Recommended products match my style

.

Finally, this study constructs a structural model of the consumer AI experience, with “social experience”, “service experience”, “intellectual experience”, “exploitation experience”, and “Classification Experience” as the main categories, as shown in Figure 1.

3.1.3. Item Generation

First, based on the qualitative analysis results discussed above and in conjunction with certain items from the existing literature, an initial pool of 26 items was developed. This included two items from the intellectual experience dimension of brand experience, which were modified to fit the context of AI personalization. These items are: “When I use AI personalized recommendations, I think a lot” and “AI personalized recommendations stimulate my curiosity and enhance my problem-solving abilities” [40]. Additionally, the items “Using AI personalized recommendations makes me want to study and explore them” and “I believe the personalized products recommended by AI align with my identity” were adapted from the thinking and relational experience dimensions of the brand experience scale developed by Zhang and Bian [41]. Apart from these 4 items, the remaining 22 items were grounded in the interview text and revised multiple times by the authors, as shown in

Table 4. Partial initial items.

I can perceive that AI personalized recommendations serve me	14. I think AI recommends personalized products that fit my identity
2. I believe AI personalized recommendations provide me with time-saving and convenient services	15. When using AI personalized recommendations, I tend to think deeply
3. I think AI personalized recommendation services give me various choices	16. AI personalized recommendations stimulate my curiosity and enhance problem-solving abilities
4. I see AI personalized recommendations as a win-win service for me and businesses	17. Using AI personalized recommendations makes me want to research and explore them
5. I believe AI personalized recommendations stimulate my consumption	18. Through long-term use and reflection, I grasp the patterns of AI personalized recommendations and utilize them
6. I can perceive that AI personalized recommendations utilize my information	19. I would delegate tasks to AI customer service that I could have done myself
7. I believe AI personalized recommendations pry into my information	20. I delegate some tasks to AI customer service, but AI cannot make decisions autonomously when performing tasks (e.g., querying product information)
8. AI personalized recommendations monitor my conversations	21. I delegate some tasks to AI customer service, and AI has some autonomy in decision-making when performing tasks (e.g., providing choices for bulk orders)
9. I feel that AI personalized recommendations protect my privacy when collecting and using my information	22. I find communication with AI customer service relatively smooth
10. I think AI personalized recommendations generally understand and position me correctly	23. I believe AI customer service can solve simple problems
11. I believe AI personalized recommendations can accurately predict my consumption preferences	24. I believe AI customer service can solve complex problems
12. I believe AI personalized recommendations can accurately predict my spending capacity	25. I think AI customer service has good humanization
13. I believe AI personalized recommendations can accurately predict my external characteristics (age, size, etc.)	26. I find AI customer service reassuring in handling tasks

.

Subsequently, the 26 items were repeatedly reviewed by three management researchers to ensure clarity, consistency, and readability. Finally, we invited three doctoral students, three master’s students, three professionals, and one senior business consultant to serve as evaluators. They were provided with a form containing all 26 measurement items, with a blank space next to each item where evaluators could assess the content validity of each item. The results showed that all 10 evaluators correctly interpreted the 26 items, meeting the minimum standard of 75%. This preliminary validation ensured that the scale had good content validity, and the research then proceeded to the data collection phase, using a 7-point Likert scale for the questionnaire.

3.2. Data Analysis

3.2.1. Data Collection

Based on the 26 initial items developed above, a questionnaire survey was conducted, primarily through an online distribution method via the Credemo platform. In terms of sample selection, a screening process was first implemented to ensure that all participants had some experience with artificial intelligence. The survey followed a voluntary participation approach, which may introduce self-selection bias, meaning that only individuals with an interest in or experience with AI technology participated. Therefore, the sample may lack representativeness, particularly as it likely overrepresents individuals with higher education and greater familiarity with technology. The questionnaire was created, modified, and collected over a period of nearly two months, from February 2023 to April 2023. A total of 500 valid responses were obtained, with invalid or duplicate responses excluded during the data collection process to ensure the representativeness of the final sample.

Subsequently, following the scale development procedures, the 500 responses were divided into two samples. Sample 1 consists of 230 responses, which were used for exploratory factor analysis to preliminarily assess the scale’s reliability and validity. Sample 2 includes 270 responses, which were used for confirmatory factor analysis to further validate the scale’s reliability and validity, ultimately completing the scale development process.

3.2.2. Exploratory Factor Analysis

This study conducted exploratory factor analysis on 230 randomly selected questionnaires using SPSS (version 26.0). The basic characteristics of the sample of 230 questionnaires are shown in

Table 5. Sample basic characteristics (N = 230).

Demographic Variable	Variable Value	Frequency	Percentage
Gender	Male	107	46.5%
	Female	123	53.5%
Age	0~20	6	2.6%
	21~30	130	56.5%
	31~40	89	38.7%
	41~50	5	2.2%
Occupation	Student	40	17.4%
	Ordinary employee	112	48.7%
	Middle and senior management	74	32.2%
	Other	4	1.7%
Education	Bachelor’s degree	178	77.4%
	Master’s degree	29	12.6%
	Doctorate	1	0.4%
	High school/Technical school and below	3	1.3%
	Associate degree	19	8.3%

below.

First, before conducting exploratory factor analysis (EFA), discriminant validity of the items was assessed. Five items were deleted due to corrected item-total correlations (CITC) less than 0.4 (CAE 4, CAE 5, CAE 10, CAE 13, and CAE 15). The remaining 21 items had CITC values all above 0.5, indicating good discriminant validity and high internal consistency. Next, EFA was performed on the remaining 21 items. The results of the exploratory factor analysis showed a Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy of 0.925 (>0.8) and Bartlett’s test of sphericity with a p-value < 0.001, confirming suitability for factor analysis. Initially, four factors with eigenvalues above 1 were extracted, but they only accounted for 54.6% of the total variance, insufficient for adequate item explanation. Consequently, five factors were extracted, which accounted for a cumulative variance of 66.059%. Items were retained if their factor loadings after varimax rotation were greater than 0.5 and did not cross-load with other factors by more than 0.4 (CAE 1, CAE 9, CAE 20, and CAE 23). Finally, 17 items remained across 5 factors.

Table 6. Exploratory factor analysis results.

Original Code	Item	Component
		1	2	3	4	5
CAE24	I believe AI chatbots can handle complex problems	0.729
CAE22	I find communication with AI chatbots smooth	0.719
CAE25	I think AI chatbots are personable	0.707
CAE26	I feel confident in the effectiveness of AI chatbots	0.668
CAE11	AI recommendations accurately predict my consumption preferences		0.654
CAE2	AI recommendations have saved me time and effort		0.635
CAE3	AI recommendations have given me a variety of choices		0.593
CAE17	Using AI recommendations makes me want to explore further			0.691
CAE16	AI recommendations stimulate my curiosity and problem-solving abilities			0.690
CAE18	Through long-term use and reflection, I understand AI recommendation patterns and utilize them			0.640
CAE21	I delegate tasks to AI chatbots, which have some autonomy in decision-making (e.g., suggesting bundle options)			0.639
CAE19	I delegate tasks to AI chatbots that I could have done myself			0.507
CAE7	I feel that AI recommendations invade my privacy				0.826
CAE6	I perceive AI recommendations as exploiting my information				0.789
CAE8	AI recommendations listen to my conversations				0.759
CAE12	AI recommendations accurately predict my spending power					0.760
CAE14	I think AI recommendations of personalized products align well with my identity					0.718
Explained Cumulative Variance (%):		15.625	29.359	42.119	54.578	66.059

presents the results of the exploratory factor analysis of the Customer AI Experience Scale.

Finally, these five factors were named as “social experience (4 items)”, “service experience (3 items)”, “intellectual experience (5 items)”, “exploitation experience (3 items)”, and “classification experience (2 items)”.

“Social experience” refers to the experience of communication and interaction with an AI partner, dependent on the AI’s communication abilities, corresponding to social experience. It includes 4 items (CAE 22, CAE 24, CAE 25, and CAE 26). “Service experience” refers to the experience of being served by AI after providing personal data, relying on AI’s listening abilities, corresponding to the service experience in data capture. It includes 3 items (CAE 2, CAE 3, and CAE 11). “Intellectual experience” refers to the thinking experience generated through interaction with AI, based on the customer’s learning abilities. It includes 5 items (CAE 16, CAE 17, CAE 18, CAE 19, and CAE 21). “Exploitation experience” refers to the experience of being exploited by AI after providing personal data, relying on AI’s listening abilities, corresponding to the exploited experience in data capture. It includes 3 items (CAE 6, CAE 7, and CAE 8). “Classification experience” refers to the experience of receiving AI’s personalized predictions, dependent on AI’s predictive abilities, corresponding to categorization experience. It includes 2 items (CAE 12 and CAE 14).

It is worth noting that the delegation experience did not emerge in the context of a large-sample questionnaire survey. This study posits that this is because previous research was based on scenarios involving social media, online shopping, and personal virtual assistants [12], whereas our study did not focus on the scenario of personal virtual assistants. Therefore, in the context of a large-sample questionnaire survey, the delegation experience did not show significant prominence.

3.2.3. Confirmatory Factor Analysis

This study employed SPSS (version 26.0) and AMOS (version 24.0) tools to conduct confirmatory factor analysis on a sample of 270 questionnaires. Since the exploratory and confirmatory factor analyses were conducted on two subsamples from the beginning and end of the data collection process, the non-random sampling may lead to differences in sample characteristics, potentially affecting the stability and generalizability of the results. The basic characteristics of the sample are presented in

Table 7. Sample basic characteristics (N = 270).

Demographic Variable	Variable Value	Frequency	Percentage (%)
Gender	Male	95	35.2
	Female	175	64.8
Age	0~20	6	2.2
	21~30	136	50.4
	31~40	113	41.9
	41~50	11	4
	51~60	4	1.5
Occupation	Student	29	10.7
	Ordinary employee	137	50.7
	Middle and senior management	95	35.2
	Other	9	3.4
Education	Bachelor’s degree	188	69.6
	Master’s degree	39	14.4
	Doctorate	3	1.1
	High school/Technical school and below	12	4.5
	Associate degree	28	10.4

.

This study examined the construct validity of the scale using both confirmatory factor analysis (CFA) and Average Variance Extracted (AVE). Firstly, CFA was conducted, and the results indicated that the original model of 5 factors with 17 items was not ideal. Subsequently, the model was refined by removing certain items, resulting in a final model with 5 factors and 14 items where the fit indices reached satisfactory levels. Specific fit indices are detailed in

Table 8. Structural model fit indices.

Fit Indices	X2/df	RMSEA	GFI	CFI	IFI	NFI	AGFI	PNFI	PGFI
5 factors, 17 items	2.783	0.081	0.877	0.903	0.904	0.858	0.827	0.688	0.625
5 factors, 14 items	2.245	0.074	0.929	0.954	0.955	0.922	0.889	0.64	0.679

below.

In addition, all standardized factor loadings of the retained items in the model meet the criterion of 0.6 < β < 1, indicating satisfactory convergent validity for each factor in this study.

Table 9. Names of five factors and corresponding items.

Factor	Factor Naming	Items	Original IDs
F1	Social Experience	I find it easy to communicate with AI chatbots	CAE22
		I believe AI chatbots can solve complex problems	CAE24
		I think AI chatbots are personable	CAE25
		I trust AI chatbots to handle tasks	CAE26
F2	Service Experience	AI personalized recommendations save me time and effort	CAE2
		AI personalized recommendations provide me with various choices	CAE3
F3	Intellectual Experience	AI recommendations spark my curiosity and improve problem-solving	CAE16
		Using AI recommendations makes me want to explore and study them	CAE17
		Through long-term use and reflection, I grasp AI recommendation patterns and utilize them	CAE18
F4	Exploitation Experience	I feel AI personalized recommendations are using my information	CAE6
		I believe AI personalized recommendations pry into my information	CAE7
		AI personalized recommendations listen in on my conversations	CAE8
F5	Classification Experience	I believe AI personalized recommendations can accurately predict my spending habits	CAE12
		I think AI recommends personalized products that fit my identity	CAE14

below presents the naming of the five factors and the final corresponding items for each factor.

The overall Cronbach’s Alpha coefficient for the scale is 0.895. The coefficients for the subscales, including social experience, service experience, intellectual experience, exploitation experience, and classification experience, are 0.648, 0.884, 0.663, 0.796, and 0.815, respectively, all of which fall within the acceptable range. In addition, as shown in

Table 10. Validity of CAE scales.

Variable	Service Experience	Exploitation Experience	Classification Experience	Intellectual Experience	Social Experience
Service Experience	0.695
Exploitation Experience	0.465 ***	0.857
Classification Experience	0.249 ***	0.491 ***	0.730
Intellectual Experience	0.464 ***	0.658 ***	0.395 ***	0.756
Social Experience	0.413 ***	0.751 ***	0.334 ***	0.629 ***	0.741

Note 1: The numbers on the diagonal represent the Average Variance Extracted (AVE) for each factor. The numbers below the diagonal represent the correlation coefficients between factors; * indicates p < 0.05; ** indicates p < 0.01; *** indicates p < 0.001.

, the AVE values for the five factors are all within the acceptable range. Furthermore, the absolute values of the correlation coefficients are smaller than the square roots of the corresponding AVEs, which demonstrates good convergent validity and discriminant validity.

3.3. Prediction Testing of Customer AI Experience Structure

The data for the predictive validation test were sourced from Sample 2, which was used for confirmatory factor analysis. Given that purchase intention, customer satisfaction, and word-of-mouth are common outcome variables of customer experience, these three variables were measured simultaneously during the collection of Sample 2, facilitating the predictive validation of the customer AI experience structure. For the study, purchase intention analysis utilized a questionnaire developed by Shan and Cui (2020) [42], consisting of 3 items; customer satisfaction analysis used a questionnaire developed by Homburg et al. (2009) [43], also consisting of 3 items; word-of-mouth analysis adopted a questionnaire developed by Ruvio et al. (2020) [44], comprising 3 items. Using SPSS tools, reliability testing was conducted on the three variables—purchase intention, customer satisfaction, and word-of-mouth—with Cronbach’s Alpha coefficients of 0.785, 0.831, and 0.823, respectively. All coefficients exceeded 0.7, indicating good reliability for the three variables.

Regression analysis. Given that the “being exploited experience” belongs to the negative dimension while others belong to positive dimensions, to ensure consistency, the collected data from the scales were reversed during the regression analysis phase, using a Likert seven-point scale. Results showed that CAE positively influenced purchase intention (β = 0.725, SE = 0.045, p < 0.001), customer satisfaction (β = 0.767, SE = 0.043, p < 0.001), and word-of-mouth (β = 0.696, SE = 0.050, p < 0.001). The better the customer’s experience with AI services, the more willing they were to purchase related products, the higher their satisfaction, and the more likely they were to spread positive word-of-mouth about AI services, recommending them to others. Therefore, the CAE scale developed in Study 1 demonstrated relatively ideal construct validity, significantly impacting the three outcome variables: purchase intention, customer satisfaction, and word-of-mouth.

4. Study 2: Mechanisms of Customer AI Experience Formation Based on Digital Interaction Platforms

Study 1 clarified the structure and implications of customer AI experiences but remained unclear about their formation process. Currently, research on the formation mechanism of customer AI experiences is still sparse. Some scholars argue that the consumer experience of AI products is influenced by AI product factors, consumer factors, and the interaction between products and consumers, but the dynamic formation process and mechanisms of customer AI experiences have not been specifically elucidated [23]. Moreover, through a literature review, it is found that under digital interaction platforms, customer AI experiences should involve multiple stages and exhibit iterative changes. Studying customer AI experiences solely from a static perspective may, therefore, be insufficient. Therefore, Study 2 will reconsider the specific formation and cyclical mechanisms of customer AI experiences from a dynamic process perspective, where the interaction between customers and AI systems is not static but rather a continuously evolving and mutually influential process. In this process, the customer’s experience is constantly fed back, adjusted, and optimized as time and usage context change, forming an ongoing cycle. The specific process will be discussed in detail in the following sections.

4.1. Research Design and Data Collection

4.1.1. Research Design

Study 2 adopts a grounded theory approach to construct the process mechanism of customer AI experiences. Grounded theory utilizes data analysis to reveal relationships between key concepts and thereby constructs theory. Therefore, Study 2 chooses the grounded theory research method, employing interviews, observations, archival data, and other methods to investigate deeply into the core elements that form customer AI experiences. This approach aims to uncover the formation mechanism of customer AI experiences.

4.1.2. Research Methodology

The respondents in this study were recruited through a snowball sampling method, with the criteria that participants had prior experience with online shopping and had interacted with AI customer service. An additional requirement was that respondents should be able to clearly recall their most recent interaction with AI customer service. Moreover, this study also collected official disclosure materials related to personalized recommendations and AI customer service from shopping apps mentioned by the participants, such as Taobao and JD, to supplement the coding process. In total, 24 interviewees were involved in in-depth interviews for Study 2, resulting in 64,000 words of interview content. The interview outline can be found in Appendix A, and the respondent information is presented in

Table 11. Interviewee information and coding table.

Code	Gender	Age	Occupation	Education	User Age (Based on Preferred Platform)
1201	Male	20	Student	Bachelor’s	2 years (Pinduoduo)
2102	Female	27	Kindergarten Teacher	Bachelor’s	5–6 years (Taobao)
2203	Female	16	Student	High School	3 years (Taobao)
2204	Female	23	Student	Master’s	5–6 years (Taobao)
2205	Female	26	Student	Master’s	7 years (Taobao)
2306	Female	27	Public Servant	Bachelor’s	7 years (Taobao)
1107	Male	25	Internet Industry Professional	Bachelor’s	3 years (Taobao)
1108	Male	26	Railway Worker	Associate Degree	Nearly 10 years (Taobao)
2209	Female	25	Student	Master’s	7 years (Taobao)
2310	Female	26	Public Servant	Bachelor’s	8 years (Taobao)
2211	Female	29	Student	Doctorate	Nearly 10 years (Taobao)
2212	Female	22	Student	Bachelor’s	4 years (Taobao)
2113	Female	25	Foreign Trade Worker	Associate Degree	5 years (Taobao)
2114	Female	26	Middle School Teacher	Bachelor’s	9 years (Taobao)
1115	Male	26	High School Teacher	Bachelor’s	7 years (JD.com)
1116	Male	25	E-commerce	Bachelor’s	7 years (Taobao)
1117	Male	27	Sales	Bachelor’s	6–7 years (Taobao)
1118	Male	27	Sales	Bachelor’s	3 years (JD.com)
1219	Male	25	Student	Master’s	6 years (JD.com)
2120	Female	29	State Grid	Master’s	10 years (Taobao)
2221	Female	32	Student	Doctorate	5 years (Pinduoduo)
1222	Male	30	Student	Master’s	9 years (Taobao)
1123	Male	35	Administrative Assistant	Master’s	9 years (Taobao)
1124	Male	41	Service Worker	High School	9 years (Taobao)

.

4.2. Data Coding and Analysis

4.2.1. Open Coding

The interview data with codes 2221, 1222, 1123, and 1124 were used for theoretical saturation testing. After multiple comparisons, no new concepts or categories emerged, indicating theoretical saturation was achieved. The data were sequentially coded, resulting in the extraction of 155 initial concepts, which were further condensed into 31 subcategories. Due to space limitations, only the subcategory of subjective knowledge is presented as an example in

Table 12. Partial examples of open coding.

Original Text Data and Corresponding Code	Initial Concept	Sub-Category
Based on my understanding of the technology used in this feature, I believe using it for recommendations and AI customer service should bring me convenience (1201). I also know it serves merely as an auxiliary means, and my expectations of it are not high. That’s my perspective (1107).	Knowledge of AI Technology (SK21)	Subjective Knowledge (SK2)
I think intelligent recommendation is essentially deduced from big data, based on my previous shopping and browsing records, to predict which products I might need, and then push them through the system (2209).	AI service knowledge (SK22)

.

4.2.2. Axial Coding

Principal axis coding involves further categorizing the subcategories formed from open coding, based on logical relationships and potential links between research categories, ultimately achieving clustering and naming of core categories. Building upon the 31 subcategories obtained from open coding, further clustering resulted in a total of 12 core categories. The core categories and their corresponding subcategories are presented in

Table 13. Main categories formed by principal axis coding.

Core Category	Sub-Category		Category Connotation
Customer Knowledge (CK1)	Past Experience (PE1)		After the customer AI experience event ends, people store the acquired information in memory.
	Subjective Knowledge (SK2)		The sum of product category information and related skills stored in customer memory, reflecting the subjective perception of the level of understanding of the service itself.
Usage Contact (UC2)	Initial Contact (IC3)		Initial practical use of products or services by customers, including services supporting such usage.
	Long-term Contact (LC4)		Long-term practical use of products or services by customers, including services supporting such usage.
AI Services (AIS3)	Personalized Recommendations (PR5)		Based on customer-related information to extract preference characteristics, recommending products, services, or other information that customers might be interested in.
	Intelligent Customer Service (ICS6)		Non-human customer service systems developed and iterated based on AI technology, providing efficient problem-solving communication to customers.
Customer Choice (CC4)	Habitual Choice (HC7)		Psychological tendency to repeat past behaviors.
	Incidental Choice (AC8)		Psychological tendency to consciously attempt new behaviors.
Customer Engagement (CE5)	Self-Engagement (SE9)		Utilizing AI services to fulfill personal psychological needs.
	Task Engagement (TE10)		Utilizing AI services to achieve goals and solve problems.
	Passive Engagement (PE11)		Customers cannot refuse or avoid AI services.
Interaction (IA6)	Short-term Interaction (SI12)		Real-time customer-AI interaction based on short-term/temporary behavioral data.
	Long-term Interaction (LI13)		Dynamic and coordinated customer-AI interaction based on long-term behavioral data.
AI Capabilities (AIC7)	Listening Capability (AC14)		AI gathers data about consumers and their living environments.
	Classification Capability (CP15)		AI analyzes and predicts customer needs and categorizes service pushes to customers.
	Communication Capability (AP16)		AI’s ability to engage in interactive communication.
Customer AI Experience (CAIX8)		Service Experience (SX17)	Based on AI’s listening capability, customers perceive themselves as being served by AI.
		Exploitation Experience (UX18)	Based on AI’s listening capability, customers perceive themselves as being utilized by AI.
		Classification Experience (CX19)	Customers receive AI’s personalized predictive experiences based on AI’s prediction capability.
		Intellectual Experience (IX20)	Thinking experience generated by customer learning ability interacting with AI services.
		Social Experience (CX21)	Experience of interaction and communication with AI partners based on AI’s communication capability.
Customer Perceived Value (CPV9)		Product Value (PV22)	Value perceived and obtained by customers from the product itself provided by AI services.
		Service Value (SV23)	Intangible value perceived and obtained by customers from enterprises throughout their interaction with AI services.
		Experience Value (XV24)	Value perceived by customers from products or services provided by AI services derived from inner feelings.
Comparison (CP10)		Comparison (CP25)	Customers balance benefits and sacrifices related to AI experience.
Customer Perceived Cost (CPC11)		Monetary Cost (MC26)	Cost incurred by customers for purchasing products recommended by AI services.
		Energy Cost (EC27)	Total time and mental cost incurred by customers during interaction with AI services.
		Psychological Cost (PC28)	Psychological “unhappiness” felt by customers during interaction with AI services.
Customer Value Judgment (CVJ12)		Value for Money (VM29)	Customers consider the value provided by AI services equals the cost they incurred.
		Value Exceeding Money (CE30)	Customers consider the value provided by AI services exceeds the cost they incurred.
		Value Not Meeting Money (NCE31)	Customers consider the value provided by AI services is less than the cost they incurred.

below.

4.2.3. Selective Coding

As the third step in implementing the procedural grounded theory coding process, selective coding involves further clustering to select core categories.

The first logical storyline surrounding the core concept of the “customer AI experience” is as follows: when customers first use or repeatedly use AI-powered personalized recommendations and intelligent customer service, they typically make choices based on past experiences and subjective knowledge. Personalized recommendations stimulate customers’ desire to purchase and enhance the shopping experience by uncovering potential purchasing needs, primarily relying on machine learning and deep learning technologies. On the other hand, intelligent customer service, when the user’s needs are clearly defined, provides precise answers using natural language processing (NLP) algorithms [10]. Customer interactions with AI services can be categorized into two types: during the initial use, customers may continue to use the service out of curiosity; with long-term use, customers are more likely to habitually repeat previous behaviors, such as viewing recommendations or consulting customer service. The logic of the customer AI experience expectation phase is illustrated in Figure 2.

The second storyline is as follows: in the realization phase of the customer AI experience, customers generate AI experiences through interaction with AI services [20]. Customer integration into AI services can occur in three ways: task integration, self-integration, and passive integration. Task integration is primarily based on the customer’s goal-oriented needs, while self-integration focuses on psychological needs, such as entertainment or passing time. Passive integration occurs when customers are compelled to engage with the service due to its setup, with no other choice. The creation of AI experiences depends on the powerful functionalities of AI, including listening, categorization, and communication abilities. Through listening, AI collects data and provides personalized services, making customers feel served. However, customers may also feel exploited due to the lack of transparency, particularly the psychological consequences of privacy breaches and loss of control [45]. In the categorization experience, customers see themselves as part of a specific group, and personalized classification enhances self-identity and satisfies identity-related motivations [46,47]. The social experience involves AI’s interactive capabilities, where robotic technology enables emotionally rich services [40]. The intellectual experience stimulates customers’ thinking through AI services, preventing boredom and helping customers understand the service rules [48]. The logic of the customer AI experience realization phase is shown in Figure 3.

The third storyline is as follows: in the reflection phase of the customer AI experience, customers make value judgments based on perceived value and cost. The perceived value consists of product value, service value, and experiential value. Product value refers to the value derived from the specific items recommended, such as “the recommended products are of good quality”. Service value refers to the perceived value from interactions with personalized recommendations and AI customer service, such as “personalized recommendations save time”. The experiential value is the psychological value customers feel, such as “the shopping experience becomes more enjoyable” or “it satisfies curiosity”. Customers also consider the costs they incur, including monetary cost, effort cost, and psychological cost. Monetary cost refers to the expense of purchasing products; effort cost refers to the time, mental effort, and information cost consumed during interaction with AI services, such as “information is too fragmented, making choices difficult”; the psychological cost includes the negative emotions experienced by customers, such as “wasting time due to irrelevant responses” or “waiting for human customer service”. After comparing value and cost, customers make one of three value judgments: worth the price, exceeds expectations, or not worth the price. The logic of the reflection phase of the customer AI experience is shown in Figure 4.

4.3. Circular Model Construction

Based on the grounded coding results and the characteristics of digital interactive platforms, the formation of the customer AI experience follows a three-phase cyclical model [27,28], consisting of the anticipation phase, realization phase, and reflection phase, with new features emerging in each stage. In the anticipation phase, customers make usage decisions based on their expectations of future events and choose AI services to achieve the anticipated value outcomes [49]. The key factor is that customers can “pre-experience” future outcomes and make their choices accordingly. The interaction may be a first-time encounter or long-term use. When the value judgment in the reflection phase is satisfactory, customers tend to continue using the AI service [50]; if the judgment is that the value does not meet expectations, they may cease using the service. The selection process is often automatic and habitual, and once a choice is made, the customer enters the realization phase [49].

The realization phase is the core stage of interaction between the customer and AI services, where the customer gradually becomes immersed and engages in long- or short-term interactions with AI functions, forming the AI experience. This interaction emphasizes the customer’s actual experience, not just simple contact. Past research has mainly defined customer experience as multi-dimensional, including cognitive, emotional, sensory, behavioral, and relational aspects (Schmitt, 1999) [13]. However, with the application of AI technology, customer experience has evolved. Puntoni et al. (2021) [12] proposed that the customer AI experience includes data capture, classification, delegation, and the social experience, reflecting how consumers interact with AI functions. During this stage, the customer is primarily driven by personal goals [20], such as task-based or self-directed immersion. Additionally, customers may be passively immersed due to AI service settings. Research also found the presence of intellectual experience in the customer AI experience.

In the reflection phase, customers evaluate the value realization based on their interactions with AI services. They weigh the perceived value against the cost and decide whether to continue using the AI service [21]. The customer’s evaluation will influence future usage decisions: if the experience is poor, the customer may reject using the AI service; if the experience is satisfactory, the customer may initiate a new interaction cycle and have higher expectations for the AI service. Therefore, the customer AI experience is a dynamic and repetitive process, consisting of three stages, as shown in Figure 5.

In summary, the specific formation and cycling mechanism of the customer AI experience is as follows: In the expectation stage, customers form an intention to further experience AI-related services based on their own knowledge and initial contact with AI services. In the realization stage, customers immerse themselves in related services and generate customer AI experiences through interaction with AI functionalities, which specifically include the service experience, exploitation experience, classification experience, social experience, and intellectual experience. The final reflection stage is primarily where customers make value judgments based on a comparison of perceived value and costs. If the value judgment is positive, customers are very likely to re-enter the expectation stage and start a new cycle. However, if the value judgment is negative, customers are very likely to discontinue using AI services.

5. Study 3: Mechanisms of Customer AI Experience on Customer Engagement Behavior in Digital Interaction Platforms

Studies 1 and 2 developed a measurement scale for customer AI experience, revealing its significant impact on customer immersion. They found that the key distinction between the customer AI experience and traditional experiences lies in the perception of AI technology. Building on this, Study 3 will further empirically examine the technological antecedents of the customer AI experience and its impact on customer immersion.

5.1. Theoretical Model and Research Hypotheses

Perceived AI innovation characteristics mainly include compatibility, observability, and result demonstrability. Compatibility refers to the degree to which a technological innovation is perceived by individuals as being consistent with their existing values, current needs, and past experiences [30]. In B2C commerce platforms, compatibility primarily refers to the extent to which AI services align with various aspects of customers’ shopping experiences, such as whether AI services match customers’ preferred shopping methods and styles. Observability refers to the degree to which individuals can see others using the innovation technology [30], also known as visibility. Given the prevalence of AI services in B2C commerce platforms, the high degree of observability of innovative technology may further promote customer usage. Result demonstrability refers to the tangibility of using the innovation, including its observability and communicability [25].

Based on the research of numerous domestic and international scholars, the Diffusion of Innovations theory in the TAM model has a strong explanatory power for perceived usefulness and perceived ease of use [51]. In B2C platforms, the primary appeal of AI technology services for most customers lies in the real-time and convenient assistance it provides for shopping, which often aligns with many customers’ shopping habits, thus making customers perceive the technology as useful and entertaining. Additionally, as social beings, the widespread use of AI technology enhances customers’ perceived usefulness and ease of use through increased observability. The higher the observability of perceived AI innovation characteristics, the easier it is for users to recognize its advantages and communicate them to others, facilitating discussions within their social circles and further enhancing their perception of the technology’s usefulness, ease of use, and entertainment value. Furthermore, the compatibility, observability, and result demonstrability of perceived AI innovation characteristics can directly affect the customer experience, as AI services can provide more precise personal recommendations, offering a more efficient, accurate, enjoyable, and satisfying shopping experience, thereby enhancing the customer AI experience. Therefore, the following hypotheses are proposed:

H1: 

Perceived AI innovation characteristics have a significant positive impact on perceived usefulness;

H2: 

Perceived AI innovation characteristics have a significant positive impact on perceived ease of use;

H3: 

Perceived AI innovation characteristics have a significant positive impact on perceived entertainment value;

H4: 

Perceived AI innovation characteristics have a significant positive impact on the customer AI experience.

Perceived usefulness reflects the extent to which an individual believes that using new information technology helps in their work and life. In the context of B2C platforms, it mainly refers to how AI technology can assist customers in shopping or solving shopping-related issues (such as after-sales services). For example, AI technology can collect and analyze data about consumers and their living environments through its listening and prediction functions [12], offering personalized services to enhance the customer experience [52]. Perceived ease of use, on the other hand, reflects the perceived difficulty or ease with which users can adopt new technology, which influences their intention and behavior toward technology adoption. When users find AI service features difficult to operate, it may lead to dissatisfaction [53]. However, when customers perceive AI technology services as easy to use, the process of searching for information about products and services becomes smoother, leading to a better customer experience [54]. Moreover, considering that the enjoyment perceived by customers when shopping on B2C platforms is often more important than the technological factors [27], consumers may become fully immersed in the shopping activity, experiencing a distorted sense of time and strong emotions [55]. In other words, time unconsciously passes for the consumer, generating positive emotions and, as a result, fostering a positive customer experience. As Hoffman and Novak (2009) [56] pointed out, if customers do not experience enjoyment in their activities, they will not achieve the optimal experience. Therefore, the following hypotheses are proposed:

H5: 

Perceived usefulness has a significant positive impact on the customer AI experience;

H6: 

Perceived ease of use has a significant positive impact on the customer AI experience;

H7: 

Perceived entertainment value has a significant positive impact on the customer AI experience.

Customer engagement includes both direct contributions through purchases and indirect profit contributions to the company, such as their recommendation behavior, online influence on potential customers and other customers’ purchases, and feedback on the products and services they consume [39]. Unlike the motivation-based definition of customer engagement in Section 4, this section focuses on customer engagement behavior, which refers to the behavioral expressions driven by customers’ intrinsic motivations toward a service or brand. On digital interaction platforms, companies establish, maintain, and develop relationships with customers that may even transcend mere transactions. A positive customer AI experience ultimately influences customer engagement by ensuring customer satisfaction and fostering emotional bonds with the company [37,57]. For example, customers may increase their consumption, recommend the service to others, engage in discussions on related topics, or even provide feedback to the platform to promote technological innovation. Based on the above analysis, if perceived AI innovation characteristics influence perceived usefulness, perceived ease of use, and perceived enjoyment, and perceived usefulness, perceived ease of use, and perceived enjoyment in turn affect customer AI experience, it can be reasonably inferred that perceived usefulness, perceived ease of use, and perceived enjoyment mediate the relationship between the perceived AI innovation characteristics and customer AI experience. Therefore, the following hypotheses are proposed:

H8: 

The customer AI experience has a significant positive impact on customer engagement behaviors;

H9: 

Perceived usefulness, perceived ease of use, and perceived entertainment have different mediating effects between perceived AI innovation characteristics and the customer AI experience.

The theoretical research model is shown in Figure 6:

5.2. Data Analysis and Hypothesis Testing

5.2.1. Source of Scale Items

Except for the measurement scale of the customer AI experience, which was developed for this study, the measurements for the other variables are derived from the research scales of relevant scholars both domestically and internationally, as shown in

Table 14. Scale items and sources.

Variable	Measurement Items	Literature Source
Perceived AI Innovation Characteristics	I believe using AI services is compatible with various aspects of my shopping (e.g., existing values, needs, and past experiences). I think using AI services aligns well with my preferred shopping methods. I feel using AI services fits my shopping style. I have seen others use AI services. I have observed others using AI services. It is easy for me to observe others using AI services. I can effortlessly explain the outcomes of using AI services to others. I believe I can articulate the outcomes of using AI services. I find the outcomes of using AI services to be obvious to me.	(Moore and Benbasat, 1991; Venkatesh et al., 2003)
Perceived Usefulness	10. I believe using AI services can enhance my shopping efficiency. 11. I believe using AI services makes my shopping easier. 12. I believe using AI services can optimize my shopping outcomes.	(Davis, 1989)
Perceived Ease of Use	13. I find AI services easy to use. 14. I find it easy to get AI services to do what I want. 15. I find interaction with AI services flexible. 16. Interaction with AI services is clear and understandable.
Perceived Enjoyment	17. I find using AI services a very enjoyable process. 18. I find using AI services a relaxed and pleasant process. 19. I really enjoy using AI services.	(Liu and Liu, 2015)
Customer AI Experience	20. I find communication with AI services (e.g., AI chatbots) smooth. 21. I believe AI services (e.g., AI chatbots) can solve complex problems. 22. I think AI services (e.g., AI chatbots) have good human-like qualities. 23. I feel reassured by the service provided by AI services (e.g., AI chatbots). 24. I find AI services provide me with convenient and time-saving services. 25. I feel AI services provide me with multiple choices. 26. I perceive that AI services are using my information. 27. I believe AI services intrude into my information. 28. AI services eavesdrop on my conversations. Intellectual Experience 29. AI services stimulate my curiosity and enhance problem-solving abilities. 30. Using AI services makes me want to research and explore them. 31. Through long-term use and reflection, I will master the rules and utilization of AI services. 32. I believe AI services can accurately predict my spending capacity. 33. I believe AI services recommend personalized products that align with my identity.	Developed in this study
Customer Engagement Behavior	34. I usually use AI services to find products of interest. 35. I typically purchase products recommended by AI services. 36. I often use AI services to answer shopping-related questions (about product information, after-sales). 37. I engage in discussions about AI services (e.g., reviews, inquiries, and sharing issues). 38. I recommend others to use AI services. 39. I share my consumer experiences using AI services with others. 40. I provide feedback on my experiences using AI services to the platform.	(Kim and Drumwright M, 2016; Wu et al., 2019)

.

5.2.2. Data Collection and Descriptive Statistics

This study collected questionnaires through a survey method, using the Credemo platform for questionnaire collection. A total of 320 questionnaires were distributed. After excluding invalid questionnaires due to missing data and inconsistent responses, 300 valid questionnaires were obtained, resulting in a valid response rate of 93.75%, and the basic characteristics of the sample are shown in

Table 15. Basic characteristics of sample (N = 300).

Demographic Variable	Variable Value	Frequency	Percentage%
Gender	Male	118	39.3
	Female	182	60.6
Age	0~20	13	4.3
	21~30	145	48.4
	31~40	102	43
	41~50	28	9.3
	51~60	12	4
Occupation	Student	45	15
	Private enterprises	131	43.7
	State-owned enterprises	74	24.7
	Foreign-funded enterprises	23	7.7
	Government-affiliated agencies	16	5.3
	Civil servants	11	3.6
Education	Bachelor’s degree	222	74
	Master’s degree	46	15.4
	Doctorate	4	1.3
	High school/Technical school and below	27	9
	Junior high school	1	0.3

.

5.2.3. Test of Common Method Bias

Since this study used a self-reported questionnaire survey, there may be concerns about common method bias. To address this, an exploratory factor analysis (EFA) was conducted on all the questionnaire items using Harman’s single-factor test. After performing a varimax rotation, six factors with eigenvalues greater than 1 were extracted. The variance explained by the first factor was 21.02%, which is below the 40% threshold, indicating that there is no significant common method bias. Additionally, a multicollinearity test was conducted on the model, and the results showed that the Variance Inflation Factors (VIFs) ranged from 2.84 to 3.44, suggesting that there is no significant multicollinearity issue among the variables.

5.2.4. Reliability and Validity Testing

The data analysis results indicate that the Cronbach’s Alpha coefficients and CR values for all scales are greater than 0.7. Furthermore, the results of the confirmatory factor analysis show that the X²/df for the six-factor model is 1.180, which is less than 3; the GFI, AGFI, CFI, and NFI are all greater than 0.9; and RMSEA is less than 0.08, suggesting that the six-factor model has a good fit with the data.

5.3. Data Processing

5.3.1. Descriptive Statistics and Correlation Analysis Between Variables

As shown in

Table 16. Correlation analysis among variables.

Scale	M	SD	1	2	3	4	5	6
Perceived AI Innovation Features	5.8902	0.61309	1
Perceived Utility	5.9802	0.81338	0.662 **	1
Perceived Usability	5.8608	0.79716	0.756 **	0.741 **	1
Perceived Enjoyment	5.8832	0.81251	0.729 **	0.744 **	0.713 **	1
Customer AI Experience	5.391	0.80007	0.652 **	0.735 **	0.740 **	0.683 **	1
Customer Engagement Behavior	5.6881	0.86262	0.654 **	0.718 **	0.706 **	0.743 **	0.801 **	1

Note: ** p < 0.01.

, the variables of perceived AI innovation characteristics, perceived usefulness, perceived ease of use, perceived entertainment value, customer AI experience, and customer engagement behavior are all significantly positively correlated with each other pairwise.

Under the condition of controlling for variables, regression analyses were conducted, and the results are shown in

Table 17. Regression analysis among variables.

Variable	Perceived Usefulness	Perceived Ease of Use	Perceived Enjoyment	Customer AI Experience				Customer Engagement Behavior
Model	M1	M2	M3	M4	M5	M6	M7	M8
Gender	−0.01	−0.01	−0.03	−0.07	−0.06	−0.06	−0.04	0.04
Age	0.04	0.00	−0.03	0.07	0.05	0.08	0.10 *	0.00
Education	−0.03	−0.08 *	−0.09 *	−0.06	−0.04	0.00	0.00	0.06
Occupation	0.02	0.06	−0.01	0.11 *	0.09 *	0.06	0.11 *	0.04
Perceived AI Innovation	0.66 ***	0.76 ***	0.73 ***	0.64 ***
Perceived Usefulness					0.72 ***
Perceived Ease of Use						0.72 ***
Perceived Enjoyment							0.68 ***
Customer AI Experience								0.79 ***
R2	0.44	0.58	0.54	0.43	0.54	0.54	0.48	0.62
F	46.39 ***	81.93 ***	69.67 ***	45.89 ***	68.58 ***	70.79 ***	55.01 ***	100.09 ***

Note: N = 300; *** p < 0.001; ** p < 0.01; * p < 0.05.

. The perceived AI innovation positively influence perceived usefulness, perceived ease of use, perceived enjoyment, and the customer AI experience (β = 0.66; β = 0.76; β = 0.73; β = 0.64). Perceived usefulness, perceived ease of use, and perceived enjoyment have direct positive predictive effects on the customer AI experience (β = 0.72; β = 0.72; β = 0.68). The customer AI experience has a direct positive predictive effect on customer engagement behavior (β = 0.79). Hypotheses H1–H8 are all supported.

5.3.2. Mediation Analysis

Firstly, we examined the mediating effects of perceived usefulness, perceived ease of use, and perceived entertainment value between perceived AI innovation characteristics and the customer AI experience. The results in

Table 18. Mediation effects of perception of usefulness, usability, and entertainment on the relationship between the perception of AI innovation and the customer AI experience.

Path	Effect Value	Boot Std. Error	Boot Lower CI	Boot Upper CI	Relative Mediation Effect
Total Direct Effect	0.749	0.095	0.557	0.929	88%
Perception of AI Innovation → Perception of Usefulness → Customer AI Experience	0.275	0.065	0.146	0.400	32%
Perception of AI Innovation → Perception of Usability → Customer AI Experience	0.332	0.082	0.180	0.498	39%
Perception of AI Innovation → Perception of Entertainment → Customer AI Experience	0.142	0.059	0.031	0.263	17%

indicate that the mediating effects of these three perceptions are all significant, with a total indirect effect of 0.749. Specifically, the 95% bootstrap confidence intervals for all three mediating paths do not include 0, indicating that the mediating effects of perceived usefulness, perceived ease of use, and perceived entertainment value are significant. Therefore, hypothesis H9 is supported.

Furthermore, we also tested the three serial mediating effects and the primary mediating effect depicted in the model diagram, with results presented in

Table 19. Chain mediation effects.

Chain Mediation	Effect Value	Boot Std. Error	Boot Lower CI	Boot Upper CI
Chain Mediation Effect 1	0.265	0.054	0.163	0.376
Chain Mediation Effect 2	0.354	0.071	0.224	0.501
Chain Mediation Effect 3	0.240	0.058	0.133	0.365
Perception of AI Innovation → Customer AI Experience → Customer Involvement Behavior	0.224	0.044	0.144	0.314

. The effects are as follows: Serial mediating effect 1: perceived AI innovation characteristics → perceived usefulness → customer AI experience → customer engagement behavior has an effect size of 0.265. Serial mediating effect 2: perceived AI innovation characteristics → perceived ease of use → customer AI experience → customer engagement behavior has an effect size of 0.354. Serial mediating effect 3: perceived AI innovation characteristics → perceived entertainment value → customer AI experience → customer engagement behavior has an effect size of 0.240. The 95% bootstrap confidence intervals for all these serial mediating paths and the primary mediating path (perceived AI innovation characteristics → customer AI experience → customer engagement behavior) do not include 0, indicating that these effects are all significant.

6. Discussion

6.1. Theoretical Contributions

Firstly, based on the perspective of the customer interaction with AI capabilities, a measurement scale for the customer AI experience was developed, which enriches the dimensions of existing experience scales and is more suitable for the evolving AI contexts. Specifically, the scale developed in this study for the customer AI experience distinguishes itself from existing measures of customer experience dimensions (such as cognitive, affective, and sensory dimensions) by adding the dimension of intellectual experience in customer AI interactions. As a result, the customer AI experience scale was structured into five dimensions and 14 items: categorical experience, social experience, intellectual experience, being served, and being utilized. This scale was validated and enriched the framework of the customer AI experience proposed by Puntoni et al. [12]. It expands the application scope of the customer experience in AI contexts, providing theoretical references and empirical tools for scholars worldwide to measure and research customer AI experiences.

Secondly, this study has uncovered the cyclic mechanism underlying the formation and development of customer AI experiences, thereby expanding the three-stage loop model of customer experience. The research identifies three stages in the formation and development of customer AI experiences: anticipation, realization, and reflection. During the anticipation stage, customers rely on their ability to anticipate experiences and outcomes, often leading them to choose to engage with AI services based on these expectations. In the realization stage, customers engage with AI capabilities within the service context, thereby experiencing customer AI interactions. In the reflection stage, customers compare the value received with their expectations; if they perceive the service as meeting or exceeding their expectations, they are likely to initiate new interactions with AI services when new shopping needs arise, potentially raising their expectations for AI services. Building upon the three-stage loop model of customer experience [20] and introducing concepts relevant to customer AI experiences, this study further extends the understanding of the cyclic formation mechanism within AI development contexts.

Finally, this study, combining the Innovation Diffusion Theory (IDT) and the Technology Acceptance Model (TAM), empirically explores the mechanism through which the customer AI experience influences customer engagement behavior via mediating variables such as perceived AI innovation features, perceived usefulness, perceived ease of use, and perceived enjoyment. This research deepens the theoretical integration and dialogue between these two frameworks. On one hand, by developing and validating a measurement tool for the customer AI experience, this study fills the gap in existing research regarding the comprehensive customer experience during the use of AI technology. On the other hand, this study further explores the heterogeneity of different user groups in the process of AI technology adoption, addressing previous research on how innovation features affect customer perceptions of usefulness [20,23], and reveals how perceived AI innovation features influence the customer experience and engagement behavior mechanisms under different user characteristics. Furthermore, by constructing a chain mediation model, this research fills the gap in the existing literature regarding the impact of AI technology usage on customer engagement behavior in digital interactive platforms, providing an empirical foundation for future research. It offers new insights into the relationship between AI technology application and customer behavior in digital environments, presenting a fresh perspective for theoretical models of customer engagement.

6.2. Managerial Insights

This study demonstrates significant practical implications for enterprises offering AI services. In addressing the anticipation stage of customer AI experiences, enterprises should continuously refine their AI-related services to attract initial and repeat usage from customers. Enterprises can set simple or complex and varied pages for different customer demands. During the realization stage, enterprises should pay particular attention to negative experiences across the five types of customer AI experiences. For instance, misunderstanding experiences in categorical experiences and isolation experiences in social experiences. Enterprises should further enhance AI-related services and capabilities in these areas. They can also promptly deploy employee services such as JD Intelligent Customer Service Technology, seamlessly connecting AI and human customer service to improve customer experiences when negative emotions are detected.

Enterprises can further explore the relevant characteristics of AI innovation to enhance compatibility with customers’ personalized styles based on data analysis. They can allow customers to openly discuss their usage experiences in online communities hosted by certain companies, facilitating communication among customers, improving the observability and communicability of the technology. This can ultimately enhance customers’ perceptions of the usefulness, ease of use, and entertainment value of AI technology, thereby delivering a better experience for customers and creating value for the enterprise. Specifically, enterprises should prioritize improving customers’ perceptions of usability and usefulness. Many customers in interviews have pointed out issues such as increasing complexity in AI personalized recommendations, numerous categories leading to cluttered pages, and recommendations not meeting their requirements. Additionally, AI customer service sometimes fails to resolve issues effectively, reducing the overall user experience. To address these concerns, enterprises should simplify AI service pages for easier and more convenient use, invest more in technological research and development, and cater to customers’ entertainment needs by incorporating features like video recommendations.

Since 15 March 2022, apps such as Douyin, WeChat, Taobao, and Weibo have implemented algorithm shutdown switches according to the regulations. Users can easily disable “personalized recommendations” with a single click. However, some apps have hidden these switches deep within their settings, and not all apps actively prompt users to turn off personalized recommendations. Given customers’ reflective processes regarding AI experiences, dissatisfaction with the experience may lead to rejection or avoidance of AI services. Some customers have expressed confusion about how to provide feedback or disable these features. Therefore, enterprises should provide multiple convenient feedback channels regarding AI services, continuously optimize algorithms, and enhance customer stickiness. This approach supports long-term development in compliance with regulations, fostering sustainable growth for enterprises.

6.3. Research Limitations and Future Prospects

First, this study focused exclusively on shopping platforms within digital interaction platforms, with AI services limited to AI personalized recommendations and intelligent customer service. While this narrow focus helped sharpen the research questions, it also limited the generalizability of the customer AI experience. Future research could explore other types of digital interaction platforms (e.g., smart healthcare platforms) and focus on other AI-related intelligent services (e.g., smart health checks) in different contexts (e.g., healthcare and tourism), further testing the validity of the customer AI experience scale developed in this study. Second, with the continuous advancement of artificial intelligence, the forms of intelligent services are becoming increasingly diverse. Therefore, future research could explore the different impacts of other AI agents, such as personal virtual assistants and social media virtual avatars, on customer AI experience. Moreover, although this study examined the dynamic formation process of customer AI experience in Research 2, it did not address situations where customers may reject or disengage from the AI service at any stage of the three-stage cyclical process. Therefore, future research could investigate the mechanisms that hinder or cause stagnation and regression in the customer AI experience. Another related aspect is that this study did not empirically test the changes in the customer AI experience over a longer period. Thus, a feasible future research direction is to conduct longitudinal studies to examine how the customer AI experience changes over time and the potential boundary conditions that affect the customer AI experience at different stages of the customer journey. Finally, the five dimensions of the customer AI experience are not independent of each other. For instance, the experience of being used often occurs together with the intellectual experience. When customers feel exploited, they may become curious and think about the underlying mechanisms of AI services or even try to counteract them. However, this study did not delve further into these relationships. Future research could explore the interrelationships between these dimensions in greater detail.