Do Trusting Belief and Social Presence Matter? Service Satisfaction in Using AI Chatbots: Necessary Condition Analysis and Importance-Performance Map Analysis

Abstract

Research indicates that perceived trust affects both behavioral intention to use chatbots and service satisfaction provided by chatbots in customer service contexts. However, it remains unclear whether perceived propensity to trust impacts service satisfaction in this context. Thus, this research aims to explore how customers’ propensity to trust influences trusting beliefs and, subsequently, their satisfaction when using chatbots for customer service. Through purposive sampling, individuals in Hong Kong with prior experience using chatbots were selected to participate in a quantitative survey. The study employed Necessary Condition Analysis, Importance-Performance Map Analysis, and Partial Least Squares Structural Equation Modelling to examine factors influencing users’ trusting beliefs toward chatbots in customer service settings. Findings revealed that trust in chatbot interactions is significantly influenced by propensity to trust technology, social presence, perceived usefulness, and perceived ease of use. Consequently, these factors, along with trusting belief, also influence service satisfaction in this context. Thus, Social Presence, Perceived Ease of Use, Propensity to Trust, Perceived Usefulness, and Trusting Belief are found necessary. By combining Importance-Performance Map Analysis, priority managerial action areas were identified. This research extends the Technology Acceptance Model by incorporating social presence, propensity to trust technology, and trusting belief in the context of AI chatbot use for customer service.

1. Introduction

Despite the increasingly widespread adoption of chatbots, customers still commonly exhibit hesitation toward their usage. Some customers have doubts about the accuracy of answers provided by chatbots, while others find their template responses unsatisfactory. At times, chatbot-generated answers may even be irrelevant or contain unnecessary information, making users rephrase their questions repeatedly and end up using more time than intended [1]. Consequently, users’ trust toward chatbots’ work performance is compromised [2]. As a matter of fact, research has indicated that individuals tend to trust humans more than chatbots [3]. Thus, dissatisfied prior users may revert to using traditional customer service delivered by humans. In fact, many prefer to call a hotline operated by human customer service representatives for customized answers, which may sometimes be more efficient than their AI counterparts. Compared to chatbots, humans can normally provide more tailor-made and precise answers in one go. Therefore, gaining deeper insight into how trust influences users’ interactions with AI chatbots is essential.

Chatbots are considered a form of artificial intelligence—they are not human beings. In today’s customer service context, with built-in algorithms, technology has enabled chatbots to provide assistance, with personalized interactions and a sense of connection [4]. In some cases, technologically advanced chatbots can even be programmed to express empathy and reflection, thereby inducing resonance among customers [5]. In turn, this creates the feeling and impression that chatbots are always there to assist and provide support, hence forming dependency and fostering trust [2].

Several recent studies have highlighted perceived trust as a key determinant shaping users’ willingness to adopt chatbots [2,6,7]. In addition, findings from Alagarsamy and Mehrolia [7] indicate that within the banking industry, perceived trust plays a notable role in driving user satisfaction.

Despite the growing adoption of chatbots in customer service, limited research has examined whether perceived propensity to trust affects service satisfaction provided by chatbots. It is thus timely to explore this relationship further. Therefore, this study aims to examine how customers’ perceived propensity to trust influences their satisfaction with chatbot-delivered services. Additionally, the role of social presence is important: Do AI chatbots need to create human-like relationships with customers? To address this, the following research question was developed:

• Does social presence affect perceived trusting belief when using chatbots in the customer service context?
• How does perceived trusting belief affect service satisfaction when using chatbots?
• What are the priority areas that must be dealt with by managers, given the limited resources available?

Necessary condition analysis (NCA) has been used with partial least squares structural equation modeling (PLS-SEM) to understand not only the determinants of service satisfaction when using chatbots but also to identify essential determinants. These determinants include perceived usefulness, perceived ease of use, perceived social presence, propensity to trust technology, and trusting belief. The importance-performance map analysis (IPMA) provides priority areas for managers to address.

2. Literature Review

2.1. Social Presence Theory

Social Presence Theory has psychological roots and explores people’s sense of being with others in different communication contexts [8]. It originates from Short et al.’s [9] views regarding social psychology in telecommunications. Communication in society can take different forms, including face-to-face non-mediated communications, as well as mediated communications. It has been argued that the former generates more social presence than the latter in different circumstances [10]. However, in today’s commercial world, there are many discussions and applications of Social Presence Theory when people or customers are involved in mediated communications. The use of online technology such as AI chatbots has facilitated this line of thinking and discussion.

Perceived existence and salience are associated with perceived quality of the communication [11], quality of interactions between people (or customers in commerce or business) and the signal/message conveyers (or merchants in commerce or business), or between people themselves [12]. Companies may often encourage clients to share experiences to create a communication platform that is driven by interaction, which can, in turn, be facilitated through social networking sites. Gefen and Straub [13] emphasized that social presence comprises several dimensions, including human warmth, sociability, personalization, and a feeling of human contact experienced during interactions with other individuals, products, or the environment.

A comprehensive analysis of research on AI-based chatbots revealed that prior studies had not fully applied Social Presence Theory [2,14]. To address this limitation, social presence was integrated into the conceptual framework proposed by Prakash et al. [2]. However, they utilized Social Response Theory instead of Social Presence Theory to explain social presence phenomena in AI-based chatbot use. In contrast, our study suggests that it is more appropriate to apply Social Presence Theory to the interactive environment and context of AI-based chatbots.

2.2. Technology Acceptance Theory

The Technology Acceptance Theory and its corresponding model have been widely adopted to study information systems and the use of advanced technologies [15,16]. In our study, Technology Acceptance Theory provides an additional theoretical base alongside Social Presence Theory in discussing people’s perceptions of and responses to using AI-based service chatbots. Although it has been adopted in this context [17], there is room for extending the theoretical application when discussing whether people accept the use of chatbots for different purposes.

According to Kesharwani and Bisht [18], two central constructs within the theory of technology acceptance are perceived ease of use (EASE) and perceived usefulness (USEFUL). Perceived usefulness focuses on whether people or users find the focal technologies able to meet their needs, such as information, payment, efficiency, and other add-on benefits [15]. Users will assess whether the technologies are acceptable and helpful when performing various activities. Perceived ease of use assesses the process and outcome of technology use from another perspective. It is an important construct in the e-commerce world, stressing the user-friendly interface, interaction, and convenience of information searching and sharing [19]. The mediated communication channel using service chatbots should be considered highly user-friendly and controllable. USEFUL and EASE are the basic functional attributes of chatbot technologies from the user or demand perspective. In accordance with Zhang et al. [20], the former can include attributes such as perceptions regarding informativeness, interactivity, and personalization, while the latter can relate to perceptions regarding accessibility. Our study adopts both to represent customers’ perceptions of service chatbot functions and attributes.

There have been some systematic reviews of Technology Acceptance Theory and its model over the past decade in understanding various new technologies and the contributing factors that are involved [21,22,23]. For instance, Momani and Jamous [23] summarized three steps in developing the Technology Acceptance Model: adoption, validation, and extension. The first stage witnessed many applications of the theory and the model, while the second stage focused on accurate measurements in different technologies. The third stage aimed to extend the model by introducing new variables and relationships, including attitudes, behavior (e.g., technology use), behavioral intention, adaptability, flexibility, previous technology experience, perceived value, and perceived playfulness [21,24].

Trust, despite being a well-researched construct [25], highly relevant to customer satisfaction in service industries, has not been thoroughly examined in service robot research [2,26]. Trust is established based on the trustee’s performance in meeting the trustor’s expectations [27]. Propensity to trust is an individual difference characteristic reflecting one’s general tendency to trust others [27].

Furthermore, previous review studies have identified limitations, including research scope and methods employed in the literature search and integration. It has been pinpointed that generalizability represents a major shortcoming in many studies they reviewed [28]. Nonetheless, they have listed or implied some future directions for the application and extension of Technology Acceptance Theory. First, a new set of variables (such as trust) can be included to further test the model and the complex relationships in different settings (such as AI-powered service chatbots). These can help validate and extend the model in alignment with future technological development trends. Second, different stakeholders, such as customers and employees, can offer their inputs to the validation and extension of the Technology Acceptance Model. Third, to better understand distinct social and business phenomena related to interactions with emerging technologies such as artificial intelligence, scholars can draw from diverse theoretical perspectives originating in fields such as anthropology, psychology, and sociology. For instance, variables from the theory of planned behavior, such as perceived behavioral control and subjective norms, can be effectively incorporated into the Technology Acceptance Model [24]. Furthermore, social presence and influence suggest that technology users not only assess whether the technologies in question are acceptable but also examine their own experiences and perceptions when interacting with the technologies and interfaces. Similarly, the integration of theoretical applications can be considered a trend for understanding and improving the applications of various technologies. The subsequent section outlines the hypotheses and research framework, highlighting the value of our theoretical integration and the various interesting variables that facilitate theoretical extension.

2.3. Research Framework and Development of Hypotheses

This study integrates both Social Presence Theory [9] and the Technology Acceptance Model [16,20] in discussing the perceptions of customers toward AI chatbots. On one hand, through the application of Technology Acceptance Theory, customers make assessments of the usefulness and ease of use of AI chatbots. In other words, the acceptance level of AI chatbots will be evaluated. On the other hand, in line with Social Presence Theory, customers’ perception of being with others (even if not humans) is assessed by emphasizing social presence when interacting with chatbots. Given the integration of Social Presence Theory and Technology Acceptance Theory, the understanding of mediated communications and interactions with chatbots involves the appraisal of the technological media and the perceived outcome of social presence between humans and technology, or AI chatbots. This enriches the literature by providing a framework for analyzing AI communication from technical and social perspectives.

This study provides further support to the findings of Prakash et al. [2] and highlights the relationship between users’ trust in technology and their personality and consumption characteristics. Some users’ mistrust of technology may extend to aspects including interface, interaction, and explanation [29]. However, in order to form a long-term, sustainable relationship between customers and technology, customers’ trust in AI is an essential foundation [30]. Moreover, trust in technology may vary significantly among people with different experiences and personality traits. Some are open to new experiences, while others are reluctant to accept them [31]. In this study, it is worthwhile to examine customers’ propensity to trust before evaluating outcomes of chatbot use, including their trusting belief and satisfaction [31].

Across various contexts, perceived ease of use influences the overall evaluation of technology use [3,18]. This not only enhances user satisfaction but also builds trust in technology [2]. To better predict the outcomes of trust and satisfaction, we have thus integrated core elements from Social Presence Theory, the Technology Acceptance Model, and the trust model in technology [2] in our framework (Figure 1). In our study, we propose that the greater the ease of use in customers’ perceptions of AI chatbots, the greater the level of service satisfaction as well as trust in technology use. Therefore, the following hypotheses were proposed:

H1: 

Perceived ease of use is positively associated with service satisfaction.

H2: 

Perceived ease of use is positively associated with trusting belief.

According to our preceding discussions, customers’ propensity to trust should be included in the model. It is a necessary determinant of customer satisfaction and trust. Propensity to trust is considered an important prerequisite for shaping trust, highlighting its vital role in the trust-building model [32]. It is also a contributing factor to general service satisfaction. Previous evidence has shown that individuals with high trust propensity have higher consumer satisfaction [32]. This is probably applicable in assessing chatbot use in our study. Therefore, the following hypotheses are developed:

H3: 

Propensity to trust technology is associated with service satisfaction.

H4: 

Propensity to trust technology is associated with trusting belief.

Social presence is considered an influential factor in driving customer satisfaction and trust. Customers’ trust in technology increases when they perceive high-quality communication and interaction [2]. The relationship between social presence and trust has been established in multiple domains [11,33], including chatbot use [34]. More specifically, social presence has been shown to increase customer satisfaction. According to Ogara et al. [35], the perception of social presence during instant messaging interactions increases user satisfaction. Following similar reasoning, it is plausible to extend this relationship to interactions involving chatbots. Drawing upon this reasoning, the hypotheses presented in this study are formulated as follows:

H5: 

Social presence can result in service satisfaction.

H6: 

Social presence can result in trusting belief.

Another technology acceptance-driven determinant in our framework is perceived usefulness. It is appropriate to postulate a positive association between customers’ overall evaluation of chatbot use and perceived usefulness. Within the service and marketing literature, customer trust can generate satisfaction in the loyalty-building process [36]. In our study, if customers trust the chatbot, it is likely that they will be satisfied in the long run. Here, perceived usefulness can help lead to customer satisfaction. For example, Han and Sa [37] applied the Technology Acceptance Model and found that students’ perceived usefulness of the technology directly influenced their satisfaction with online classes. Additionally, it has been shown that a relationship between customer trust and perceived usefulness exists [38]. In addition, customers’ trust in the medium (e.g., technology use) can be attributed to their perceived usefulness [39]. Given the previous discussions, the relevant hypotheses were developed:

H7: 

Trusting belief will result in service satisfaction.

H8: 

Perceived usefulness will result in service satisfaction.

H9: 

Perceived usefulness will result in trusting belief.

3. Materials and Methods

3.1. Methodology

In 2024, individuals in Hong Kong with experience in chatbot use were selected via purposive sampling to participate in a quantitative survey. Participants provided their informed consent prior to taking part in the study. To screen out ineligible participants, a qualifying question about prior chatbot usage in customer service contexts over the past three months was included. Quota sampling was employed to enhance sample representativeness, with age and sex as criteria. Approximately 60% of respondents were female and aged 18–40, 30% were 41–60, and 10% were over 60. A total of 200 people were approached, and finally, 158 responses were collected. The response rate was 79%. They were given a USD 6 supermarket coupon as an incentive in order to compensate their time and effort doing the questionnaire. The effect of using an incentive is to encourage respondents to complete the questionnaire responsibly. The amount is small relative to the local living standard, so it would not create any extra social desirability effect. Constructs from validated scales were included and measured. The constructs of perceived ease of use, perceived usefulness, and social presence were adapted from Gefen and Straub [40], whereas the constructs of trusting belief and propensity to trust technology originated from Lankton et al. [41]. Additionally, the customer satisfaction construct was adapted based on Ashfaq et al. [42]. All items for these constructs were assessed using a seven-point Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree). The established scales were adapted for the customer service context using AI chatbots. The Chinese translation of the questionnaire was listed side by side. The back-translation method was used to ensure the meanings remained consistent.

3.2. Data Collection

This study involved 158 participants. A minimum sample size was determined using the G*Power 3.1.9.7 analysis tool, indicating that at least 89 participants were required based on five predictors, an effect size of 0.15, and a 95 percent alpha level. An effect size of 0.15 was used, representing a small effect.

Table 1. Demographic data.

Category		Frequency	Percentage %
Gender	Male	62	39.2
	Female	96	60.8
Age	18–30	61	38.6
	31–40	31	19.6
	41–50	42	26.6
	51–60	9	5.7
	61 or above	15	9.5
Organization size	Less than 5	6	3.8
	5–20	16	10.1
	21–50	16	10.1
	51–100	12	7.6
	101 or above	68	43.0
	N/A	40	25.3
Industry	Career break	8	5.1
	Civil servant	8	5.1
	Cultural and creative	6	3.8
	Financial services	5	3.2
	Professionals	53	33.5
	Retailing and customer service	15	9.5
	Retired	29	18.3
	Student	22	13.9
	Tourism	9	5.7
	Trading & Logistics	3	1.9

provides a summary of the sample’s demographic characteristics. A total of 39.2% of the sample were men, while 60.8% were women. Regarding age, 26.6% of respondents were aged 41–50, 38.6% aged 18–30, 5.7% aged 51–60, 19.6% aged 31–40, and 9.5% were above 61. A total of 43% of the respondents worked in large companies (>100 employees); 33.5% were professionals, 9.5% worked in retailing and customer service, 5.1% were civil servants, 5.7% worked in the tourism industry, 3.8% worked in the cultural and creative industry, 1.9% worked in the trading and logistics industry, and 3.2% worked in the financial services sector; 18.3% were retired and 5.1% on a career break. The demographic composition of variables such as age and gender was generally consistent with that reported by the Census and Statistics Department [43] of the Hong Kong government.

3.3. Analysis Method

Partial least squares structural equation modeling (PLS-SEM) was used in the analysis because of the predictive nature of our model. It has been widely used in satisfaction models and technology acceptance frameworks. Works are mainly related to ‘consumer behavior’ and ‘trust’ [44]. IPMA has been used with structural equation modeling (SEM) to examine the effects of chatbots’ attributes on customer relationships [45]. PLS-SEM provides sufficient evidence on independent variables. However, NCA has not yet been employed to explore necessary conditions. NCA provides evidence for the necessity logic. IPMA was used to assess variables’ importance and performance [46]. This research employed NCA and SEM to achieve a holistic understanding. Specifically, IPMA was further incorporated to provide an advanced approach to the decision-making process, termed combined importance-performance map analysis (CIPMA) [46].

4. Results

All indicator loadings exceeded the 0.708 threshold for Cronbach’s alpha, showing good internal consistency reliability. Good convergent validity was supported by Average Variance Extracted (AVE) values > 0.50. The composite reliability (rho a) was used, as it reflects the construct’s internal consistency well (

Table 2. Validity and reliability of measurement model.

Scale and Item	Average Variance Extracted	Composite Reliability (rho a)	Cronbach’s Alpha
Perceived usefulness (USEFUL)	0.872	0.931	0.927
Perceived ease of use (EASE)	0.802	0.905	0.878
Social presence (SOCIAL)	0.834	0.938	0.934
Propensity to trust technology (PROP)	0.808	0.887	0.881
Trusting belief (TRUST)	0.663	0.913	0.898
Service satisfaction (SATISFACTION)	0.910	0.967	0.967

), meaning the constructs have good reliability in our measurement model [47].

Table 3. Discriminant validity of the model.

	Mean	Std	East	Prop	SatisFaction	Social	Trust	Useful
Perceived ease of use (EASE)	4.395	1.326
Propensity to trust technology (PROP)	4.377	1.245	0.426
Service satisfaction (SATISFACTION)	3.829	1.396	0.747	0.381
Social presence (SOCIAL)	3.066	1.412	0.599	0.318	0.607
Trust belief (TRUST)	4.057	1.162	0.700	0.496	0.705	0.576
Perceived usefulness (USEFUL)	4.083	1.459	0.766	0.321	0.762	0.585	0.665

shows that constructs of our measurement model have good discriminant validity (with all values below the 0.85 threshold).

The hypotheses testing results are shown in

Table 4. Hypotheses testing results.

Path	Hypothesis	Path Coefficient	t-Statistic	p-Value	LCI (2.5%)	UCI (97.5%)	Support
EASE  -> SATISFACTION	H1	0.249	2.879	0.004 ***	0.078	0.420	Yes
EASE -> TRUST	H2	0.264	3.791	<0.001 ***	0.126	0.405	Yes
PROP  -> SATISFACTION	H3	0.033	0.499	0.618	−0.097	0.163	No
PROP  -> TRUST	H4	0.185	3.114	0.002 **	0.070	0.305	Yes
SOCIAL -> SATISFACTION	H5	0.141	2.045	0.041 *	0.008	0.280	Yes
SOCIAL -> TRUST	H6	0.139	2.255	0.024 *	0.017	0.264	Yes
TRUST -> SATISFACTION	H7	0.255	2.925	0.003 **	0.081	0.424	Yes
USEFUL -> SATISFACTION	H8	0.323	4.175	<0.001 ***	0.166	0.470	Yes
USEFUL -> TRUST	H9	0.210	2.992	0.003 **	0.067	0.343	Yes

Note: p < 0.001 ***; p < 0.01 **; p < 0.05 *.

:

In Table 4, the hypotheses testing results indicate that eight out of nine hypotheses were supported. Hypothesis three, which proposed an association between perceived propensity to trust and satisfaction, was not supported. The confidence interval contains zero. The effect sizes for all supported hypotheses were small (

Table 5. Effect sizes.

Path	Hypothesis	Path Coefficient	Effect Size (f2)	Effect
EASE -> SATISFACTION	H1	0.249	0.064	Small
EASE  -> TRUST	H2	0.264	0.088	Small
PROP  -> SATISFACTION	H3	0.033	0.002	N/A
PROP -> TRUST	H4	0.185	0.072	Small
SOCIAL -> SATISFACTION	H5	0.141	0.035	Small
SOCIAL -> TRUST	H6	0.139	0.040	Small
TRUST -> SATISFACTION	H7	0.255	0.059	Small
USEFUL -> SATISFACTION	H8	0.323	0.141	Small
USEFUL -> TRUST	H9	0.210	0.071	Small

). Figure 2 shows the structural model. The R-square values for trusting belief and service satisfaction are 54.1% and 64.6%, respectively. Our model has moderate explanatory power. The f-squared effect sizes ranged from 0.035 to 0.141 (except for the unsupported results), i.e., showing small to medium effect sizes on service satisfaction. The Q-squared values ranged from 0.523 to 0.561, indicating large predictive relevance for the dependent construct—service satisfaction. Finally, the collinearity statistics (VIF) pertaining to the inner model are all less than 3 (

Table 6. Collinearity statistics (VIF)–inner model.

Path	VIF
EASE -> SATISFACTION	2.462
EASE -> TRUST	2.264
PROP -> SATISFACTION	1.266
PROP -> TRUST	1.181
SOCIAL -> SATISFACTION	1.626
SOCIAL -> TRUST	1.563
TRUST -> SATISFACTION	2.178
USEFUL -> SATISFACTION	2.287
USEFUL -> TRUST	2.135

).

The structural model is presented below (Figure 2).

Hypothesis testing results indicated that all the independent variables were sufficient to affect service satisfaction, except for propensity to trust technology, which can be attributed to the mediation effect of trusting belief. Propensity to trust technology affects service satisfaction via trusting belief. Willingness to rely on technology cannot establish service satisfaction; trust in technology must be established first (Figure 3). The direct effect is 0.033, which is not significant. The indirect effects are 0.141 and 0.255 (Figure 2), resulting in a total indirect effect of 0.036. The total effect is 0.069.

Since all the outer weights of the model are positive, we can continue our analysis by following step five from the guideline provided by Richter et al. [48]. Using the standard importance-performance map, performance data for the antecedent constructs can be obtained (Figure 4).

Importance data have been captured by the total effect, which refers to the combined value of indirect and direct effects. The indicator data contributes to the latent variable scores and performance [49]. The importance of perceived usefulness is much higher than that of perceived ease of use. The performance of social presence is lower than that of other variables (Figure 5).

Upon completion of IPMA, we proceed to step five using latent variable scores. Following steps six to eight of the procedure, NCA was conducted. By checking the bottleneck table (

Table 7. Bottleneck table for service satisfaction.

	LV Scores—SATISFACTION	LV Scores—EASE	LV Scores—PROP	LV Scores—SOCIAL	LV Scores—TRUST	LV Scores—USEFUL
0.000%	0.000	NN	NN	NN	NN	NN
5.000%	5.000	NN	5.347	NN	NN	NN
10.000%	10.000	NN	5.347	NN	NN	NN
15.000%	15.000	NN	5.347	NN	NN	NN
20.000%	20.000	4.868	10.695	NN	11.340	5.968
25.000%	25.000	16.667	10.695	NN	11.340	5.968
30.000%	30.000	16.667	10.695	NN	11.340	16.667
35.000%	35.000	16.667	10.695	NN	11.340	16.667
40.000%	40.000	26.984	10.695	NN	11.340	16.667
45.000%	45.000	26.984	10.695	NN	11.340	16.667
50.000%	50.000	26.984	10.695	NN	11.340	16.667
55.000%	55.000	26.984	10.695	NN	11.340	16.667
60.000%	60.000	30.951	18.253	NN	11.340	16.667
65.000%	65.000	30.951	28.098	NN	24.558	16.667
70.000%	70.000	50.000	28.098	NN	24.558	16.667
75.000%	75.000	53.967	28.098	NN	24.558	16.667
80.000%	80.000	53.967	28.098	NN	24.558	16.667
85.000%	85.000	53.967	28.098	30.185	24.558	50.000
90.000%	90.000	60.898	50.000	30.185	44.331	55.655
95.000%	95.000	60.898	50.000	47.612	44.331	55.655
100.000%	100.000	81.852	50.000	50.199	56.051	55.655

) using a satisfaction outcome level of 85, the percentage of cases that do not meet the necessary condition can be obtained. Finally, all the necessary effect sizes are larger than 0, and the p-values are smaller than 0.05 (

Table 8. cIPMA results.

Antecedent Construct	Importance	Performance	Percentage of Cases That Do Not Meet the Necessary Condition *	Necessity Effect Size d (p Value)
Perceived usefulness	0.376	51.391	39.873	0.197 (0.000)
Perceived ease of use	0.317	56.586	33.544	0.302 (0.000)
Trusting belief	0.25	50.948	8.228	0.174 (0.000)
Social presence	0.176	34.434	47.468	0.068 (0.021)
Propensity to trust technology	0.08	56.29	9.494	0.194 (0.001)

* based on a satisfaction outcome level of 85.

).

The desired level of service satisfaction is set at 85 out of 100, based on the common understanding of what constitutes an acceptable service satisfaction result [46]. The combined IPMA was prepared using the desired level of service satisfaction (Figure 6). All the independent variables are white circles, indicating they are the necessary conditions for the desired level. Perceived usefulness of chatbots has the highest importance rating, followed by ease of use of chatbots. Propensity to trust technology has the least importance. In terms of performance, perceived ease of use has the highest performance rating, followed by propensity to trust technology. There are cases that do not meet the necessary conditions, especially for perceived ease of use and perceived usefulness (larger white circles), which belong to the prioritized action areas. In the lower left-hand corner of the combined IPMA map (Figure 6), social presence also appears as a prioritized action area due to unmet conditions, indicated by a large white circle and lower performance.

To achieve the desired level of service satisfaction of 85%, the corresponding levels are as follows: perceived ease of use at 54%, propensity to trust technology at 28%, social presence at 30%, trusting belief at 25%, and perceived usefulness at 50%, according to the bottleneck table. However, for a desired level of service satisfaction of 80% or lower, social presence is not necessary (Table 7).

5. Discussion

By incorporating propensity to trust technology, social presence, and trusting belief into the Technology Acceptance Model, our study enriches current literature in the context of AI chatbot usage for customer service.

Instead of behavioral intention as an outcome variable, service satisfaction was selected in the customer service context. The modified Technology Acceptance Model was supported by our empirical data. Higher trust propensity is associated with higher consumer satisfaction [32]. However, the association between propensity to trust technology and consumer satisfaction was not supported in our study. Instead, the linkage between propensity to trust technology and consumer satisfaction is through trusting belief. It is more logical to the researchers because trust is essential when people face a new technological environment.

From the PLS-SEM structural model results, trusting belief, perceived usefulness, and perceived ease of use contribute to customer satisfaction with significant path coefficients with larger values compared to social presence. In a competitive marketplace, we all aim for almost perfect service satisfaction, thus social presence is essential. Based on Hauff et al. [46], there are some managerial recommendations. In the upper right-hand corner of the combined IPMA map (Figure 5), the white circles for trusting belief, perceived ease of use, and perceived usefulness indicate prioritized action areas.

In high service satisfaction standards (e.g., 85%), perceived ease of use, propensity to trust, social presence, trusting belief, and perceived usefulness are necessary. In contrast, at medium service satisfaction standards (e.g., 80% or below), social presence is not necessary. This means social presence appears in the model for high service satisfaction standards, but not for medium or low standards.

6. Conclusions

Within the customer service context, factors such as perceived usefulness, perceived ease of use, social presence, and an individual’s propensity to trust technology collectively influence users’ trusting beliefs toward chatbots. Additionally, customer satisfaction with chatbot-based services is influenced by perceived ease of use and usefulness, as well as by social presence and trusting beliefs. To achieve the 85% service satisfaction standard, common in the service industry, all independent variables are necessary. The combined IPMA map identifies prioritized action areas as social presence, perceived usefulness, and perceived ease of use.

Practical measures are needed to improve chatbots’ perceived usefulness, ease of use, and social presence. Humor could enhance chatbots’ social presence. More functions with friendly interfaces could improve perceived usefulness. Showing suggested alternative questions may increase perceived ease of use by reducing users’ need to type exact questions.

Future research could extend from AI chatbots to robots. As this is a cross-sectional study, longitudinal research would better capture service satisfaction and its antecedents. Comparing AI chatbots with human agents in customer service contexts presents another research opportunity.