Destination Competitiveness Improvement: Insights From Causal Counterfactual AI Analysis

Abstract

Previous methods for destination competitiveness improvement have mainly focused on identifying and prioritizing competitive disadvantages of destinations. Although effective, this approach may not be optimal as it may require more change than improving combinations of other competitive disadvantages. Furthermore, these methods neglect the differing foci of travel experiences between tourist groups and have been unable to identify targeted competitiveness improvement strategies for different tourist groups. This study addresses these research gaps by developing an analytical framework that can identify targeted strategies that entail minimal changes to improve the competitiveness of destinations for different tourist groups, based on user-generated data, aspect-level sentiment analysis, and the optimization-based causal counterfactual Al algorithm. The application of the framework is demonstrated through a case study involving four destinations in Australia. The proposed analytical framework and findings are valuable in assisting destinations to improve their competitiveness in today’s increasingly competitive experiential tourism market.

Keywords

causal counterfactual AI algorithm user-generated data aspect-level sentiment analysis decision analytics destination competitiveness improvement

Introduction

Achieving a high level of competitiveness is important for tourist destinations to obtain a favorable market position (Abreu-Novais et al., 2016; Dwyer & Kim, 2003), which contributes not only to the economic development but also to the social and environmental development of tourist destinations (Ritchie & Crouch, 2003). Given its importance, a considerable amount of academic effort has been devoted to understanding the competitive position of destinations around the world and exploring knowledge that may help destinations improve their competitiveness (Crouch, 2011; Y. R. Kim et al., 2022). These existing studies typically identify potential directions for destinations to improve their competitiveness by applying established competitiveness evaluation models (e.g., Crouch & Ritchie, 1999) to capture competitive disadvantages (Andrades & Dimanche, 2017; Ayikoru, 2015; Mustafa et al., 2020). Due to constraints such as resource limitations and policy restrictions, destinations are often unable to address all competitive disadvantages simultaneously. Consequently, some initial studies (e.g., Cracolici & Nijkamp, 2009; Sánchez & López, 2015) have explored methods to prioritize competitive disadvantages (e.g., importance performance analysis). In this way, destination managers can focus on addressing the most pressing disadvantages to improve their destinations’ competitiveness.

Although prioritizing competitive disadvantages is an important extension to applying established competitiveness evaluation models to guide destinations in making informed decisions to improve their competitiveness, there are still some limitations in these existing approaches that deserve further investigation. Firstly, focusing on the most concerning competitive disadvantages may not always be optimal, as addressing them may require more changes than addressing a combination of other disadvantages to achieve the desired level of competitiveness. Suppose that the competitive disadvantages of a destination are “ticket prices,”“transport” and “accommodation” experiences, and the disadvantage that tourists are most concerned about is the “accommodation” experience. Improving the “accommodation” experience (e.g., refurbishing all the hotels) may require substantially more changes than simultaneously enhancing the “transport” experience and “ticket prices” (e.g., providing free shuttle buses and reducing ticket prices) or simultaneously ameliorating the experiences of “accommodation” and “transport” (e.g., refurbishing a small number of hotels and providing free shuttle buses) to raise the destination’s competitiveness to the desired level. To the best of our knowledge, no efforts have been made to help destinations identify optimal strategies that entail minimal changes to improve their competitiveness to a desired level, likely due to the absence of an available method for completing this task.

Second, existing studies typically assume that tourists in a destination are homogeneous, and thereby focus on identifying the same set of changes (e.g., providing free shuttle buses) to improve the competitiveness of a destination for all tourists (Ayikoru, 2015; Mustafa et al., 2020). However, the changes necessary to improve the competitiveness of a destination may vary for different tourist groups (i.e., segments), as their focus on the travel experience differs (Neto et al., 2020). Changes to the experience dimension that are irrelevant to a particular tourist group will not contribute to improving the competitiveness of a destination for that group. Suppose that tourist group A focuses more on the “transport” experience and tourist group B focuses more on the “accommodation” experience. Changes in “accommodation” will increase the competitiveness of the destination for tourist group B, but not for tourist group A, as they value the “transport” experience more. So far, there is no method available to help managers develop such targeted strategies to improve the competitiveness of destinations for different tourist groups.

To address these limitations and fill the research gap regarding optimal (i.e., minimal change in competitive disadvantages) and targeted strategy formulation for improving destination competitiveness, this study proposes framing the improvement of destination competitiveness as an optimization problem. We introduce a destination competitiveness improvement strategy identification framework drawn from aspect-level sentiment analysis, k-means clustering, optimization-based AI methods and online user-generated data. The foundation of this analytical framework (i.e., method) is the causal counterfactual AI algorithm (Xia et al., 2023), which identifies the strategy that entails the most minimal change in tourist touchpoint (De Keyser et al., 2020) and experience dimensions to improve a destination’s competitiveness for different tourist groups by solving an elaborate counterfactual optimization problem. The questions this study aims to answer through the proposed analytical framework are: (1) What minimal change strategies could a destination adopt to improve its competitiveness for different groups of tourists to a desired level? (2) What is the required change for each item in the identified strategy? (3) What is the required total change in each dimension of a destination to improve its competitiveness? The application of the framework is demonstrated through a case study involving four tourist destinations in Australia. The results provide detailed insights that can help the studied destinations make informed decisions to improve their competitiveness in today’s increasingly competitive and complex experiential tourism market.

Literature Review

This section first briefly reviews the definition of destination competitiveness and its evaluation methods, which form the theoretical basis of this study. It then reviews the literature on destination competitiveness improvement and the causal counterfactual AI-based strategy formulation that underpins the methodology of this study.

Destination Competitiveness Evaluation

Destination competitiveness is a multifaceted concept that has been conceptualized and evaluated differently in the literature (Abreu-Novais et al., 2016; Mazanec et al., 2007). Some researchers define destination competitiveness from a supply perspective, relying on expert evaluations of determinants from established models (e.g., Crouch & Ritchie, 1999; Ritchie & Crouch, 2003) or economic indicators to evaluate destination competitiveness (e.g., Croes, 2011; Song et al., 2023). Other researchers define destination competitiveness from a demand perspective and rely on tourist evaluations (e.g., Andrades-Caldito et al., 2014; Neto et al., 2020), or tourist satisfaction (Song et al., 2012) to appraise it.

This study follows the demand perspective, arguing that destination competitiveness can be measured based on tourists’ satisfaction at key touchpoints—the interactions between tourists and destinations—that are important to their travel experience (Stein & Ramaseshan, 2016). We adopt this demand perspective, with an emphasis on tourists’ satisfaction at touchpoints that are important to their experience, because the tourist experience is the only product in the tourism industry and there has been a consistent call for more research from this perspective (Neto et al., 2020; Shoval & Birenboim, 2019; Xia et al., 2024). The theoretical underpinning of our method is touchpoint theory in the customer experience management literature (Holmlund et al., 2020), which conceptualizes and evaluates customer experience through customer satisfaction with direct or indirect interactions (i.e., touchpoints) between customers and organizations (De Keyser et al., 2020). As positive customer experiences lead to positive consumer behavior such as revisiting and loyalty (Srivastava & Kaul, 2016), this research argues that touchpoints with high tourist satisfaction, which indicates positive experiences, can then represent the competitive advantages of a destination and vice versa (H. Kim & Fung So, 2024).

Destination Competitiveness Improvement

Numerous empirical studies have explored ways in which destinations can improve their competitiveness. A typical approach in these studies is applying established competitiveness evaluation models (Dwyer & Kim, 2003; Ritchie & Crouch, 2003) to identify competitive disadvantages that destinations can improve (Gomezelj & Mihalič, 2008). The knowledge identified in this line of research usually includes a set of disadvantages that a destination needs to change to improve its competitiveness, such as foreign direct investment, human resource development, and tourism information dissemination (Chens et al., 2008).

As destinations are often unable to improve all their competitive disadvantages simultaneously due to resource and other constraints, studies therefore go a step further to identify the priority of competitive disadvantages so that destinations can focus on improving the most concerning (i.e., important) disadvantages (Albayrak et al., 2018; Enright & Newton, 2004, 2005). Several methods have been used to identify the priority of competitive disadvantages, such as interviewing stakeholders (Ayikoru, 2015; Azzopardi & Nash, 2016), calculating importance coefficients (Cracolici & Nijkamp, 2009; Sánchez & López, 2015) and applying importance performance analysis (Albayrak et al., 2018; Mustafa et al., 2020). Among them, applying importance performance analysis is the most common. The method provides a prioritization of competitive disadvantages by plotting their performance and importance on a two-dimensional grid (Abreu-Novais et al., 2016).

Focusing on the most concerning disadvantages is undoubtedly a straightforward strategy for destinations to improve their competitiveness within practical constraints. However, this may not always be the most optimal approach. Addressing the most concerning disadvantages could require substantially more changes than improving multiple disadvantages collectively to achieve the desired level of improvement in destination competitiveness. Additionally, targeting only the most concerning disadvantages may be insufficient due to complex interactions between competitive disadvantages.

Moreover, many existing methods implicitly assume that tourists are homogeneous, suggesting that the same set of changes can improve the competitiveness of a destination for all tourists (e.g., Mustafa et al., 2020). However, as indicated in recent literature, destination competitiveness is a “dynamic concept” and different tourist groups (i.e., segments) have varying perceptions about the competitiveness of a destination (Neto et al. 2017, 2020). This means that targeted strategies for different tourist groups are required to effectively improve the competitiveness of a destination for them, as improvement strategies that are effective for one tourist group may not be effective for another (Kumar & Dhir, 2020). To overcome these limitations, this study develops an analytical framework that identifies strategies requiring minimum changes to improve a destination’s competitiveness for various groups of tourists by formulating an elaborate counterfactual optimization problem. The foundation of this framework is the causal counterfactual AI algorithm, which is detailed in the following subsection.

Counterfactuals and Related AI Algorithms for Strategy Formulation

Counterfactuals refer to hypothetical scenarios that explore what would have happened if certain factors had been altered, and they represent a form of causal judgment technique commonly used in everyday human life (Pearl & Mackenzie, 2018). This approach to causal judgment is concluded as causal counterfactual theory (Hannart et al., 2016), which states that an event Y is causally dependent on another event X, if and only if X and Y occur simultaneously in the actual and counterfactual scenarios. If X had not occurred in one scenario, Y would not have occurred either (Lewis, 1973). Causal counterfactual theory holds that the cause of an event always produces different outcomes from what would happen in its absence (Lewis, 1973). For example, if a tourist group’s dissatisfaction stems from an old car park, a scenario with new car parks might meet their expectations. However, if there is no substantial improvement in their experience, then the old car park is not the cause of their dissatisfaction.

This causal judgment approach has been widely used in management research to support strategy formulation because managerial decisions often involve causal reasoning (Durand & Vaara, 2009; McGill, 2000; T. Yang, 2020). A major challenge in counterfactual-based strategy formulation is constructing counterfactuals for complex events, which often have numerous potential causes (e.g., low competitiveness). The interaction between these causes requires the construction of many counterfactuals (i.e., combinations of causes) for potential outcome comparisons (Xia et al., 2023). However, this process is labor-intensive and often impractical for humans, which hinders the use of this strategy formulation approach for complex events. Recently proposed causal counterfactual AI algorithms (Sharma & Kiciman, 2020; Verma et al., 2020) have addressed this challenge by transferring the complexity of counterfactual construction and outcome estimation from humans to algorithms through constructing optimization problems (Huenermund et al., 2022), providing efficient decision-making tools for tourism research (Xia et al., 2023).

Compared to existing methods for destination competitiveness improvement (Albayrak et al., 2018; Enright & Newton, 2004), the causal counterfactual AI algorithm is more cost-effective because it can be applied to easily collected tourism big data and does not require implementation of changes, which usually require extensive resources (Xia et al., 2023). More importantly, the causal counterfactual AI algorithm is able to identify the minimal changes that may cause different outcomes from reality, thus enabling destination managers to identify optimal strategies (i.e., strategies entail minimal changes in tourist touchpoints and experience dimensions) to improve the competitiveness of their destinations to a desired level, which existing methods that focus on identifying all or the most concerning disadvantages cannot achieve.

Methodology

This section details the analytical framework developed to identify the most optimal and targeted strategies to help destinations improve their competitiveness for different groups of tourists based on user-generated data and AI methods (Figure 1). The framework consists of three steps: (1) user-generated data collection, where user-generated data in the form of tourist reviews and ratings are collected; (2) destination competitiveness analysis, where aspect-level sentiment analysis and k-means clustering are used to quantify the competitiveness of the destination and the experiences of different groups of tourists at key touchpoints important to their travel experience; (3) counterfactual-based competitiveness improvement strategy identification, where targeted strategies entailing minimum changes that destinations can adopt to improve their competitiveness for specific groups of tourists are identified through the causal counterfactual AI algorithm and qualitative coding. The detailed processes involved in these steps are visualized in Appendix 1.

Figure 1.

Overview of the proposed competitiveness improvement framework.

User Generated Data Collection

In recent years, user-generated data, in the form of online reviews and ratings, have emerged as an important source of tourist experience big data in tourism research due to advantages such as large-scale availability and non-intrusiveness to human participants (Xia et al., 2023). Examples of platforms containing such user-generated data include TripAdvisor (https://www.tripadvisor.com/), Yelp (https://www.yelp.com/), Agoda (https://www.agoda.com/), and Google Maps (https://www.google.com/maps/). In order to collect tourist reviews and ratings from these online platforms to conduct destination competitiveness analysis, our first step is to identify the tourism service providers (e.g., hotels and attractions) within the target destinations. This can be done by using the search or location features of these platforms. After that, the review page of each tourism service provider is browsed to collect all tourists’ reviews, ratings and overall service ratings. Tourism service providers that do not have this information are excluded. Only English reviews are collected due to their predominant use in Google Maps, and reviews with fewer than 10 words are excluded for insufficient information. Finally, we assign a label to each tourist to indicate their overall satisfaction with their travel experience in the destination based on their ratings. Tourists who give ratings lower than the overall service rating of the provider are labeled “0,” indicating dissatisfaction; otherwise, they are labeled “1,” indicating satisfaction.

Destination Competitiveness Analysis

The objective of destination competitiveness analysis is to accurately measure the satisfaction of different groups of tourists at key touchpoints that are integral to their experience and the competitiveness of destinations at these touchpoints. This involves identifying these key touchpoints and quantifying tourist satisfaction in response to them. The techniques of aspect-level sentiment analysis (Li et al., 2023) and k-means clustering (Frochot, 2005) are employed. Aspect-level sentiment analysis is a type of fine-grained sentiment analysis technique that aims to extract the aspects discussed in a sentence and identify the sentiment toward those aspects (H. Yang et al., 2021). It has been confirmed as an effective technique for capturing tourists’ perceptions of touchpoints that are important to their travel experience (Gao et al., 2024). For an example review, “The staff is friendly, but the car park is too small!”, aspect-level sentiment analysis focuses on extracting the aspects (touchpoints) of “staff” and “car park” discussed by tourists and identifying the sentiment (i.e., positive, negative or neutral) toward these two aspects (touchpoints), namely, “friendly” as “positive” and “too small” as “negative.”

There are several open-source toolkits for aspect-level sentiment analysis. We chose the one provided by H. Yang et al. (2021) due to its strong performance and thorough documentation.¹ Once the aspects in all tourists’ reviews and the sentiments toward these aspects have been extracted by the aspect-level sentiment analysis technique, a post-processing process is applied to identify the key touchpoints that tourists value in their experiences at a destination. First, all aspects are converted to lowercase, typographical errors and sentiments of aspects are checked, and aspects in the form of stop words such as “is” and “the” are filtered out. Second, part-of-speech tagging is performed to filter out non-noun aspects and leave noun aspects as the touchpoints (and their sentiments) for subsequent analysis. Finally, a user-specified touchpoint frequency threshold $ε$ is introduced to identify key touchpoints. Specifically, the frequency of each unique touchpoint is counted and touchpoints with a frequency greater than $ε$ are considered key touchpoints, as a high frequency indicates that these touchpoints are valued by a large number of tourists and can therefore be used to describe tourists’ experiences at a destination. Inspired by the measure used in aspect-aware recommendation systems to quantify user satisfaction for different aspects of a product (J. Tan et al., 2021), a normalized group touchpoint opinion score is proposed below to quantify the experiences of tourists and the competitiveness of destinations in terms of these key touchpoints.

Let $T = {t_{1}, t_{2}, . . ., t_{n}}$ and $C = {c_{1}, c_{2}, . . ., c_{n}}$ represent $n$ unique key touchpoints and the frequency of these touchpoints extracted by the aspect-level sentiment analysis technique from a given text corpus (i.e., a specific tourist or all tourists’ reviews), respectively. $Φ (t_{i})$ represents a function that denotes the sentiment of a touchpoint $t_{i}$ , with values ${1 : positive, 0 : neutral, - 1 : negative}$ . For each of the key touchpoints $t_{i}$ , its normalized group touchpoint opinion score can be calculated as:

\begin{array}{l} S (t_{i}, c_{i}) = φ (W_{d_{g}} * \sum_{1}^{c_{i}} Φ (t_{i}))) \\ = φ (\frac{c_{i}}{n} * \sum_{1}^{c_{i}} Φ (t_{i})), \end{array}

(1)

where $W_{g} = \frac{c_{i}}{n}$ is a weight that indicates the intensity of tourists’ attention to touchpoint $t_{i}$ , with a higher value signifying a greater focus on touchpoint $t_{i}$ . $φ (\cdot) = \frac{1}{1 + e^{- (\cdot)}}$ is the sigmoid function that is used to scale the normalized group touchpoint opinion score within the range of (0, 1).

A higher normalized group touchpoint opinion score (close to 1) at the touchpoint $t_{i}$ indicates greater tourist focus and satisfaction, constituting a competitive advantage. Conversely, a lower normalized group touchpoint opinion score (close to 0) indicates greater tourist focus and dissatisfaction and represents a competitive disadvantage; 0.5 serves as the demarcation threshold. Based on the reviews of each tourist and the reviews of all tourists in a destination, the experience of each tourist $e^{i}$ and the competitiveness of the destination $d^{k}$ can be individually quantified as vectors $e^{i} = [S (t_{1,} c_{1}^{i}), S (t_{2,} c_{2}^{i}), \dots, S (t_{n,} c_{n}^{i})]$ and $d^{k} = [S (t_{1,} c_{1}^{k}), S (t_{2,} c_{1}^{k}), \dots, S (t_{n,} c_{n}^{k})]$ , according to Equation 1.

After this quantification, the most widely used algorithm in the segmentation literature—k-means—is used to identify tourist groups according to their experience vector $e^{i}$ (Dolnicar et al., 2014). Touchpoints not mentioned by a tourist are imputed with the average normalized group touchpoint opinion score of the other tourists who mentioned these touchpoints (i.e., mean imputation; Donders et al., 2006) before clustering, to avoid biasing the results by missing values. By conducting k-means clustering, tourists with similar experience vectors, namely, satisfaction at touchpoints of their travel experience, are clustered together. The center of each cluster $g^{j} = c - kmeans (e^{1}, \dots e^{t})$ is used to represent the experiences of tourists in respective groups. Intuitively, $g^{j}$ can be considered as the vector representation of the experience of the most representative tourist (also known as the prototype; P. N. Tan et al., 2023) in each of the tourist groups.

Counterfactual-Based Competitiveness Improvement Strategy Identification

This step identifies the optimal strategies entailing minimal changes that destinations could adopt to improve their competitiveness for different groups of tourists, based on quantified experiences ( $g^{j}$ ) and destination competitiveness ( $d^{k}$ ), through the causal counterfactual AI algorithm (J. Tan et al., 2021). In the causal counterfactual AI algorithm, strategy formulation is achieved by solving a counterfactual optimization problem with the goal of finding the optimal strategy to achieve a given outcome (Xia et al., 2023). In our study, the outcome is to effectively improve the unsatisfactory experience of a specific group of tourists at key touchpoints in order to increase the competitiveness of a destination for them—and optimal means minimal change.

Let $y_{i, k} = f (e^{i}, d^{k} | Θ)$ be a neural network that predicts whether or not tourists are satisfied with their experience in a destination, where $Θ$ is the network parameter, and $y_{i, k}$ is the output of the neural network, which equals 1 if tourists are satisfied with their overall experience or 0 if tourists are dissatisfied with their overall experience. $e^{i}$ and $d^{k}$ are the tourist experience vector and the destination competitiveness vector, respectively. $g^{j}$ is a subtype of $e^{i}$ , as it represents the experience of the representative tourist in each group. The objective of the counterfactual optimization in this study is to identify the minimal change strategy $Δ_{j}$ for a group of tourists who had unsatisfactory experiences (i.e., $y_{j, k} = 0$ ). Such that when $Δ_{j}$ is applied to $d^{k}$ —namely, the destination implements this strategy (in a hypothetical counterfactual scenario) and improves the experiences delivered at the suggested touchpoints in $Δ_{j}$ —they will have a satisfying experience (i.e., $y_{j, k} = 1$ ). Formally, this optimization problem can be expressed as:

\begin{matrix} minimise C (Δ) = | | Δ | |_{2}^{2} + β | | Δ | |_{0} \\ s . t ., f (g^{j}, d^{k} + Δ | Θ) = 1 if f (g^{i}, d^{k} | Θ) = 0 \end{matrix}

(2)

where $C (Δ)$ denotes the minimal change strategy that minimizes the total change in the current competitiveness of the destination, as all changes incur costs, quantified as the $l_{2}$ -norm $| | Δ | |_{2}^{2}$ . The minimal change strategy also seeks simplicity by minimizing the change in the number of touchpoints, indicated as the $l_{0}$ -norm $| | Δ | |_{0}$ . An example might be focusing only on either improving staff training or upgrading facilities, rather than doing both. $β$ is a weight to balance between these two criteria.

One challenge in solving the counterfactual optimization problem of Equation 2 is that the $l_{0}$ -norm in the optimization objective is not differentiable. This non-differentiability prevents the use of efficient gradient-based optimizers to solve this optimization problem. Following the practice of the existing literature (J. Tan et al., 2021), this study relaxes the $l_{0}$ -norm to $l_{1}$ -norm $| | Δ | |_{1}$ and models the constraint of Equation 2 as the binary cross entropy loss $BCE (f (g^{j}, d^{k} + Δ | Θ), 1)$ , where $B C E (\hat{f}, f) = - [f * \log (\hat{f}) + (1 - f) * \log (1 - \hat{f})]$ , and 1 is the target label representing a satisfied tourist experience. By incorporating the binary cross entropy loss as a Lagrange term into the objective, the resulting differentiable optimization function is expressed as:

\begin{matrix} minimize \\ Δ \end{matrix} {| | Δ | |}_{2}^{2} + β {| | Δ | |}_{1} + BCE (f (g^{j}, d^{k} + Δ | Θ), 1)

(3)

Considering that not all touchpoints focused by tourists are changeable (e.g., location), a one-hot mask vector $a$ is then introduced into Equation 3 to ensure that the potential strategy includes only changeable touchpoints.² If a touchpoint is changeable, its mask value in $a$ is 1; otherwise, 0. The final optimization function is expressed as:

\begin{matrix} \begin{matrix} minimise \\ Δ \end{matrix} | | a ⊙ Δ | |_{2}^{2} + β | | a ⊙ Δ | |_{1} \\ + BCE (f (g^{j}, d^{k} + a ⊙ Δ | Θ), 1) \end{matrix}

(4)

where ⊙ is the Hadamard product (Horn & Yang, 2020) that eliminates the changes to the unchangeable touchpoints in $Δ$ by multiple zeros in the mask vector $a$ . Solving this optimization problem using the gradient-based optimizer (Kingma & Ba, 2014) yields the optimal strategy $Δ_{j}$ , entailing minimal changes that the destination could adopt to improve the unsatisfactory experience of the tourist group $g^{j}$ . Since multiple counterfactuals can exist for a single event, the solution to the optimization problem in Equation 4 is therefore not unique (Molnar, 2023). Users of our framework can solve this optimization problem multiple times to identify the most appropriate strategy according to prior knowledge or specified criteria. Suppose the number of changeable touchpoints in the identified strategy is r, this strategy can then be represented as $s_{j}^{r} = {δ_{1}^{j}, δ_{2}^{j}, \dots, δ_{r}^{j}}$ , where $δ_{r}^{j}$ is the required change for the r-th touchpoint for this group of tourists.

Recognizing that different tourist touchpoints might address the same dimension of a destination, such as “breakfast” and “lunch” experiences both indicating the food quality of the destination, and changes in these dimensions are informative for managers to obtain an overview of the changes required to improve the competitiveness of their destination. Two of the authors therefore independently coded the dimensions of the identified touchpoints to obtain the changes at the dimension level. Disagreements on codes were discussed and resolved within the authoring team. The cumulative changes across unique touchpoints within the same dimension are summarized as the dimension change.

Case Study: Application in Four Australian Destinations

This section presents a case study to demonstrate the application of the developed analytical framework in formulating strategies entailing minimal changes to improve the competitiveness of destinations for different groups of tourists. The data collection is presented first, followed by the analysis of the competitive advantages and disadvantages of the destinations and the strategies to improve competitiveness.

Data Collection

Our proposed framework can be adapted to any online tourism platform that has user-generated data in the form of tourist reviews and ratings. In this study, we use Google Maps for demonstration purposes due to its global popularity as a review platform, the widespread public availability of reviews and its frequent use in tourism studies (Mathayomchan & Taecharungroj, 2020). Four tourist destinations in Australia were selected for analysis: the Great Ocean Road, Phillip Island Nature Park, Cradle Mountain and Rottnest Island (Bilbytravel, 2024). These destinations were chosen for demonstration because they have a relatively high percentage (over 24%) of tourists reporting low satisfaction with their travel experiences (i.e., labeled 0). This indicates the need for these destinations to formulate strategies to improve their competitiveness in providing tourists with more satisfactory experiences, which is the focus of this study.

The search terms “Great Ocean Road Itinerary,”“Phillip Island Tour,”“Cradle Mountain Tour” and “Rottnest Island Tour” were used in Google Maps to retrieve relevant tourism service providers in these destinations for data collection. Following the method described in Section 3.1, we collected a dataset of 12,532 tourist user-generated data in July of 2023. Details of the number of service providers, the number of data collected, the percentage of dissatisfied tourists and the date range of the data for each destination are shown in Table 1.

Table 1.

Statistics of the Collected User-Generated Data.

Destination	No. of collected tourist user-generated data	% of dissatisfied tourists	No. of service providers	Date range of tourist user-generated data
Great Ocean Road	8,589	24.05%	36	2013–2023
Phillip Island	2,813	35.37%	10	2013–2023
Cradle Mountain	765	30.58%	5	2016-2023
Rottnest Island	365	29.04%	4	2016-2023
Total	12,532	27.13%	55	–

Key Touchpoint Discovery

The collected reviews were analyzed using the process presented in Section 3.1 to identify key touchpoints that are important to the tourist travel experience. Following the aspect-aware recommendation literature (e.g., J. Tan et al., 2021), the touchpoint frequency threshold $ε$ was set at 10 for the dense sub-dataset—Great Ocean Road—and five for the sparse sub-dataset—Phillip Island and Cradle Mountain. For the sparsest dataset—Rottnest Island, the threshold $ε$ is set to three. The word cloud of key touchpoints identified, and their statistical information for each destination, are shown in Table 2. The Great Ocean Road has 111 key touchpoints, with a mean of 1.35 and a standard deviation of 0.93 per review.³ Phillip Island has 53 key touchpoints, with a mean of 1.25 and a standard deviation of 0.81 per review. Cradle Mountain has 35 key touchpoints, with a mean of 1.60 and a standard deviation of 0.94. Rottnest Island has 45 key touchpoints, with a mean of 1.64 and a standard deviation of 0.95. This indicates that our framework is able to identify the most important key touchpoints of tourists in each destination. The high number of key touchpoints identified also confirms the complexity of the tourist experience.

Table 2.

The Identified Key Touchpoints That Comprise Tourists’ Experiences and Their Statistics.

Destination	Total no. of touchpoints	Mean (SD) of touchpoints for each review
Great Ocean Road	111	1.35 (0.93)
Phillip Island	53	1.25 (0.81)
Cradle Mountain	35	1.60 (0.94)
Rottnest Island	45	1.64 (0.95)

The popular key touchpoints that tourists focus on in the Great Ocean Road include “place” (i.e., the “place” aspect extracted from reviews, indicating the place that tourists visit in the destination), “view,”“scenery,” and “beach.” Consistent with the fact that tourists are deeply impressed by the stunning beach views of the Great Ocean Road. For Phillip Island, key touchpoints of “penguin,”“kangaroo,”“koala,”“penguin parade” and “wildlife” stand out. Consistent with the fact that wildlife on Phillip Island, particularly penguins, kangaroos and koalas, attracts substantial attention from tourists (Bilbytravel, 2024). The key touchpoints that stand out are “guide,”“staff,”“shuttle bus” for Cradle Mountain, and “tour guide,”“quokka,”“staff,”“segway,” for Rottnest Island respectively. These touchpoints align with the unique experiences each destination offers to tourists (Kentish, 2024; Wadjemup Rottnest, 2024).

Competitive Advantage and Disadvantage Identification

The normalized group touchpoint opinion score of the identified key touchpoints was calculated to quantify the tourist experience and understand the competitive advantages and disadvantages of the four destinations studied. As shown in Figure 2, the presence of touchpoints with high normalized group touchpoint opinion scores (close to 1) indicates that these destinations have unique competitive advantages that attract tourists. Meanwhile, the presence of touchpoints with low normalized group touchpoint opinion scores (below 0.5) indicates that there is room for improvement in the overall competitiveness of these destinations.

Figure 2.

Normalized group touchpoint opinion score distribution of the four destinations at identified key touchpoints: (a) Great Ocean Road, (b) Phillip Island, (c) Cradle Mountain, and (d) Rottnest Island.

To demonstrate the identified competitive advantages and disadvantages of these four destinations, Table 3 shows the top five and last five touchpoints, ranked by the normalized group touchpoint opinion score for each destination. The most unique competitive advantages of the Great Ocean Road (i.e., the top five touchpoints) include the “view,”“place,”“scenery,”“chocolate” and “staff” touchpoints, reflected in its stunning views and famous chocolate factories (e.g., Great Ocean Road Chocolaterie & Ice Creamery⁴). The most unique competitive disadvantages (i.e., the last five touchpoints) of the Great Ocean Road are the “toilet,”“crowd,”“snake,”“bus” and especially “step” touchpoints. This is probably because the prevalence of steps at Great Ocean Road attractions makes it difficult for some tourists to visit, especially in bad weather when the steps become wet and slippery.

Table 3.

Top Five and Last Five Touchpoints Ranked by Normalized Group Touchpoint Opinion Score for Each Destination.

Destination	Top 5 touchpoints (normalized group touchpoint opinion score)	Last 5 touchpoints (normalized group touchpoint opinion score)
Great Ocean Road	View (1.000)	Step (0.072)
	Place (1.000)	Toilet (0.436)
	Chocolate (1.000)	Crowd (0.463)
	Staff (1.000)	Snake (0.492)
	Scenery (0.999)	Bus (0.497)
Phillip Island	Kangaroo (1.000)	Crowd (0.463)
	Penguin (1.000)	Cost (0.486)
	Place (1.000)	Toilet (0.494)
	Staff (0.998)	Service (0.496)
	Penguin Parade (0.960)	Enclosure (0.497)
Cradle Mountain	Staff (1.000)	Food (0.491)
	Guide (0.999)	Drink (0.496)
	Place (0.987)	Weather (0.498)
	Shuttle Bus (0.908)	Toilet (0.500)
	Visitor Center (0.777)	Price (0.500)
Rottnest Island	Staff (1.000)	Price (0.468)
	Tour Guide (0.999)	Bus (0.476)
	Place (0.983)	Restaurant (0.495)
	Food (0.820)	Fly (0.495)
	Quokka (0.735)	Toilet (0.497)

Phillip Island’s most unique competitive advantages are its “kangaroo,”“penguin,”“place,”“staff,” and “penguin parade” touchpoints, reflected in its popularity as a wildlife destination with a variety of Australian wildlife, such as penguins and kangaroos (Bilbytravel, 2024). Its unique competitive disadvantages lie in the “cost,”“toilet,”“service,”“enclosure,” and especially “crowd” touchpoints. This is probably because the destination has attracted too many tourists, and the crowds of tourists have reduced the quality of services and the availability of facilities, such as toilets. For Cradle Mountain, its unique competitive advantages lie in the “staff,”“guide,”“place,”“shuttle bus” and “visitor centre” touchpoints, reflected in its fantastic, guided tour and convenient shuttle bus service (Kentish, 2024). However, its competitive disadvantages are the “toilet,”“price,”“weather,” particularly the “food” and “drink” touchpoints. For Rottnest Island, its key competitive advantages include the “staff,”“tour guide,”“place,”“food” and “quokka” touchpoints. Its competitive disadvantages, however, are the “restaurant,”“fly,”“toilet,”“bus” and especially the “price” touchpoints, as reflected in tourists’ perceptions of higher travel/service costs (Tripadvisor, 2024).

Competitiveness Improvement Strategies Formulation

Changeable Touchpoints and Dimensions

Table 4 presents the identified changeable touchpoints that could be considered in strategy formulation for each of the destinations, along with their respective dimension. The Great Ocean Road and Cradle Mountain have 41 and 22 changeable touchpoints, respectively, grouped into five dimensions: “facilities,”“food & drink,”“service,”“staff” and “value for money (VFM).” Phillip Island and Rottnest Island have 24 and 27 changeable touchpoints, respectively, grouped into six dimensions, of which “activities” is an additional dimension.

Table 4.

The Identified Changeable Touchpoints and Dimensions Related to the Respective Touchpoints of Each Destination.

Destination (no. of touchpoints)	Changeable touchpoints	Changeable dimensions
Great Ocean Road (41)	Amenity, café, car park, facility, pathway, picnic area, signage, step, table, toilet, track, trail, viewpoint, information center, shop	Facilities
	Breakfast, cake, chocolate, coffee, dessert, drink, food, ice cream, lunch, meal, chicken, pizza	Food & Drink
	Crowd, service, atmosphere, communication, traffic, helicopter, bus, accommodation	Service
	Guide, pilot, staff, crew	Staff
	Cost, price	Value for Money (VFM)
Phillip Island (24)	Activity, event	Activities
	Boardwalk, café, enclosure, facility, seat, shop, sign, toilet, viewing spot, visitor center	Facilities
	Food, snack, water	Food & Drink
	Booking, crowd, service, line	Service
	Ranger, staff	Staff
	Cost, price, ticket	Value for Money (VFM)
Cradle Mountain (22)	Amenity, cafe, carpark, facility, shop, toilet, track, visitor center	Facilities
	Chips, coffee, drink, food, lunch, meal	Food & Drink
	Information, service, shuttle bus	Service
	Driver, guide, ranger, staff	Staff
	Price	Value for Money (VFM)
Rottnest Island (27)	Activity	Activities
	Center, equipment, facility, locker, restaurant, toilet	Facilities
	Food, seafood, lunch	Food & Drink
	Service, accommodation, bus, bike, ferry, segway, map, souvenir	Service
	Captain, crew, staff, tour guide, driver, instructor, skipper, staff	Staff
	Price	Value for Money (VFM)

Tourist Clustering and Strategy Generation Performance

The quantified experience vectors obtained in the previous section were fed into the k-means algorithm to cluster tourists with unsatisfactory experiences according to their satisfaction (i.e., normalized group touchpoint opinion score) at destination touchpoints. Only tourists with unsatisfactory experiences were selected for this step, as the main focus of this study is to help destinations formulate strategies to improve their competitiveness in providing satisfactory experiences to these tourists. Following the existing literature, the elbow method (Sánchez-Rivero et al., 2023) was used to determine the number of clusters k. Figure 3 illustrates the variation of the within-cluster sum of squares against the number of clusters for the k-means clustering of unsatisfied tourists in the four destinations. Analysis of the “elbow” points in Figure 3 reveals five distinct clusters for the Great Ocean Road and four distinct clusters for Phillip Island, Cradle Mountain and Rottnest Island.

Figure 3.

The variation of the within-cluster sum of squares against the number of clusters for unsatisfied tourists: (a) Great Ocean Road, (b) Phillip Island, (c) Cradle Mountain, and (d) Rottnest Island.

Table 5 presents the top five touchpoints that can best differentiate each tourist group and their normalized group touchpoint opinion score in the four destinations. These touchpoints were identified by summarizing the pairwise absolute difference between each group’s normalized group touchpoint opinion score and those of the other groups. The results show distinct normalized group touchpoint opinion scores between tourist groups, and even different touchpoints for some groups, highlighting the need for targeted strategies to improve the perceived competitiveness of the destination for each tourist group.

Table 5.

The Top Five Touchpoints That Differentiate Each Tourist Group in the Four Destinations and Their Normalized Group Touchpoint Opinion Score.

Destination	Tourist groups	Touchpoints (normalized group touchpoint opinion score)
Great Ocean Road	G1	service (0.567), carpark (0.535), chocolate (0.634), staff (0.598), place (0.374)
	G2	service (0.540), carpark (0.535), chocolate (0.633), place (0.666), staff (0.328)
	G3	service (0.570), chocolate (0.635), staff (0.598), place (0.653), carpark (0.340)
	G4	price (0.492), carpark (0.535), staff (0.598), place (0.657), chocolate (0.427)
	G5	service (0.568), carpark (0.538), chocolate (0.636), staff (0.599), place (0.666)
Phillip Island	G1	place (0.641), ranger (0.606), cost (0.487), penguin (0.609), staff (0.563)
	G2	place (0.627), ranger (0.608), cost (0.483), penguin (0.296), staff (0.554)
	G3	penguin (0.614), ranger (0.598), bird (0.588), place (0.522), staff (0.406)
	G4	penguin (0.733), staff (0.556), seat (0.570), staff (0.556), cost (0.490)
Cradle Mountain	G1	service (0.579), visitor center (0.595), shuttle bus (0.541), place (0.639), staff (0.626)
	G2	visitor center (0.597), shuttle bus (0.549), service (0.559), place (0.638), staff (0.325)
	G3	service (0.584), visitor center (0.597), shuttle bus (0.580), place (0.638), staff (0.718)
	G4	service (0.584), shuttle bus (0.557), visitor center (0.553), staff (0.639), place (0.291)
Rottnest Island	G1	bike (0.507), quokka (0.588), food (0.598), service (0.600), staff (0.639)
	G2	quokka (0.597), driver (0.514), food (0.578), service (0.596), staff (0.322)
	G3	quokka (0.597), driver (0.525), food (0.600), staff (0.641), service (0.269)
	G4	quokka (0.597), driver (0.525), food (0.600), service (0.597), staff (0.731)

Having identified tourist groups, a counterfactual strategy identifier based on a three-layer neural network was developed to predict whether tourists were satisfied or not with their experience and to identify the corresponding improvement strategies (i.e., $f$ in Equation 4). The most popular first-order gradient descent-based optimizer—Adam (Kingma & Ba, 2014)—was used to train the neural network and solve the counterfactual optimization problem in Equation 4. For each tourist group, the optimization problem was solved 20 times to identify the final competitiveness improvement strategies. Table 6 summarizes the success rate of the competitiveness improvement strategies that our framework can generate (i.e., successfully finding an optimal solution to Equation 4) for each tourist group. The success rate exceeds 85% for all tourist groups, with an average of 97%. This high success rate demonstrates the reliability of our framework in supporting the development of destination competitiveness improvement strategies.

Table 6.

The Success Rate of the Developed Framework in Generating Competitiveness Improvement Strategies for Tourist Groups in the Four Destinations.

Destination	Tourist groups	No. generated strategies/total	Success rate of generation
Great Ocean Road	G1	19/20	95%
	G2	18/20	90%
	G3	20/20	100%
	G4	20/20	100%
	G5	17/20	85%
Phillip Island	G1	20/20	100%
	G2	20/20	100%
	G3	20/20	100%
	G4	20/20	100%
Cradle Mountain	G1	19/20	95%
	G2	20/20	100%
	G3	20/20	100%
	G4	20/20	100%
Rottnest Island	G1	17/20	85%
	G2	20/20	100%
	G3	20/20	100%
	G4	19/20	95%
Average		329/340	97%

Competitiveness Improvement Strategies

This study assumes that destinations prefer strategies that address the most negative changeable touchpoints of each tourist group and of the destination while minimizing the number of touchpoints changed and the total change. The results of the identified strategy according to this assumption for each tourist group are presented in Table 7. The number in parentheses indicates the required change in terms of the normalized group touchpoint opinion score for each touchpoint to effectively improve the competitiveness of the destination for each tourist group. This number provides managers with an overview of the extent to which a touchpoint needs to be changed to improve their destination’s competitiveness. For tourist groups of the Great Ocean Road, our identified strategies suggest that changes to the “crowd” touchpoint (G1); the “toilet” touchpoint (G2); the “bus” touchpoint (G3); the “crowd” and “step” touchpoints (G4); and the “bus” touchpoint (G5), respectively, can effectively improve the perceived competitiveness of each tourist group. Furthermore, the identified strategies emphasize a greater need to change the “step” touchpoint for tourist group G4.

Table 7.

Touchpoint-Level Competitiveness Improvement Strategies for Tourist Groups in the Four Destinations.

Destination	Tourist groups	Competitiveness improvement strategy (change)
Great Ocean Road	G1	crowd (0.501)
	G2	toilet (0.522)
	G3	bus (0.503)
	G4	crowd (0.537), step (0.705)
	G5	bus (0.503)
Phillip Island	G1	cost (0.514), enclosure (0.503), line (0.502)
	G2	cost (0.514), enclosure (0.503)
	G3	crowd (0.537), snack (0.500)
	G4	cost (0.514)
Cradle Mountain	G1	price (0.500), toilet (0.500)
	G2	information (0.490), price (0.500)
	G3	information (0.490), toilet (0.500)
	G4	food (0.509)
Rottnest Island	G1	driver (0.500), restaurant (0.505)
	G2	facility (0.501), price (0.532)
	G3	toilet (0.503)
	G4	driver (0.500), map (0.502)

For tourist groups of Phillip Island, our identified strategies suggest that changes to the “cost,”“enclosure” and “line” touchpoints (G1); the “cost” and “enclosure” touchpoints (G2); the “crowd” and “snack” touchpoints (G3); and the “cost” touchpoint (G4) can effectively improve the perceived competitiveness of each tourist group. The touchpoint requiring more change is “cost” for tourist groups G1 and G2, and “crowd” for tourist group G3. For tourist groups of Cradle Mountain, our identified strategies suggest that changes to the “price” and “toilet” touchpoints (G1); the “information” and “price” touchpoints (G2); the “information” and “toilet” touchpoints (G3); and the “food” touchpoint (G4), respectively, can elevate the perceived competitiveness of each tourist group. The touchpoint requiring more change is “price” for tourist group G2 and “toilet” for tourist group G3. For tourist groups of Rottnest Island, our identified strategies suggest that changes to the “driver” and “restaurant” touchpoints (G1); the “facility” and “price” touchpoints (G2); the “toilet” touchpoint (G3); and the “driver” and “map” touchpoints (G4), respectively, can improve the perceived competitiveness of each tourist group. The touchpoint requiring more change is “restaurant” for tourist group G1, “price” for tourist group G2 and “map” for tourist group G4.

Unlike the strategies identified through existing methods, which focus on identifying several most concerning competitive disadvantages (i.e., touchpoints), the strategies identified in this study include a combination of touchpoints that entails minimal changes for each tourist group and required changes for each touchpoint, which is more informative to help managers make informed decisions to improve the competitiveness of their destinations to the desired level. As minimal changes indicate that these strategies require the least effort to improve the competitiveness of destinations, while targeting specific tourist groups allows managers to address the unique needs of each tourist group to improve the competitiveness of destinations for them.

Table 8 presents a comparison of the total required changes and potential outcomes of the strategies identified by our proposed framework versus the most concerning approach at the touchpoint level. The total required change for the strategies identified by our proposed method is calculated by summing the changes outlined in Table 7. In contrast, the total change for the most concerning approach is determined by summing the changes required to improve the $n$ lowest scoring touchpoints in Table 3 to the highest satisfaction level (i.e., 1). To ensure a fair comparison, $n$ is set to the maximum number of touchpoints in the strategies identified for each destination. Namely, two for Great Ocean Road (i.e., step and toilet), Cradle Mountain (i.e., food and drink), Rottnest Island (i.e., price and bus) and three for Phillip Island (i.e., crowd, cost, and toilet). The potential outcomes of the strategies are predicted by the neural network $f$ , where 1 indicates successful improvement of a destination’s competitiveness to a desired level, and 0 indicates failure to achieve the desired improvement. We can see that our method can identify strategies that require fewer changes than the most concerning approach to improve the destination competitiveness for all identified tourist groups. In addition, our method is more reliable because improving the most concerning touchpoints is more likely to fail to achieve the desired competitiveness improvement for some tourist groups (e.g., G3 for Phillip Island).

Table 8.

Comparison of the Total Change Required and the Potential Outcomes of the Strategies Identified by the Proposed Method and Most Concerning Approach.

		The proposed method		Most concerning approach
Destination	Tourist groups	Total change	Potential outcomes	Total change (touchpoints)	Potential outcomes
Great Ocean Road	G1	0.501	1	1.492 (step, toilet)	1
	G2	0.522			1
	G3	0.503			1
	G4	1.242			1
	G5	0.503			0
Phillip Island	G1	1.519	1	1.557 (crowd, cost, toilet)	0
	G2	1.017			1
	G3	1.037			0
	G4	0.514			1
Cradle Mountain	G1	1.000	1	1.013 (food, drink)	0
	G2	0.990			1
	G3	0.990			1
	G4	0.509			1
Rottnest Island	G1	1.005	1	1.056 (price, bus)	0
	G2	1.033			1
	G3	0.503			1
	G4	1.002			1

Based on the touchpoint-level strategies identified in Table 7, the dimension-level changes that the four destinations could make to improve their competitiveness for different tourist groups are examined and presented in Table 9. For tourist groups G1, G3 and G5 of the Great Ocean Road, our results suggest changes to the “service” dimension to improve their perceived competitiveness. For tourist groups G2 and G4 of Great Ocean Road, our results suggest changes to the “facilities” dimension; and both the “service” and “facilities” dimensions, to improve their respective perceived competitiveness, and the dimension that requires more change for tourist group G4 is “facilities.” For tourist groups G1, G2, G3 and G4 of Phillip Island, our results recommend that changes to the “value for money (VFM),”“facilities,” and “service” dimensions; the “value for money (VFM)” and “facilities” dimensions; the “service” and “food” dimensions, and the “value for money (VFM)” dimension, respectively, can improve the perceived competitiveness for each tourist group. The dimension requiring more change for tourist groups G1 and G2 is “value for money (VFM),” and, for tourist group G3, it is “service.”

Table 9.

Dimension-Level Competitiveness Improvement Strategies for Tourist Groups in the Four Destinations.

Destination	Tourist groups	Dimension (total change)
Great Ocean Road	G1	Service (0.501)
	G2	Facilities (0.522)
	G3	Service (0.503)
	G4	Service (0.537), Facilities (0.705)
	G5	Service (0.503)
Phillip Island	G1	Value for Money (VFM; 0.514), Facilities (0.503), Service (0.502)
	G2	Value for Money (VFM; 0.514), Facilities (0.503)
	G3	Service (0.537), Food (0.500)
	G4	Value for Money (VFM; 0.514)
Cradle Mountain	G1	Value for Money (VFM; 0.500), Facilities (0.500)
	G2	Service (0.490), Value for Money (VFM; 0.500)
	G3	Service (0.490), Facilities (0.500)
	G4	Food (0.509)
Rottnest Island	G1	Staff (0.500), Facilities (0.505)
	G2	Facilities (0.501), Value for Money (VFM; 0.532)
	G3	Facilities (0.503)
	G4	Staff (0.500), Service (0.502)

For tourist groups G1, G2, G3 and G4 of Gradle Mountain, our results suggest that changes to the “value for money (VFM),” and “facilities” dimensions; the “service” and “value for money (VFM)” dimensions; the “service” and “facilities” dimensions; and the “food” dimension, respectively, can improve the perceived competitiveness for each tourist group. The dimension requiring more change is “value for money (VFM)” for tourist group G2 and “facilities” for tourist group G3. For tourist groups G1, G2, G3 and G4 of Rottnest Island, our results suggest that changes to the “staff” and “facilities” dimensions; the “facilities” and “value for money (VFM)” dimensions; the “facilities” dimension; and the “staff” and “service” dimensions, respectively, can improve the perceived competitiveness for each tourist group. The dimension requiring more change is “facilities” for tourist group G1, “value for money (VFM)” for tourist group G2 and “service” for tourist group G4. These dimension-level strategies offer detailed insights into the accurate dimensions that require change to improve the competitiveness of the studied destinations for each tourist group, which is valuable in guiding overall resource allocation decisions for targeted competitiveness improvement.

Figure 4 provides an overview of the total changes required in each dimension to improve the competitiveness of the four destinations, calculated by summing up the corresponding touchpoint changes in each dimension. The maximum changes of the same touchpoints are used as the touchpoint change during the summation. For the Great Ocean Road, the dimension requiring the greatest number of changes to improve competitiveness is “facilities,” followed by the “service” dimension. For Phillip Island, the dimension that requires the greatest number of changes to improve competitiveness is “service,” followed by the dimensions of “value for money (VFM),”“facilities” and “food.” For Cradle Mountain, the dimension that requires the most changes to improve competitiveness is “food,” followed by “value for money (VFM),”“facilities” and “service.” For Rottnest Island, the dimension requiring the most changes to improve competitiveness is “facilities,” followed by “value for money (VFM),”“service,” and “staff.”

Figure 4.

Overview of the dimension-level changes to improve the competitiveness of the four destinations: (a) Great Ocean Road, (b) Phillip Island, (c) Cradle Mountain, and (d) Rottnest Island.

Discussion and Implications

With the proposed framework and results, this study makes several contributions to the existing literature. First, this study extends the methodology of destination competitiveness improvement. Unlike methods such as importance performance analysis (Albayrak et al., 2018; Mustafa et al., 2020) and calculation of importance coefficients (Sánchez & López, 2015), which focus on prioritizing competitive disadvantages (i.e., touchpoints/dimensions) to improve destination competitiveness, our framework is able to identify the most optimal touchpoints/dimensions combinations that entail minimal changes to improve destination competitiveness using AI methods. The proposed framework serves as an initial attempt to improve existing destination competitiveness improvement methods by integrating optimization-based AI decision analytic techniques. These techniques have attracted considerable attention in the recent information systems literature (Ara et al., 2024), yet they remain underexplored in the context of destination competitiveness. Furthermore, unlike existing methods that recommend a set of improvement strategies for all tourists (Ayikoru, 2015; Azzopardi & Nash, 2016), our framework captures differences between tourist groups to identify tailored competitiveness improvement strategies for each group. To the best of our knowledge, this is the first study to develop an analytical framework to identify group-specific competitiveness improvement strategies.

The proposed analytical framework also extends the methodology of AI-based analysis in the tourism literature. Currently, most AI methods used in the tourism literature are association based, such as descriptive AI methods (e.g., associate rule mining; Vu et al., 2018), which aim to capture the patterns in data to help researchers understand complex phenomena, and predictive AI methods (e.g., long short-term memory neural networks; Zhang et al., 2021), which use historical data to make predictions. While these methods have contributed substantially to the understanding of a wide range of tourism problems, the inability of these methods to capture causality and estimate potential outcomes of strategies limits their capacity to formulate more effective strategies to support decision-making (Choudhury et al., 2021). This study is one of the first attempts to explore the use of causal (i.e., prescriptive) AI methods (Langen & Huber, 2023) to support decision-making in the tourism context. Given that tourism management practice often involves interventions (i.e., changing something) and estimating outcomes for decision-making, we call for future studies to explore the use of causal AI methods (Huenermund et al., 2022) to advance tourism management research and practice.

Second, this study has enriched the theoretical underpinnings of the destination competitiveness improvement literature. To the best of our knowledge, this is the first study to challenge the most concerning/important view of strategy identification for improving destination competitiveness and the first to propose the minimal change view of strategy identification by formulating an optimization problem. The main argument of this study is that focusing on the most concerning touchpoints/dimensions may not be the optimal way to identify directions in which to improve destination competitiveness. This is because the touchpoints/dimensions most concerning to tourists may be difficult to change (e.g., “place” for G1 on the Great Ocean Road, in Table 5), and improving the combination of touchpoints/dimensions may require fewer changes to achieve the desired improvement in destination competitiveness (see Table 8). This new perspective on strategy identification opens new avenues for destination managers to take advantage of novel optimization-based AI methods to support strategy development for the improvement of destination competitiveness.

Furthermore, this study is one of the initial attempts to enrich destination competitiveness improvement literature using theoretical underpinnings from the market segmentation literature (Floyd & Gramann, 1997; Prayag et al., 2021). Most previous studies have assumed that tourists in a destination are homogeneous and, therefore, that the same changes can be applied to improve a destination’s competitiveness for all tourists (Ayikoru, 2015; Chens et al., 2008; Chin, 2022). However, as the tourism industry enters the customized experiential era, it is essential to explore the differences between different tourist groups and to develop targeted strategies to improve destination competitiveness (Neto et al., 2020; Peng & Tzeng, 2012). Our study responds to this need, and our results provide detailed strategies that the studied destinations can adopt to improve their competitiveness for different tourist groups, which has not been explored in the literature.

Identifying a comprehensive list of touchpoints that determine tourist experiences (Table 2), this study also lays theoretical foundations for future research to conceptualize and analyze destination competitiveness from the perspective of tourist experience. Currently, most studies conceptualize and analyze destination competitiveness based on established models that include a wide range of pre-defined factors (Dwyer & Kim, 2003; Ritchie & Crouch, 2003). Despite its popularity, one limitation of this approach is that it does not allow for the identification of unique aspects of the product (i.e., the tourist experience) that make a destination competitive in the tourism market. The touchpoint theory adopted in this study provides researchers with the opportunity to understand the unique touchpoints of the tourist experience that compose the competitive advantages and disadvantages of destinations. The analysis conducted in this study also echoes Shoval and Birenboim’s (2019) call for using disaggregated tourist experiences rather than an aggregated satisfaction score to understand destination competitiveness. The changeable touchpoints and dimensions identified also provide detailed insights into what can be changed at the touchpoints (i.e., the micro) level and dimension (i.e., meso) level to improve destination competitiveness, which forms the knowledge base for subsequent studies to improve destination competitiveness from the perspective of tourist experience.

Third, this study also has several practical implications. To begin with, our results reveal the competitive advantages and disadvantages of the destinations studied, which can help managers understand their destination’s unique competitive position. For example, our results indicate that the unique competitive advantages of the Great Ocean Road lie in its scenery-related touchpoints, while the unique competitive advantage of Phillip Island can be identified as its wildlife-related touchpoints. Furthermore, our results provide managers of the studied destinations with detailed insights into what could be changed to improve the competitiveness of their destination for different tourist groups at the touchpoint level and the dimension level, while outlining the required changes at each level. These insights are useful for helping managers allocate budgets and make informed decisions to improve the competitiveness of the studied destinations in the tourism market.

Conclusions

Due to the importance of high levels of competitiveness for destination prosperity, existing studies have devoted extensive efforts to explore knowledge that may help destinations improve their competitiveness (Assaker et al., 2014; Y. R. Kim et al., 2022). Most of these studies have directly applied competitiveness evaluation models to identify competitive disadvantages that destinations need to change to improve their competitiveness. However, as destinations cannot change all their competitive disadvantages, some initial studies have focused on prioritizing these factors and recommending improvements to the most concerning ones. Despite the contributions, this approach may not be optimal, as the most concerning competitive disadvantages may be difficult to change and may require more changes than a combination of developments to other disadvantages would to improve a destination’s competitiveness to a desired level. Furthermore, most existing methods assume that tourists are homogeneous and are therefore unable to capture the differences between tourist groups to identify targeted strategies that will improve the competitiveness of destinations for different tourist groups. However, as suggested in recent studies, destination competitiveness is a dynamic concept, and it is perceived differently by different tourist groups (Neto et al., 2017, 2020); hence, there is a need to identify targeted strategies for individual tourist groups.

To address the limitations of these existing methods, this study developed an analytical framework based on optimization-based AI methods and online user-generated data for identifying strategies to improve destination competitiveness. The framework can take advantage of the interaction of competitive disadvantages and capture the differences between different tourist groups to identify optimal strategies that entail minimal changes to improve the perceived competitiveness of a destination for particular tourist groups. The application of the proposed framework is demonstrated through a case study involving four tourist destinations in Australia. The results reveal detailed touchpoint- and dimension-level strategies that the four destinations could adopt to improve their competitiveness for different groups of tourists, which has not been explored in the existing literature.

Similar to other studies, this study has several limitations that could serve as a starting point for future studies. To begin with, the strategies identified in our framework are derived from computer simulations. Although the simulation results of the causal counterfactual AI algorithm used can identify optimal strategies that entail minimal changes to improve destination competitiveness for different tourist groups, how the resulting strategies can be effectively integrated into destination management practice still needs further investigation. Future research could build on our study by employing experimental designs to identify the detailed steps and the best practices for integrating AI algorithms with the managerial expertise of managers to improve destination competitiveness. Furthermore, the dimension-level competitiveness analysis conducted in this study may not be applicable to other destinations, as these dimensions were derived based on our understanding of the touchpoints. Future studies can explore different dimensions according to their research purpose and the expertise of the researchers involved. Finally, our data only include reviews and overall ratings; future studies could extend this research by exploring the use of aspect ratings in travel platforms and integrating reviewers’ profiles to examine the demographic characteristics of different tourist groups.

Footnotes

Appendix

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is supported by an Australian Government Research Training Program (RTP) Scholarship

ORCID iDs

Haiyang Xia

Birgit Muskat

Marion Karl

Gang Li

Notes

Author Biographies

Haiyang Xia is a PhD candidate at the Research School of Management, Australian National University. His research interests include machine learning and its application in tourism and management.

Birgit Muskat is an associate professor at the Research School of Management, Australian National University. Her research interests include experience management entrepreneurship and innovation with a contextual focus on tourism and hospitality.

Marion Karl is a senior lecturer at the School of Hospitality and Tourism Management, University of Surrey. Her research focuses on tourists’ travel behaviour and relevant influencing factors of travel decisions, including risk perceptions and emotions.

Qian Li is a lecturer at the School of Electrical Engineering, Computing and Mathematical Sciences, Curtin University. Her research interests include causal machine learning, data mining and their applications in management.

Gang Li is a professor at the School of Information Technology, Deakin University. His research interests are data science, data privacy and technology applications to tourism and hospitality.

References

Abreu-Novais

Ruhanen

Arcodia

(2016). Destination competitiveness: What we know, what we know but shouldn’t and what we don’t know but should. Current Issues in Tourism, 19(6), 492–512. https://doi.org/10.1080/13683500.2015.1091443

Albayrak

Caber

Rosario González-Rodríguez

Aksu

(2018). Analysis of destination competitiveness by IPA and IPCA methods: The case of Costa Brava, Spain against Antalya, Turkey. Tourism Management Perspectives, 28, 53–61. https://doi.org/10.1016/j.tmp.2018.07.005

Andrades-Caldito

Sánchez-Rivero

Pulido-Fernández

J. I.

(2014). Tourism destination competitiveness from a demand point of view: An empirical analysis for Andalusia. Tourism Analysis, 19(4), 425–440. https://doi.org/10.3727/108354214X14090817031035

Andrades

Dimanche

(2017). Destination competitiveness and tourism development in Russia: Issues and challenges. Tourism Management, 62, 360–376. https://doi.org/10.1016/j.tourman.2017.05.008

Ara

Maraj

M. A. A.

Rahman

M. A.

Bari

M. H.

(2024). The impact of machine learning on prescriptive analytics for optimized business decision-making. International Journal of Management Information Systems and Data Science, 1(1), 7–18. https://doi.org/10.62304/ijmisds.v1i1.112

Assaker

Hallak

Vinzi

V. E.

O’Connor

(2014). An empirical operationalization of countries’ destination competitiveness using partial least squares modeling. Journal of Travel Research, 53(1), 26–43. https://doi.org/10.1177/0047287513481275

Ayikoru

(2015). Destination competitiveness challenges: A Ugandan perspective. Tourism Management, 50, 142–158. https://doi.org/10.1016/j.tourman.2015.01.009

Azzopardi

Nash

(2016). A framework for island destination competitiveness–perspectives from the island of Malta. Current Issues in Tourism, 19(3), 253–281. https://doi.org/10.1080/13683500.2015.1025723

Bilbytravel. (2024). Great Ocean Road and Phillip Island in one day. https://bilbytravel.com.au/great-ocean-road-and-phillip-island-in-one-day

10.

Chens

C. Y.

Sok

(2008). Evaluating the competitiveness of the tourism industry in Cambodia: Self-assessment from professionals. Asia Pacific Journal of Tourism Research, 13(1), 41–66. https://doi.org/10.1080/10941660701883367

11.

Chin

C. H.

(2022). Empirical research on the competitiveness of rural tourism destinations: A practical plan for rural tourism industry post-COVID-19. Consumer Behavior in Tourism and Hospitality, 17(2), 211–231. https://doi.org/10.1108/CBTH-07-2021-0169

12.

Choudhury

Allen

R. T.

Endres

M. G.

(2021). Machine learning for pattern discovery in management research. Strategic Management Journal, 42(1), 30–57. https://doi.org/10.1002/smj.3215

13.

Cracolici

M. F.

Nijkamp

(2009). The attractiveness and competitiveness of tourist destinations: A study of Southern Italian regions. Tourism Management, 30(3), 336–344. https://doi.org/10.1016/j.tourman.2008.07.006

14.

Croes

(2011). Measuring and explaining competitiveness in the context of small island destinations. Journal of Travel Research, 50(4), 431–442. https://doi.org/10.1177/0047287510368139

15.

Crouch

G. I.

(2011). Destination competitiveness: An analysis of determinant attributes. Journal of Travel Research, 50(1), 27–45. https://doi.org/10.1177/0047287510362776

16.

Crouch

G. I.

Ritchie

J. R. B.

(1999). Tourism, competitiveness, and societal prosperity. Journal of Business Research, 44(3), 137–152. https://doi.org/10.1016/S0148-2963(97)00196-3

17.

De Keyser

Verleye

Lemon

K. N.

Keiningham

T. L.

Klaus

. (2020). Moving the customer experience field forward: Introducing the touchpoints, context, qualities (TCQ) nomenclature. Journal of Service Research, 23(4), 433–455. https://doi.org/10.1177/1094670520928390

18.

Dolnicar

Grün

Leisch

Schmidt

(2014). Required sample sizes for data-driven market segmentation analyses in tourism. Journal of Travel Research, 53(3), 296–306. https://doi.org/10.1177/0047287513496475

19.

Donders

A. R.

van der Heijden

G. J.

Stijnen

Moons

K. G.

(2006). A gentle introduction to imputation of missing values. Journal of Clinical Epidemiology, 59(10), 1087–1091. https://doi.org/10.1016/j.jclinepi.2006.01.014

20.

Durand

Vaara

(2009). Causation, counterfactuals, and competitive advantage. Strategic Management Journal, 30(12), 1245–1264. https://doi.org/10.1002/smj.793

21.

Dwyer

Kim

(2003). Destination competitiveness: Determinants and indicators. Current Issues in Tourism, 6(5), 369–414. https://doi.org/10.1080/13683500308667962

22.

Enright

M. J.

Newton

(2004). Tourism destination competitiveness: A quantitative approach. Tourism Management, 25(6), 777–788. https://doi.org/10.1016/j.tourman.2004.06.008

23.

Enright

M. J.

Newton

(2005). Determinants of tourism destination competitiveness in Asia Pacific: Comprehensiveness and universality. Journal of Travel Research, 43(4), 339–350. https://doi.org/10.1177/0047287505274647

24.

Floyd

M. F.

Gramann

J. H.

(1997). Experience-based setting management: Implications for market segmentation of hunters. Leisure Sciences, 19(2), 113–128. https://doi.org/10.1080/01490409709512243

25.

Frochot

(2005). A benefit segmentation of tourists in rural areas: A Scottish perspective. Tourism Management, 26(3), 335–346. https://doi.org/10.1016/j.tourman.2003.11.016

26.

Gao

Xia

Deng

Muskat

Law

(2024). Value creation in wine tourism–an exploration through deep neural networks. Journal of Vacation Marketing, 30(3), 376–391. https://doi.org/10.1177/13567667221140605

27.

Gomezelj

D. O.

Mihalič

(2008). Destination competitiveness—Applying different models, the case of Slovenia. Tourism Management, 29(2), 294–307. https://doi.org/10.1016/j.tourman.2007.03.009

28.

Hannart

Pearl

Otto

F. E. L.

Naveau

Ghil

(2016). Causal counterfactual theory for the attribution of weather and climate-related events. Bulletin of the American Meteorological Society, 97(1), 99–110. https://doi.org/10.1175/BAMS-D-14-00034.1

29.

Holmlund

Van Vaerenbergh

Ciuchita

Ravald

Sarantopoulos

Ordenes

F. V.

Zaki

(2020). Customer experience management in the age of big data analytics: A strategic framework. Journal of Business Research, 116, 356–365. https://doi.org/10.1016/j.jbusres.2020.01.022

30.

Horn

R. A.

Yang

(2020). Rank of a Hadamard product. Linear Algebra and Its Applications, 591, 87–98. https://doi.org/10.1016/j.laa.2020.01.005

31.

Huenermund

Kaminski

J. C.

Schmitt

(2022). Causal machine learning and business decision making. Academy of Management Proceedings, 2021(1), 12517. https://doi.org/10.2139/ssrn.3867326

32.

Kentish. (2024). Cradle Mountain. https://www.sheffieldtasmania.com.au/attraction/cradle-mountain

33.

Kim

Fung So

K. K

. (2024). Customer touchpoints: Conceptualization, index development, and validation. Tourism Management, 103, 104881. https://doi.org/10.1016/j.tourman.2023.104881

34.

Kim

Y. R.

Liu

Williams

A. M.

(2022). Competitiveness in the visitor economy: A systematic literature review. Tourism Economics, 28(3), 817–842. https://doi.org/10.1177/13548166211034437

35.

Kingma

D. P.

(2014). Adam: A method for stochastic optimization [arXiv preprint]. arXiv:1412.6980. https://doi.org/10.48550/arXiv.1412.6980

36.

Kumar

Dhir

(2020). Associations between travel and tourism competitiveness and culture. Journal of Destination Marketing & Management, 18, 100501. https://doi.org/10.1016/j.jdmm.2020.100501

37.

Langen

Huber

(2023). How causal machine learning can leverage marketing strategies: Assessing and improving the performance of a coupon campaign. PLoS One, 18(1), e0278937. https://doi.org/10.1371/journal.pone.0278937

38.

Lewis

(1973). Counterfactuals and comparative possibility. Journal of Philosophical Logic, 2(4), 418–446. https://www.jstor.org/stable/30226074

39.

B. X. B.

Gao

(2023). Restaurant survival prediction using customer-generated content: An aspect-based sentiment analysis of online reviews. Tourism Management, 96, 104707. https://doi.org/10.1016/j.tourman.2022.104707

40.

Mathayomchan

Taecharungroj

(2020). “How was your meal?” Examining customer experience using Google maps reviews. International Journal of Hospitality Management, 90, 102641. https://doi.org/10.1016/j.ijhm.2020.102641

41.

Mazanec

J. A.

Wöber

Zins

A. H.

(2007). Tourism destination competitiveness: From definition to explanation? Journal of Travel Research, 46(1), 86–95. https://doi.org/10.1177/0047287507302389

42.

McGill

A. L.

(2000). Counterfactual reasoning in causal judgments: Implications for marketing. Psychology and Marketing, 17(4), 323–343. https://doi.org/10.1002/(SICI)1520-6793(200004)17:4<<323::aid-mar4>3.0.CO;2-X

43.

Molnar

(2023). Interpretable machine learning. https://christophm.github.io/interpretable-ml-book/

44.

Mustafa

Omar

Mukhiar

S. N. S.

(2020). Measuring destination competitiveness: An importance-performance analysis (IPA) of six top island destinations in South East Asia. Asia Pacific Journal of Tourism Research, 25(3), 223–243. https://doi.org/10.1080/10941665.2019.1687534

45.

Neto

A. Q.

Dimmock

Lohmann

Scott

(2020). Destination competitiveness: How does travel experience influence choice? Current Issues in Tourism, 23(13), 1673–1687. https://doi.org/10.1080/13683500.2019.1641070

46.

Neto

A. Q.

Lohmann

Scott

Dimmock

(2017). Rethinking competitiveness: Important attributes for a successful scuba diving destination. Tourism Recreation Research, 42(3), 356–366. https://doi.org/10.1080/02508281.2017.1308086

47.

Pearl

Mackenzie

(2018). The book of why: The new science of cause and effect. Basic Books.

48.

Peng

K. H.

Tzeng

G. H.

(2012). Improvement tourism competitiveness using hybrid fuzzy MCDM [Conference session]. 2012 International Conference on Fuzzy Theory and Its Applications (iFUZZY2012) (pp. 24–29). IEEE. https://doi.org/10.1109/iFUZZY.2012.6409670

49.

Prayag

Alrawadieh

(2021). Motivation, emotion and world heritage status in discerning the heritage tourists: A segmentation perspective. Tourism Management Perspectives, 40, 100906. https://doi.org/10.1016/j.tmp.2021.100906

50.

Ritchie

J. R. B.

Crouch

G. I.

(2003). The competitive destination: A sustainable tourism perspective. CABI.

51.

Sánchez

A. G.

López

D. S.

(2015). Tourism destination competitiveness: The Spanish Mediterranean case. Tourism Economics, 21(6), 1235–1254. https://doi.org/10.5367/te.2014.0405

52.

Sánchez-Rivero

Rodríguez-Rangel

M. C.

Ricci-Risquete

(2023). K-means segmentation of tourism accommodation based on the active use of websites: Its application to an emerging destination (Extremadura, Spain). Journal of Vacation Marketing, 29(4), 654–669. https://doi.org/10.1177/13567667221117303

53.

Sharma

Kiciman

(2020). Causal inference and counterfactual reasoning [Conference session]. Proceedings of the 7th ACM IKDD CoDS and 25th COMAD (pp. 369–370). ACM. https://doi.org/10.1145/3371158.3371231

54.

Shoval

Birenboim

(2019). Customization and augmentation of experiences through mobile technologies: A paradigm shift in the analysis of destination competitiveness. Tourism Economics, 25(5), 661–669. https://doi.org/10.1177/1354816618806428

55.

Song

Jiao

Park

(2023). Sectoral productivity and destination competitiveness. Annals of Tourism Research, 103, 103645. https://doi.org/10.1016/j.annals.2023.103645

56.

Song

van der Veen

Chen

J. L.

(2012). The Hong Kong tourist satisfaction index. Annals of Tourism Research, 39(1), 459–479. https://doi.org/10.1016/j.annals.2011.06.001

57.

Srivastava

Kaul

(2016). Exploring the link between customer experience–loyalty–consumer spend. Journal of Retailing and Consumer Services, 31, 277–286. https://doi.org/10.1016/j.jretconser.2016.04.009

58.

Stein

Ramaseshan

(2016). Towards the identification of customer experience touch point elements. Journal of Retailing and Consumer Services, 30, 8–19. https://doi.org/10.1016/j.jretconser.2015.12.001

59.

Tan

Chen

Zhang

(2021). Counterfactual explainable recommendation. Proceedings of the 30th ACM International Conference on Information & Knowledge Management (pp. 1784–1793). ACM. https://doi.org/10.1145/3459637.3482420

60.

Tan

P. N.

Steinbach

Karpatne

Kumar

(2023). Introduction to data mining. Pearson. https://www-users.cse.umn.edu/∼kumar001/dmbook/index.php

61.

Tripadvisor. (2024). Rottnest Island on a budget. https://www.tripadvisor.com.au/ShowTopic-g488366-i16108-k8897321-Rottnest_Island_on_a_budget-Rottnest_Island_Fremantle_Greater_Perth_Western_Australia.html

62.

Verma

Dickerson

Hines

(2020). Counterfactual explanations and algorithmic recourses for machine learning: A review [arXiv preprint]. arXiv:2010.10596. https://doi.org/10.48550/arXiv.2010.10596

63.

H. Q.

Law

Zhang

(2018). Travel diaries analysis by sequential rule mining. Journal of Travel Research, 57(3), 399–413. https://doi.org/10.1177/0047287517692446

64.

Wadjemup Rottnest

I. S

. (2024). Connect to all that is here. https://www.rottnestisland.com/

65.

Xia

Muskat

Karl

Law

(2024). Destination competitiveness research over the past three decades: A computational literature review using topic modelling. Journal of Travel & Tourism Marketing, 41(5), 726–742. https://doi.org/10.1080/10548408.2024.2332278

66.

Xia

Muskat

Prayag

(2023). AI-based counterfactual reasoning for tourism research. Annals of Tourism Research, 101, 103617. https://doi.org/10.1016/j.annals.2023.103617

67.

Yang

Zeng

Yang

Song

(2021). A multi-task learning model for Chinese-oriented aspect polarity classification and aspect term extraction. Neurocomputing, 419, 344–356. https://doi.org/10.1016/j.neucom.2020.08.001

68.

Yang

(2020). Understanding commuting patterns and changes: Counterfactual analysis in a planning support framework. Environment and Planning B Urban Analytics and City Science, 47(8), 1440–1455. https://doi.org/10.1177/2399808320924433

69.

Zhang

Muskat

Law

(2021). Tourism demand forecasting: A decomposed deep learning approach. Journal of Travel Research, 60(5), 981–997. https://doi.org/10.1177/0047287520919522