Effects of digital readiness on digital competence of AEC companies: a dual-stage PLS-SEM-ANN analysis

ABSTRACT

To what extent does a firm’s digital readiness influence its competence in implementing digital initiatives? This study employs a deep-learning-based dual-stage approach using Partial Least Squares Structural Equation Modelling (PLS-SEM) and Artificial Neural Network (ANN) to demonstrate and quantify this relationship. Data were sourced from a questionnaire survey involving 428 architecture, engineering and construction (AEC) firms in New Zealand. The PLS-SEM analysis confirmed the positive correlation between the digital readiness of an organization and its competence towards seven types of DT, including immersive technologies, sensing technology, robotics, 3D printing, digital fabrication, artificial intelligence and big data. The ANN analysis further quantified the importance of the investigated readiness indicators in influencing digital competence. The results highlighted four most significant readiness attributes influencing the digital competence of AEC firms: (1) organizational culture, (2) perception of the leadership team, (3) hardware & software systems and (4) strategy plans. The findings can serve as a baseline for developing effective change management strategies and contribute to reducing the digital divide within AEC organizations, facilitating the effectiveness of organizational digital transformation.

KEYWORDS:

• Digital technology (DT)
• digital readiness
• digital competence
• partial least squares structural equation modelling (PLS-SEM)
• artificial neural network (ANN)
• architecture, engineering, and construction (AEC)

Introduction

The architecture, engineering and construction (AEC) sector is grappling with challenges such as escalating project complexities, growing environmental concerns and intensified stakeholder engagement (Olawumi & Chan, 2019; Saieg et al., 2018). Digital technologies (DT) have emerged as pivotal solutions, revolutionizing the planning, design, construction and maintenance of AEC projects through digital tools and solutions (Chen et al., 2022). The integrated use of DT offers advanced capabilities, including multiple-dimensional (nD) digital representation, complex models with predictive and prescriptive capabilities, reduced human errors, decreased material wastage and quality-assured project outcomes (Ebekozien & Samsurijan, 2024; Love & Matthews, 2019).

Incorporating multiple DT into AEC projects requires revamped construction processes, refined crew schedules, optimized equipment and material logistics, and strengthened stakeholder collaboration (Boje et al., 2020; Stute et al., 2021). However, central to these is an ability to comprehend digital competence of AEC practitioners (Hubschmid-Vierheilig et al., 2020; Khin & Ho, 2019; Vieru, 2014). Such an understanding will facilitate the customization of DT to meet project-specific requirements and user needs (Abdelaal & Guo, 2021; Okpala et al., 2021). Inadequate digital competence has been observed in various forms, such as a lack of proficiency in using digital tools (Bosch-Sijtsema et al., 2021), ineffective collaboration and communication (Shojaei et al., 2023) and resistance to change traditional workflows (Tabatabaee et al., 2021). Despite the growing recognition of the importance of digital competence in the AEC sector, existing literature such as those by Adeosun and Oke (2022), Gamil and Rahman (2019) and Jiang et al. (2021) have investigated the status of digital competence in the AEC industry. However, the influential factors affecting digital competence remain relatively unexplored.

Deploying DT in real-world AEC practices requires collaborative efforts encompassing changes to people, processes, technology and interactions (Ezcan et al., 2020). Recent research by Hubschmid-Vierheilig et al. (2020) and Mungra et al. (2024) indicates that the relationship between practitioners and firms in the AEC sector concerning technology adoption is inherently collaborative. The collaboration is grounded in mutual trust, effective communication and a shared goal of delivering projects on time, with quality and within budget. The digital competence of AEC practitioners, specifically their knowledge and practices towards new inventions, is closely linked to their past project experiences and job-related goals achieved through innovative tools (Lee & Meng, 2021). In corporate settings, digital competence is strongly correlated with the innovativeness and motivation of organizational personnel to embrace and utilize technology (Vieru, 2014). However, there is a dearth of scholarly research on the organizational readiness to adopt DT, also known as ‘organizational digital readiness’, and its impact on a firm’s competence in DT. This knowledge gap may impede the recognition and integration of digital solutions into a firm's core business (Ranta et al., 2021). The situation underscores the need for empirical studies into the question of ‘To what extent does organizational digital readiness influence a firm's competence in implementing DT initiatives, and how are they interconnected?’

This paper seeks to answer the aforementioned research question by adopting a dual-stage approach that combines Partial Least Squares Structural Equation Modelling (PLS-SEM) with Artificial Neural Network (ANN) techniques. This approach aims to elucidate the interaction between organizational digital readiness and firms’ DT competence. Two research objectives are specified:

1. Investigate the impact of organizational digital readiness on firms’ competence in implementing DT initiatives, considering both linear and potential non-linear relationships

2. Quantify the influence of readiness indicators, prioritizing those with the most significant impact on firms’ digital competence.

Background

Digital readiness

Digital readiness refers to the preparedness of an individual, organization, or society to harness and adapt to digital advancements (Chen et al., 2023). The concept extends beyond technological preparedness and encompasses a holistic view that includes mindset, skills, strategy, governance and cultural considerations brought about by DT. Digital readiness can be understood at multiple levels. At the individual level, it relates to an individual's knowledge, attitude and skills toward digital solutions (Cobos et al., 2016). Organizational-level readiness pertains to a firm's capability to adopt and integrate DT into its operations, services or products.

Exploring digital readiness and its influential indicators in organizational settings has been a significant area of research. For instance, Elshafey et al. (2020) employed the Technology Acceptance Model (TAM) (Davis, 1989) in a survey conducted in Malaysia, Egypt, Saudi Arabia and Turkey to assess the readiness of building information modelling integrated with augmented reality (AR-BIM) in the construction industry. The findings identified the perceived ease of use and perception of external control as the most influential to the readiness of AR-BIM. Juan et al. (2017) applied the Readiness for Workplace Change Management (RWCM) framework (Becker, 2005) to evaluate the BIM readiness in architectural firms in Taiwan. They pinpointed significant determinants, including government regulations, competitive encouragement, financial incentives and technical assistance. According to these studies and technology readiness theories, organizational digital readiness can be evaluated by indicators under multiple dimensions (Ezcan et al., 2020). Specifically, the Technology–Organization–Environment (TOE) framework (Tornatzky & Fleischer, 1990) provides insights into DT adoption from technological, organizational and environmental aspects. Cobos et al. (2016) and Xu et al. (2022) have highlighted the significance of the human aspect, particularly top and senior management, in influencing organizational digital readiness.

Digital competence

Digital competence can be defined as the knowledge (know-what), attitudes (know-why) and skills (know-how) required for organizations to identify and evaluate DT, implement and manage digital solutions and develop and enhance digital skills and capabilities within the organization (Osmundsen, 2020; Vieru, 2014; Wiesböck & Hess, 2018). Khin and Ho (2019) and Vieru (2014) indicated that the ability to align business strategies with existing digital transformation skills could significantly influence the adoption and utilization of DT. The importance of organizational digital competence has been increasingly recognized by the AEC industry. Evidence shows that organizations that can develop and leverage their digital competence are often better positioned to improve their operational efficiency, enhance their competitiveness and serve the needs of customers and stakeholders (Al-Edenat, 2023; Konopik et al., 2022; Trenerry et al., 2021).

However, inadequate digital competence has been observed in the literature. For instance, through workshops and online surveys with AEC practitioners, Bosch-Sijtsema et al. (2021) observed deficiencies in the competence in using advanced digital tools, highlighting the necessity for DT training and the introduction of new competence from other sectors to incorporate new DT. Shojaei et al. (2023), through semi-structured interviews and case studies, observed a lack of effective communication and collaboration, both internally within teams and externally with supplier networks. Tabatabaee et al. (2021) identified habitual resistance to adopt new technologies as a primary reason behind technology adoption failure. Recognizing the importance of digital competence, recent literature has explored further the status of digital competence in the AEC industry. For instance, Adeosun and Oke (2022) investigated the awareness and usage of cyber physical systems among AEC practitioners, while Gamil and Rahman (2019) examined awareness regarding BIM. Additionally, Jiang et al. (2021) investigated the practical applications of smart construction sites in the industry. However, despite these efforts, the influential factors affecting digital competence remain relatively underexplored.

Linkages between digital readiness and digital competence in organizational setting

Improvement of organizational digital competence can be achieved through a positive relationship between employees and the organization (Hubschmid-Vierheilig et al., 2020; Mungra et al., 2024). Employees with digital skills and openness to change are more likely to accept and utilize new technologies, whereas those lacking digital skills or resisting change may impede technology adoption (Gamil & Rahman, 2019). From the individual’s perspectives, TAM (Davis, 1989) suggests that improving digital readiness, such as by upgrading infrastructure (perceived ease of use) and providing necessary resources (perceived usefulness), can positively influence individuals’ perceptions of DT, thereby facilitating greater digital competence. To enhance employees’ digital competence, organizations play a crucial role in facilitating DT adoption and usage by providing necessary resources and support and fostering collaboration and knowledge sharing among employees (Al Hadwer et al., 2021). Drawing on organizational technology adoption theories, such as TOE (Tornatzky & Fleischer, 1990), RWCM (Becker, 2005) and Resource-Based View (RBV) theory (Barney, 2001), organizations could create an environment conducive to digital skill development by investing in technological infrastructure, fostering a supportive organizational culture, strengthening leadership support and providing training and resources. The presence of digital readiness initiatives, therefore, enables employees to access and engage with DT more effectively, leading to improvements in digital competence over time. The Diffusion of Innovations Theory (DOI) (Rogers, 1995) also emphasizes the importance of navigating the adoption process effectively to achieve and improve digital competence within organizations.

Similarly, the digital competence of AEC practitioners, particularly their knowledge, attitude and practice towards innovations, can be influenced by various factors within organizations. Past studies have investigated the knowledge and practice of different types of DT in AEC organizations and the associated influential factors. For instance, Castaneda and Cuellar (2020) conducted a systematic review and highlighted knowledge sharing as a critical factor that promotes innovation and organizational knowledge creation and application. Other crucial organizational factors linked with the organizational digital readiness include organizational culture (Martínez-Caro et al., 2020), leadership commitment and support (Razkenari & Kibert, 2022), training and development (Sukanthan Rajendra et al., 2022), resources and infrastructure (Axmann & Harmoko, 2020) and a supportive organizational structure (Lin et al., 2019). According to Lee and Meng (2021) and Zhen et al. (2021), these factors are considered significant mediator variables in determining the organizational digital competence. Although theoretical frameworks, such as TAM, TOE, RWCM, RBV and DOI, provide insights into how different factors shape digital readiness and subsequently influences digital competence, the direct impact of digital readiness on digital competence remains insufficiently established (Lee & Meng, 2021; Zhen et al., 2021). This understanding is crucial for AEC organizations seeking to enhance their organizational digital competence and improve their market competitiveness (Hubschmid-Vierheilig et al., 2020; Vieru, 2014).

Conceptual research framework and hypotheses development

This paper is a part of a broader research project on DT-enabled transformation in the AEC sector. The investigation covers various research themes, including the global technology implementation landscape, the digital readiness of AEC firms and the digital competence of AEC industry practitioners. A preceding segment of this research project, i.e. Chen et al. (2023), introduced a digital readiness model and a self-assessment tool for AEC organizations to evaluate their digital capabilities. A thorough review of technology adoption theories, technology readiness indicators and related studies on digital readiness has also been detailed by Chen et al. (2023). Building on this foundational work, the current study extends the relationship between digital readiness and digital competence in organizational contexts.

Chen et al. (2023) identified a list of indicators for measuring organizational digital readiness, from which we sourced 15 indicators for this study. The technological, organizational and environmental indicators were based on the TOE theory. In contrast, the remaining indicators under leadership and workforce dimensions were grounded in the RWCM theory (see Figure 1a). Through statistical analysis, Chen et al. (2023) further grouped these indicators into two core components: (1) the LEi component, which included indicators related to leadership and the internal environment and (2) the TOWEx component, which included indicators related to technology, organization, workforce and external environment. Further, 26 technologies implemented in the global AEC sector were identified through a systematic literature review by Chen et al. (2022). These technologies were classified into three categories: emerging DT, construction methods and building materials. Through a pilot study which will be explained in the research methodology section, we focused on the 14 types of emerging DT outlined by Chen et al. (2022) for further investigation (see Figure 1b). Other technologies, such as virtual prototyping, multi-dimensional modelling, eye-tracking and digital twins technologies, were excluded as they were considered as being less prevalent or underdeveloped among AEC practitioners in New Zealand. The selection of these 14 DT aimed to provide an overview of the digital landscape within the New Zealand context, covering a diverse range of tools and solutions currently utilized across various sectors by AEC firms. The detailed descriptions and examples for each DT have been provided in Appendix A.

Figure 1. Lists of readiness indicators and digital technologies included in the study.

Building upon prior research while expanding its depth, this study investigates the relationship between digital readiness and digital competence within organizations. Specifically, this study examines and quantifies the effects of the LEi and TOWEx components on firms’ competence in 14 types of DT. A conceptual research framework was proposed and presented in Figure 2, accompanied by three hypotheses proposed:

H1. The overall digital readiness of organizations significantly influences their digital competence.

H1a. The LEi component (leadership and internal environmental indicators) significantly influences organizational digital competence.

H1b. The TOWEx component (technological, organizational, workforce and external environmental indicators) significantly influences organizational digital competence.

Figure 2. The conceptual research framework.

Research methodology

A quantitative approach was employed, utilizing a structured questionnaire survey. As a powerful and adaptable tool, structured questionnaire surveys are efficient in collecting views and experiences of experts (Jin & Gambatese, 2020), enabling the quantitative exploration of complex phenomena and trends across diverse populations (Creswell & Creswell, 2017). The survey design and pre-test, sampling, data collection and data analysis methods are described in the following sub-sections.

Questionnaire design and pilot study

As stated in the prior section, the survey was designed based on the list of readiness indicators and DT obtained from Chen et al. (2022, 2023). The questions in the survey were carefully crafted to capture various dimensions of organizational digital readiness and firms’ DT competence. By employing a combination of Likert-scale questions, multiple-choice questions and matrix tables, it increases the level of complex statistical testing. We also provided detailed explanations and instructions for each question to ensure clarity and consistency in respondents’ interpretations, thus enabling future researchers to replicate our methodology effectively.

Before distributing the industry-wide survey, a pilot study was conducted with seven experienced industry experts, four prominent scholars in engineering and construction and two advisors from a research agency to ensure the questionnaire’s contextual relevance, reliability and usability. Based on the feedback, the wording of each indicator was amended into statements to improve clarity, and the list of DT was adjusted to adapt to the context of New Zealand. For instance, the statement for indicator L1 was expressed as ‘To what extent do you agree that leadership perception is important to ensure an organization is DT ready’. All participants agreed that the measurement items were complete and did not suggest any changes.

The finalized questionnaire was structured into three parts, including (1) participant demographic details, (2) perceived level of importance of the 15 readiness indicators (Figure 1a) and (3) perceived level of competence of the 14 DT (Figure 1b). For the readiness indicators, participants were asked to rate their level of agreement with each readiness statement on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), with 4 representing neither agree nor disagree. Regarding digital competence, participants were asked to rate their current knowledge and practice of the 14 DT on a nine-point Likert scale including 1 (no prior knowledge), 2 (aware, but not interested), 3 (aware, not currently using, but interested), 4 (not currently using but exploring further), 5 (limited usage), 6 (moderate usage), 7 (extensive usage), 8 (extensive usage with a desire to expand application) and 9 (extensive usage with an endeavour to develop custom applications). Based on the ‘seven plus or minus two’ principle by Miller (1956), the employment of a Likert scale in this manner could facilitate a convenient means for participants to articulate their views. The seven-point for the readiness indicators scale was chosen for certain constructs where a moderate level of granularity was sufficient for capturing respondents’ perceptions. In contrast, the nine-point scale for the digital competence was employed for constructs where a finer level of differentiation was desired to better capture nuances in respondents’ attitudes and behaviours. The scale positions 1–4 are intended to reflect the accumulation of knowledge or shifts in attitude towards technology. A rating of 5 signifies occasional technology use, while a rating of 6 indicates regular usage for moderately complex tasks. Scale points from 7 to 9 denote extensive, frequent and mission-critical technology use. By utilizing both scales, we ensured that the measurement instruments were tailored to the specific characteristics of each construct, thereby enhancing the validity and reliability of our data collection process. Both Likert scales were found to be unambiguous and comprehensible by participants in the pilot study, consistent with findings in the literature (e.g. Habibi et al., 2014; Preston & Colman, 2000).

Sampling and data collection

The targeted population for this study comprised all the firms in the AEC sector of New Zealand. The prioritized respondents included professionals directly engaged in DT implementation activities and senior personnel responsible for adopting technology or implementing decisions within their organizations. To minimize recall error and obtain reliable statistical data, an introductory page providing research background and questionnaire contents was added to the front page of the questionnaire. The targeted population was also clearly indicated on the cover page to ensure that those who responded to the survey met the conditions of the questionnaire. The data collection for this research was approved by the University of Auckland Human Participants Ethics Committee with reference number 023682.

As the competence level toward multiple DT is relatively new among industry practitioners, a completely random sampling method was not deemed appropriate (Etikan et al., 2016). Instead, snowball sampling was used to elicit data, as it allows for the inclusion of participants who may be difficult to reach through traditional random sampling methods (Etikan & Bala, 2017). Respondents with DT experience were approached through several methods, including searching through related industry publications, requesting a distribution from industry associations and contacting professionals through LinkedIn. The large user group and their professional social network enabled the questionnaire to be widely distributed, resulting in a sufficient number of respondents in the relevant field. Additionally, respondents were invited to share the survey link with individuals who met the targeted population criteria. These approaches ensured a diverse pool of participants representing different segments of the AEC industry.

Using the sample size calculator on Qualtrics (https://www.qualtrics.com/blog/calculating-sample-size/), we obtained an optimal sample size of 383 by setting a confidence level of 95%, a margin of error of 5% and a total of 70,629 AEC enterprises as of February 2021 reported by the Ministry of Business, Innovation & Employment (MBIE, 2022). Finally, we received 428 fully completed questionnaires for statistical analysis, exceeding the statistically estimated sample size. This could ensure data saturation and helps mitigate potential sampling biases.

Data analysis

A dual-stage approach combining PLS-SEM and ANN techniques was used to analyse the questionnaire data. The proposed PLS-SEM-ANN analytical procedure comprises six steps, as shown in Figure 3. The first two steps were presented in previous sections. The subsequent sections describe the analyses from step 3 to step 6.

Figure 3. The proposed analytical procedure.

PLS-SEM analysis

PLS-SEM is a prevalent multivariate analysis method for calculating variance-based structural equation models, particularly in social sciences fields. PLS-SEM works efficiently with complex models and has no data normality distribution and large sample size requirements, thereby contributing to its widespread adoption in recent project management research (e.g. Chen et al., 2019; Ghansah et al., 2024; Mardani et al., 2020; Zeng et al., 2021; Zuo et al., 2018). This study explored the impact of organizational digital readiness on digital competence, requiring a robust analytical approach capable of handling complex relationships and diverse variables. To achieve this objective, we adopted PLS-SEM to assess the measures employed in defining research constructs and to test hypotheses concerning these constructs within a single process.

PLS-SEM involves two main steps: the measurement model and the structural model. The measurement model assessment involves evaluating the reliability and validity of the constructs. The internal reliability was assessed using Cronbach’s alpha and composite reliability, and values of 0.70 are considered satisfactory in exploratory research studies (Hair et al., 2019). The convergent validity was assessed by the average variance extracted (AVE), which has a recommended level of over 0.50 (Al-Sharafi et al., 2023). The discriminant validity was assessed by comparing the correlations among the latent variables with the square root of AVE and the Heterotrait–Monotrait (HTMT) ratio (Ab Hamid et al., 2017). An HTMT ratio below the threshold of 0.85 suggests that the two constructs in comparison are distinct and not interchangeable (Ab Hamid et al., 2017). The structural model assessment involves identifying significant path coefficients using a bootstrapping procedure of 5000 subsamples and evaluating predictive relevance (Q²) through a blindfolding method with an omission distance of 7 (Al-Sharafi et al., 2023). The bootstrapping technique generates the variance inflation factor (VIF), the coefficient of determination (R²), t-values and p-values for model parameters, allowing hypothesis testing. SmartPLS software was used for conducting PLS-SEM analysis.

ANN analysis

Although PLS-SEM is effective for hypotheses testing, it can only examine the linear relationship, potentially leading to oversimplification of complexities in decision-making processes (Ng et al., 2022; Wang et al., 2022). To address this limitation and quantify the relationship between digital readiness and digital competence with sufficient accuracy, a deep learning-based ANN approach with two hidden layers (Zhang et al., 2022) was conducted. ANN allows us to capture potential non-linear relationships that may exist among the readiness indicators and firms’ digital competence. By combining both techniques, we aim to offer a more nuanced understanding of how different readiness indicators influence firms’ digital competence, considering both linear and non-linear effects. Unlike shallow ANN approaches with one hidden layer, deep learning-based ANN approaches have two or more hidden layers, providing enhanced accuracy of capturing non-compensatory and non-linear relationships within the model (Akour et al., 2022; Ashaari et al., 2021). Models with three or more layers were not considered due to the increased complexity requiring larger datasets for training and testing, and the inclusion of extra layers does not ensure improved outcomes (Aghaei et al., 2023). Therefore, as shown in Figure 4, the proposed ANN model includes an input layer, two hidden layers and an output layer.

Figure 4. The ANN model bridging digital readiness to digital competence.

The input layer X represents the scores of indicators from survey participants, and the neurons were classified into two groups, corresponding to the two components of the readiness model. The shape of the layer can be expressed as ($\sum _{i=1}^2{J}_i$, N) in which i = 1 and 2 is the number of components, J_i is the number of indicators under component i, and N is the number of survey participants. Each group in the first hidden layer has m_i neurons, which can be estimated from the input layer through the equation below: (1) $a_i=f(S_{\mathrm{Ji},\mathrm{mi}}\ast X_i+B_{\mathrm{Ji}})$(1) where f(), S_Ji,mi and B_Ji are the activation function, strength and bias for group i, respectively; X_i and a_i are the neurons in the input and first hidden layers, respectively. The second hidden layer has k neurons and connects the first hidden layer by using S_m,k and B_m in which m = m₁ + m₂. The number of neurons in the two hidden layers was determined as m₁ = m₂ = 16 and k = 32 through a trial-and-error process to achieve adequate modelling performance (Sharma et al., 2016). The output layer Y is connected with the second hidden layer through S_k,T and B_k in which T is the number of DT. To improve the modelling performance (Ashaari et al., 2021), the scores of neurons in the input and output layers were normalized within [0, 1].

The model was trained in Python using the Keras machine learning tool. To prevent overfitting, this study engaged a 10-fold cross-validation method which is a common validation method used in similar studies, e.g. Al-Sharafi et al. (2023) and Khayer et al. (2020). This method divided the dataset into 10 equally sized folds, with 9 (90%, i.e. 385 participants) used for training and the remaining 1 (10%, i.e. 43 participants) for validation in each iteration. Each fold could serve as the validation set once during the 10 iterations, thereby reducing the risk of overfitting and providing a comprehensive assessment of the model's predictive capability. The predictive accuracy was assessed using root mean squared error (RMSE), mean and standard deviation for both training and testing data. Preliminary optimization was conducted to determine activation functions, optimizer and loss function. The Rectified Linear Units (ReLU) function has been selected as the active function in the hidden layers due to its ability to accelerate model training (Zhang et al., 2021). The sigmoid function, i.e. the default active function for binary classification problems, was applied to the final output matrix. The cross-entropy loss function was adopted with values close to zero representing high degrees of similarity between predicted and actual results. The Adam optimizer, with a learning rate of 0.001, yielded the highest prediction accuracy after preliminary training.

Sensitivity analysis

Sensitivity analysis with the ANN was conducted to assess the robustness and stability of the model by systematically varying the input variables within a certain range and observing the corresponding changes in the outputs of the model (Al-Sharafi et al., 2023; Zhang et al., 2022). Therefore, the sensitivity analysis allows us to understand the relative importance of different input variables and identify which variables have the most significant impact on the model’s predictions.

The outcomes of the sensitivity analysis include the average importance of each readiness indicator, which is crucial for feature selection, dimensionality reduction and model interpretability (Al-Sharafi et al., 2023). The importance of each indicator was calculated by multiplying the strength of layers in the trained ANN model, using the connection weights approach, as it allows for learning and inheritance of rationality from the topology of the trained ANN (Zhang et al., 2022). The product of strength S_m,k and S_k,T was divided into two parts with shapes of (m₁,T) and (m₂,T), corresponding to the two components of the readiness model, and the importance for indicators under component i was then calculated by using the following equation: (2) $S_{\mathrm{Ji},T}=S_{\mathrm{Ji},\mathrm{mi}}\cdot (S_{m,k}\cdot S_{k,T}{)}_i$(2) where S_Ji,T includes the importance of J_i indicators for T technologies under component i and the mean strength for T technologies was taken as the total strength for each indicator.

Results

Profile of survey respondents

The respondent demographics are presented in Figure 5. The largest group of respondents were engineers (28%), followed by building contractors/subcontractors (24%) and consultants (18%). As for the firm size, the majority of the sample was 242 (57%) SMEs that have less than 50 employees (Small Business Council, 2019), including 23% with 1–5 employees, 20% with 6–19 employees, 8% with 20–49 employees and 6% without any employees. The remaining respondents were from 175 (41%) large businesses with 50 or more employees, including 28% with 100 or more employees and 13% with 50–99 employees. In terms of operation year, most respondents were from more established organizations that have operated for more than 20 years (39%), followed by those that operated for 0–5 years (25%), 5–10 years (20%) and 10–20 years (15%).

Figure 5. Demographic details of survey respondents.

Reliability and validity of constructs

To achieve the recommended AVE, some indicators of digital competence with factor loadings ranging from 0.40 to 0.70 were removed, as suggested by Hair et al. (2019). Following this criterion, the PLS-SEM constructs were modified by removing DT1, DT2, DT3, DT4, DT6, DT7 and DT8. Although the factor loadings of some items (i.e. DT11, DT12, W1 and W2) are still less than the desired value of 0.7, they were kept in the model as the recommended AVE value was achieved (Hair et al., 2021). The modified PLS-SEM model is presented in Figure 6. Table 1 demonstrates that the reliability and convergent validity of the constructs and indicators are accomplished in this study as Cronbach’s alpha and composite reliability values were larger than 0.70, and the AVE values were larger than 0.50 (Hair et al., 2021). Finally, the discriminant validity was established, as presented in Table 2, in which the square roots of AVE are larger than the HTMT ratios.

Figure 6. PLS-SEM model.

Table 1. Reliability and validity results of PLS-SEM.

Construct	Indicators	Factor loading	Cronbach’s alpha	Composite reliability	AVE
Digital competence	DT13	0.802	0.841	0.878	0.509
	DT5	0.750
	DT9	0.724
	DT14	0.710
	DT10	0.706
	DT11	0.654
	DT12	0.633
Digital readiness	–	–	0.825	0.914	0.843
LEi	L2	0.901	0.851	0.909	0.769
	L1	0.877
	E1	0.851
TOWEx	O4	0.825	0.933	0.942	0.577
	O5	0.818
	T2	0.812
	O1	0.802
	T1	0.786
	T3	0.775
	O6	0.774
	O2	0.769
	O3	0.732
	E2	0.727
	W1	0.697
	W2	0.562

Table 2. Discriminant validity results of PLS-SEM.

Construct	Digital competence	Digital readiness	LEi	TOWEx
Digital competence	0.713	0.360	0.236	0.380
Digital readiness	0.335	0.918	–	–
Lei	0.210	–	0.877	0.802
TOWEx	0.368	–	0.703	0.760

Notes: Diagonal and italicized are the square roots of the AVE. Below the diagonal elements are the correlations between the construct's values. Above the diagonal elements are the HTMT ratio of correlation values.

Structural model analysis

The structural model was evaluated using VIF, R², t-values, p-values and Q². The VIF was evaluated first to examine potential multicollinearity concerns. Table 3 presents the VIF values among the dimensions are below 5, indicating the absence of multicollinearity (Hair et al., 2021). R² for digital competence is 0.14, indicating that the theorized model is statistically meaningful. The two components of digital readiness jointly explain a 14% variance in the digital competence among the investigated organizations. The resulting Q² is 0.059, above 0, indicating that the predictive capability is established (Hair et al., 2019).

Table 3. Path estimates from PLS-SEM.

Path relationship	β	SD	t-value	p-value	VIF	Decision
H1: Digital readiness → Digital competence	0.335	0.038	8.714	0.000	–	Supported
LEi → Digital readiness	0.874	0.025	35.158	0.000	1.975	–
TOWEx → Digital readiness	0.960	0.008	117.877	0.000	1.975	–
LEi → Digital readiness → Digital competence	0.293	0.066	2.134	0.033	–	Supported
TOWEx → Digital readiness → Digital competence	0.322	0.059	7.411	0.000	–	Supported
DT5 → Digital competence	–	–	–	–	1.770	–
DT9 → Digital competence	–	–	–	–	1.505	–
DT10 → Digital competence	–	–	–	–	1.763	–
DT11 → Digital competence	–	–	–	–	1.714	–
DT12 → Digital competence	–	–	–	–	1.608	–
DT13 → Digital competence	–	–	–	–	1.903	–
DT14 → Digital competence	–	–	–	–	1.716	–

In further assessment of the goodness of fit, the hypotheses were tested to ascertain the significance of the relationship. The results presented in Table 3 revealed that digital readiness significantly and positively influenced digital competence (β = 0.335, t = 8.714 and p = 0.000). Hence, hypothesis H1 is supported. The results also confirmed that readiness indicators under the LEi and TOWEx components have significant positive relationships with digital competence (β = 0.293 and 0.322 for the two structural paths, and p < 0.05). As such, hypotheses H1a and H1b are supported. Due to the larger path coefficient, the TOWEx component has a more substantial overall effect on digital competence than the LEi component.

ANN and sensitivity analyses

The 15 significant indicators of digital readiness and seven DT of digital competence obtained from the PLS-SEM were implemented as the inputs and outputs in the ANN analysis, respectively. Based on the 10-fold cross-validation, the RMSE values, representing the error in the training and testing phases, were recorded and displayed in Table 4. The mean RMSE values for training and testing are 0.050 and 0.051, respectively, which are relatively small, indicating the consistency of the neural network and accurate prediction (Parhi et al., 2022). Thus, the developed PLS-SEM-ANN model has achieved accurate and reliable results. As presented in Table 4, the predictive accuracy of the ANN model was found to be R² = 30%, which is more than twice the accuracy obtained by the PLS-SEM technique. This suggests that the adopted PLS-SEM-ANN approach is more effective in consistently articulating endogenous constructs than the stand-alone PLS-SEM technique.

Table 4. RMSE values from ANN analysis.

Network	RMSE (Training)	RMSE (Testing)	R^2 of PLS-SEM	R^2 of ANN
ANN1	0.056	0.062	14%	30%
ANN2	0.055	0.042
ANN3	0.048	0.038
ANN4	0.045	0.051
ANN5	0.048	0.051
ANN6	0.060	0.057
ANN7	0.049	0.052
ANN8	0.045	0.043
ANN9	0.046	0.050
ANN10	0.047	0.062
Mean	0.050	0.051
SD	0.0049	0.0077

The results of the sensitivity analysis are presented in Table 5. The relative importance of each readiness indicator in the input layer was calculated by using Equations (1) and (2). It is noted that the importance of each DT in the output layer was taken as the average for the seven DT identified in the modified PLS-SEM model. Additionally, the mean importance (MI) is used against the highest value to calculate the relative normalized importance (NI) in percentage. The values of NI range from 5% to 100%, with higher values indicating more significant importance. Among the indicators in the two components of digital readiness, the most influential indicator is E1 (NI = 100%), followed by L1 (NI = 88%), T3 (NI = 85%) and O1 (NI = 73%). Findings also reveal that O3 (NI = 5%) has the most minor effect on digital competence. The total MI for the TOWEx and LEi components are 4.02 and 2.24, respectively, showing that indicators in the TOWEx component have a more significant effect than those in the LEi component on digital competence, confirming the PLS-SEM findings. Comparing between the significance of indicators in the two components, indicators E1 and L1 in the LEi component are more significant than those in the TOWEx component in predicting digital competence. The remaining indicator L2 in the LEi component is also more significant than most of the indicators in the TOWEx component except for T3, O1 and W2.

Table 5. Importance of readiness indicators from sensitivity analysis.

Construct	ANN										MI	NI (%)
	1	2	3	4	5	6	7	8	9	10
T1-TOWEx	0.08	0.11	0.11	0.13	0.13	0.21	0.28	0.27	0.22	0.24	0.18	18
T2-TOWEx	0.29	0.42	0.44	0.43	0.38	0.34	0.31	0.31	0.16	0.17	0.33	34
T3-TOWEx	0.47	0.77	0.85	0.87	0.88	0.90	0.98	0.98	0.77	0.81	0.83	85
O1-TOWEx	0.47	0.61	0.71	0.76	0.80	0.85	0.79	0.82	0.61	0.64	0.70	73
O2-TOWEx	0.05	0.29	0.36	0.41	0.42	0.46	0.49	0.51	0.42	0.46	0.39	40
O3-TOWEx	0.07	0.02	0.03	0.01	0.01	0.00	0.07	0.11	0.10	0.09	0.05	5
O4-TOWEx	0.02	0.04	0.00	0.06	0.17	0.22	0.20	0.25	0.26	0.28	0.15	15
O5-TOWEx	0.21	0.28	0.36	0.40	0.37	0.33	0.25	0.29	0.13	0.14	0.28	28
O6-TOWEx	0.09	0.13	0.18	0.19	0.17	0.17	0.10	0.13	0.10	0.11	0.14	14
E2-TOWEx	0.33	0.37	0.29	0.25	0.25	0.30	0.38	0.39	0.36	0.39	0.33	34
W1-TOWEx	0.03	0.08	0.14	0.17	0.23	0.27	0.33	0.37	0.30	0.30	0.22	23
W2-TOWEx	0.22	0.34	0.40	0.43	0.50	0.51	0.48	0.51	0.40	0.44	0.42	44
E1-Lei	1.00	1.00	1.00	1.00	1.00	1.00	1.00	1.00	0.85	0.84	0.97	100
L1-Lei	0.77	0.70	0.66	0.72	0.80	0.87	0.96	1.00	1.00	1.00	0.85	88
L2-Lei	0.26	0.46	0.50	0.46	0.47	0.44	0.47	0.48	0.32	0.35	0.42	43

Note: MI: mean importance; NI: normalized importance.

Discussion

Insights of the PLS-SEM-ANN model

The findings of this study indicate that all the investigated digital readiness indicators are significant in predicting digital competence. Among these indicators, organizational culture (E1) was identified by the ANN analysis as the strongest predictor of digital competence. This indicator is followed by leadership perception (L1), hardware & software systems (T3) and strategy plan (O1). The leading position of organizational culture in affecting digital competence is consistent with its effect on digital readiness identified by Chen et al. (2023). This finding further addresses the interconnection between digital readiness and digital competence, highlighting the importance of addressing organizational culture as a fundamental aspect of digital transformation initiatives. The organizational culture could involve a willingness to change, openness to innovation, continuous professional development and social collaboration (Razkenari & Kibert, 2022). A positive and receptive organizational culture towards technology can foster innovation, collaboration and a willingness to learn and improve, ultimately improving digital competence (Martínez-Caro et al., 2020). Therefore, the results reported in this article reinforce the leading role of organizational culture in affecting digital competence in AEC organizations, addressing the gap in research regarding the limited focus on organizational culture during digital transformation (Martínez-Caro et al., 2020; Razkenari & Kibert, 2022).

The existing literature has explored the significant role of leadership or top management in technology implementation in organizations (Ijab et al., 2019; Razkenari & Kibert, 2022). This is also consistent with the findings in Chen et al. (2023) which indicated that leadership support and perception are among the top three indicators of the digital readiness of organizations. The findings from the current research provide empirical evidence supporting this notion and further enrich the body of knowledge by highlighting the relative importance of this indicator compared with other indicators. The technology perceptive view of top management could significantly influence employees’ dominant values, beliefs, and norms and form a positive organizational culture toward technology knowledge and practice (Hu et al., 2012). Therefore, the effectiveness of integrating the leadership dimension of RWCM into the TOE framework in examining essential indicators of organizational digital readiness was supported.

The dual-stage analysis results also recognize having reliable and effective hardware & software systems as third influential indicator of organizational digital competence. Nevertheless, the results of Chen et al. (2023) did not fully appreciate the significance of this indicator for organization digital readiness, indicating the advantage offered by the PLS-SEM-ANN approach. The findings from this research underscore the importance of prioritizing investments in advanced hardware & software systems to support digital transformation efforts. Existing literature, such as Chen et al. (2019) and Ijab et al. (2019), also emphasized the foundational position of hardware & software in the application of DT in the AEC industry. It is imperative for organizations invest in high-quality hardware & software systems that can support the specific needs of AEC projects.

Besides, the results reported in this research also addressed the significant effect of a well-defined strategy plan on digital competence compared to the remaining indicators. The importance of this indicator has also been recognized in existing literature such as Ngo et al. (2020) and Sukanthan Rajendra et al. (2022). While Chen et al. (2023) recognized the strategy plan as a critical indicator of digital readiness, it was not ranked particularly high. The current study indicated that a strategic plan, ranked as the fourth most important indicator for digital competence, could serve as a roadmap from the awareness to the practice usage of DT and help align the organization’s goals with its capabilities. One of the key benefits of a strategic plan is that it can help to clearly articulate the goals and objectives regarding the practical usage of DT, formulate the organizational culture to be innovative and ensure consistency and alignment across the organization.

This study adopted a dual-stage PLS-SEM-ANN approach to evaluate the impact of digital readiness on the digital competence of AEC firms. Compared with the usage of PLS-SEM or ANN approaches, the superiority of the usage of the dual-stage PLS-SEM-ANN analysis lies in the improved prediction accuracy and robustness. While traditional PLS-SEM assumes linear relationships among variables, ANN can discern nonlinear relationships, thereby providing more precise predictions for unseen data. By integrating the strengths of PLS-SEM for structural equation modelling and ANN for nonlinear pattern recognition, the dual-stage analysis yields more accurate and dependable outcomes compared to individual methods (Khayer et al., 2020; Zhang et al., 2022). This allows for a more nuanced understanding of the complex interactions between digital readiness factors and digital competence in AEC firms, which may not be adequately captured by linear models. From the PLS-SEM results, the indicators in the TOWEx component with top three factor loadings are O4, O5 and T2. However, the further ANN analysis reveals that the top three indicators in terms of importance levels in the TOWEx component are T3, O1 and W2. Similar findings are observed for the LEi component, with factor loadings ranked as L2, L1 and E1, whereas importance levels are ranked inversely. The ANN analysis includes the interconnections between digital readiness and digital competence. Additionally, the sensitivity analysis of ANN highlights the significance levels of all digital readiness indicators to digital competence, providing insights beyond those obtained from PLS-SEM, which focuses solely on the factor loadings of the two components.

Theoretical implications

The previous research by Chen et al. (2023) developed a technology readiness model to evaluate AEC firms’ readiness for DT adoption via descriptive statistical analysis, exploratory factor analysis and fuzzy synthetic evaluation. The current research contributes theoretically in several ways. First, it extends the prior research by examining the impacts of digital readiness on digital competence. Utilizing the PLS-SEM method, the study revealed that the two components of digital readiness, namely LEi and TOWEx, have a significant impact on digital competence in the organizational context. These findings not only align with existing theories, such as TOE, RBV and RWCM, but also fill the gaps in these theories, which indicated only indirect relationships between digital readiness and digital competence. Moreover, utilizing subsequent ANN analysis to account for the non-linear relationships, this study identifies organizational culture, leadership perception, hardware & software systems and strategy plans as leading contributors to digital competence. These results reinforce the pivotal role of these indicators in promoting innovation and adaptation, aligning with established theories. The improved prediction accuracy further ensures the reliability of these readiness indicators in predicting digital competence compared to existing literature (e.g. Al-Edenat, 2023; Mungra et al., 2024). Additionally, by employing a dual-stage PLS-SEM-ANN approach, this study quantitatively evaluates the relative importance of investigated readiness indicators in predicting digital competence, extending the qualitative relationships found in existing literature (e.g. Konopik et al., 2022; Trenerry et al., 2021; Vieru, 2014). This contributes to advancing the theoretical understanding of organizational readiness indicators in enhancing digital competence.

Practical implications

The research findings may be applied in real-world settings to inform and guide the utilization and investment of multiple DT in AEC organizations. The inclusion of seven DT can help practitioners and policymakers better understand the breadth and depth of DT implementation in the AEC industry. The identification and prioritization of critical indicators for enhancing digital competence may help AEC firms to improve their technology implementation strategies and competitiveness. Identifying the most significant indicators for digital competence could also help AEC practitioners focus their attention and resources on the most critical areas with the greatest potential for improvement. For instance, AEC organizations can develop leadership development programs focusing on cultivating a technology-perceptive culture within organizations and empowering the top management to effectively advocate for digital initiatives. DT providers can also offer training programs, workshops and online resources to promote the utilization of advanced hardware & software. Industry associations play a crucial role in encouraging collaboration and communication between DT experts and industrial leaders, fostering an environment conducive to DT implementation. By offering support and guidance, industry associations can assist AEC organizations in shaping leadership perception, making informed choices regarding hardware & software systems and formulating strategy plans aligned with their digital objectives. Ultimately, these collaborative efforts contribute to the acceleration of digital transformation initiatives within the AEC industry, driving innovation and enhancing overall competitiveness.

Methodological contributions

Methodologically, the adopted dual-stage PLS-SEM-ANN approach demonstrates superior performance compared to the traditional PLS-SEM method. The employment of this approach enables quantitative evaluations of the relative importance of investigated readiness indicators in predicting digital competence with more reliable accuracy than conventional methods (Ng et al., 2022; Wang et al., 2022). This approach captures both the linear and non-linear interactions among the influential factors, improving the understanding of how different factors interact and influence digital competence in a complex context. Furthermore, the sensitivity analysis of ANN enhances the reliability of the ANN model by providing insights into its behaviour and performance under different conditions.

Limitations and future work

This study is subjected to several limitations. To ensure the reliability and validity of the PLS-SEM model, 7 of the 14 DT have been excluded from the modified PLS-SEM model. These excluded DT, including RFID, UAV, IoT, GPS, BIM, Blockchain and CT, have very low factor loadings in the original PLS-SEM model, indicating that they are not strongly related to the measured construct. The exclusion of the seven DT may not necessarily mean that they are not influential factors but did not meet the required statistical criteria to be included in the modified PLS-SEM model. Additional research may be needed to fully understand the impact of these excluded DT on measuring the overall digital competence in the AEC industry. Additionally, readiness indicators with low importance scores may still be important in specific contexts or for certain firms. Accordingly, further research may be necessary to explore the nuances of digital readiness indicators with low importance scores through comparative studies between firms or countries.

Conclusion

This study utilized deep-learning-based dual-stage PLS-SEM and ANN techniques to investigate the effects of 15 digital readiness indicators on the competence towards 14 types of DT in AEC firms. The PLS-SEM analysis revealed that digital readiness positively amplifies competence toward seven types of DT, including immersive technologies, sensing technology, robotics, 3D printing, digital fabrication, artificial intelligence and big data. The ANN sensitivity analysis further revealed that organizational culture is the strongest predictor of digital competence, emphasizing its pivotal role in fostering innovation, collaboration and a willingness to learn and improve within AEC organizations. Moreover, leadership perception, hardware & software systems and strategy plan also significantly contribute to digital competence, underscoring the importance of leadership support, technological infrastructure and strategic planning in driving digital transformation initiatives. By integrating the strengths of PLS-SEM for complex modelling and ANN for nonlinear pattern recognition, this approach enhances prediction accuracy and robustness, providing a more nuanced understanding of the complex interactions between digital readiness factors and digital competence.

The theoretical contributions of this study include an improved understanding of the relationships between digital readiness and digital competence of the AEC organizations, addressing the fragmented nature of digital competence studies. By extending previous research, this study confirms the substantial influence of digital readiness and its components on organizational digital competence. The utilization of the PLS-SEM-ANN approach allows for a quantitative evaluation of the relative importance of digital readiness indicators on digital competence. The implications of the findings mean that organizations can take appropriate measures to enhance their digital competence and effectively navigate the digital landscape. For instance, organizations should give utmost importance to fostering a culture that appreciates and encourages digital innovation. Equally critical is the active demonstration of a positive attitude towards digital transformation by top management. Lastly, it is suggested that organizations in the AEC sector should make substantial investments in state-of-the-art technological infrastructure and formulate a well-defined strategy plan in order to bolster digital capabilities and guarantee a smooth transition towards digital competence.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The authors would like to thank the China Scholarship Council (CSC) and the funding from the Building Research Association of New Zealand (Project number LR12069). Without the funding supports from both sources, this research would not have been possible.

Appendix

Table A1. Digital technologies included in the questionnaire survey.

Table

Radio Frequency Identification (RFID) RFID is the wireless non-contact use of radio frequency waves to transfer data. Tagging items with RFID tags allows users to automatically and uniquely identify and track inventory and assets

Drone/Unmanned Aerial Vehicle (UAV) Drone can be used in surveying or carry out visual inspections of construction areas

The internet of things (IoT) It is a term for a network of devices, sensors, and people connected to the internet that collect and exchange data. Examples in construction might be equipment you can control or read remotely, or site monitoring via sensors on site

Global Positioning Systems (GPS) GPS is a satellite-based navigation system. GPS can be installed in vehicles to locate materials vehicles, workers or equipment

Augmented/Virtual/Mixed Reality (AR/VR/MR) While AR supplements the real world by laying computer-generated data over it, VR aims to introduce users into an artificial real world. AR headset such as the Hololens allows designers and contractors to work with information projected on what the wearer sees in real life and spot construction issues ahead of time and improve communication

Building Information Modelling (BIM) BIM is a process that creates a 3D model that enables document management, coordination, and simulation during the lifecycle of a project. A BIM model can be used to explore design options and to create visualizations that help stakeholders understand what the building will look like before it's built

BlockChain It is a chain of blocks that store digital information about transactional records. Because it is a decentralized system that exists between all permitted parties without needing to pay intermediaries, it can be used for creating smart contracts to allow the performance of credible transactions without third parties to save time and conflict

Cloud-Based Technology The ‘cloud’ is the ability to host a software platform or service from a remote location that can be freely accessed and used anywhere via Internet access

Sensor technology Sensors are used to gather data. Some are used to monitor the properties of concrete during curing, while others are used to track environmental conditions or project equipment. Other types of sensors have been developed to increase workers’ safety on the jobsite by monitoring potentially dangerous conditions such as heat, noise levels, dust, and the presence of hazardous fumes

Robotic technology It is to automate processes. With robotic technology, you can expect traditional construction activities like welding, material handling, packing, dispensing, cutting and packing to be fully automated

3D printing, or additive manufacturing (AM) It is the construction of a three-dimensional object from a CAD model or a digital 3D model. There are 3D printed moulds for concrete casting and 3D printed concrete structure is underway in New Zealand

Digital fabrication This technology is based on a combination of computational design methods and automated construction processes. Examples in construction might be equipment to fabricate building components automatically according to design principles and algorithms

Artificial intelligence (AI) AI (machine learning or deep learning) is a wide-ranging branch of computer science concerned with building smart machines capable of performing tasks that typically require human intelligence. Machine learning is a subset of AI and it is the study of computer algorithms that improve automatically through experience. Deep learning is a subset of machine learning in AI that has networks capable of learning unsupervised from data that is unstructured or unlabelled

Big data It is a combination of structured, semi-structured, and unstructured data collected by organizations that can be mined for information and used in machine learning projects, predictive modelling, and other advanced analytics applications