Introduction

The growing need to understand trends within vast datasets has attracted significant interest in big data analytics from both academic and industrial sectors (Hariri et al., 2019). BDA refers to techniques used to analyze and acquire intelligence from big data (Gandomi and Haider, 2015). Based on previous meta-analyses, BDA or related concepts are widely believed to have a positive impact on firm performance (FP) (Oesterreich et al., 2022; Ansari and Ghasemaghaei, 2023). However, an early study based on the Brazil national survey has uncovered a surprising outcome: the utilization of big data collection and analysis is negatively correlated with the anticipated benefits in product performance (Dalenogare et al., 2018). Furthermore, a past study suggested that the big data-FP relationship is complex. They have found that when BDA capability reached a specific inflection point, using big data did not enhance FP; instead, it may have a negative impact (Hao et al., 2019). A past study suggested that the success rate of BDA is not as high as some non-academic reports claim. The authors believed that, in practice, many BDA projects fail among firms, although these failures often go unreported (Reggio and Astesiano, 2020). Therefore, this study attempted to identify a feasible mechanism for converting BDA to FP, aiming to help firms realize the value of their BDA investments.

A seminal study argued that organizational agility (OA) may serve as a key mechanism through which information technology (IT) impacts FP (Sambamurthy et al., 2003). OA is frequently defined as a firm’s ability to identify and react to opportunities, threats, and changes (Tallon et al., 2019). A prior meta-analysis has examined the connection between IT capability, OA, and FP. Although this study recognized that OA plays a mediating role in the IT capability-FP relationship, it did not delve into this specific mechanism (Werder and Richter, 2022). This left a research gap for subsequent studies in the information systems (IS) research field. In the studies on BDA, many authors indicated that firms must gain OA from their big data investments to create value and competitive advantage (Vesterinen et al., 2024). Whether OA was a recognized mediator for promoting the BDA-FP relationship required verification. From the dynamic capabilities view (DCV), BDA is considered a low-level dynamic capability, serving as the foundation for acquiring OA. In contrast, OA represents a high-level dynamic capability (Ghasemaghaei et al., 2017). Moreover, dynamic capabilities exhibit a hierarchical structure, with high-order dynamic capabilities building upon low-order ones (Schilke, 2014). Thus, this study posits that OA serves as a crucial mechanism for firms to transform BDA into FP. As shown in Table 1, the current meta-analyses focused exclusively on the relationship between BDA (or similar concepts) and performance. However, no study has synthesized the mediating role of OA in the BDA-FP relationship.

Table 1 Related works.
Full size table

Meta-analysis can identify appropriate moderating variables and attempt to explain differences across different studies (Card, 2015). In meta-analyses conducted in the information systems (IS) research field, NC is a commonly studied moderator (Jeyaraj and Dwivedi, 2020). NC refers to the collective programming of the mind distinguishing the members of one group or category of people from others (Geert Hofstede, 2001). As shown in Table 1, in meta-analyses related to BDA, although NC-related moderating variables were considered, they were merely operationalized as binary variables without an in-depth exploration of NC’s impact on BDA usage and outcomes. Furthermore, most of these studies are based on the resource-based view (RBV) theory, which is considered overly static and fails to account for external environmental changes. In contrast, the DCV theory is regarded as an extension of RBV (Teece et al., 1997). However, scholars tended to overlook that dynamic capabilities are influenced, even constrained, by external factors, such as NC (Schilke et al., 2018). A study suggests that dynamic capabilities are context-dependent, and NC plays a central role in understanding dynamic capabilities (Bitencourt et al., 2020). This study is grounded in the DCV theory. From a macro perspective, each study is conducted within specific countries, making it challenging to avoid the influence of NC on the BDA–OA–FP relationship. Furthermore, current meta-analyses often lack exploration of the moderating role of NC in BDA–OA–FP relationships.

In addition to contextual moderators, methodological moderators are frequently investigated in meta-analyses. In simple terms, methodological moderators are associated with different study design characteristics, which may affect the strength or direction of the meta-analytical estimates (Rudolph et al., 2020). As shown in Table 1, only two meta-analyses on BDA have focused on methodological moderators, such as Oesterreich et al. (2022) and Ansari and Ghasemaghaei (2023), with their primary focus being on the measurement methods of variables and types of responses. However, they have overlooked other methodology-related factors, such as sample size, sampling techniques, time dimensions, and analytical techniques. Suppose the focal relationship in a meta-analysis is influenced by certain methodological characteristics not included in the actual analysis. In that case, this unobserved heterogeneity will obscure the accurate estimation of parameters (Rudolph et al., 2020). Therefore, this study will continue exploring methodological moderators’ impact on the BDA-FP relationship.

In summary, there were currently two main research gaps in the meta-analysis regarding BDA. Firstly, existing meta-analyses focused on the direct relationship between BDA or similar concepts and performance while overlooking the existence of mediating effects. Ansari and Ghasemaghaei (2023) suggested that future meta-analyses should synthesize mechanisms through which BDA impacts FP. In response to this, our study aimed to investigate the mediating role of OA in the BDA-FP relationship. We constructed a research model based on the DCV and validated it using meta-structural equation modeling (MASEM). MASEM complements the limitations of meta-analysis and structural equation modeling, as meta-analysis cannot perform multivariate analysis, and structural equation modeling cannot replicate the same study (Bergh et al., 2016).

Secondly, most existing meta-analyses on BDA have been grounded in the RBV theory, focusing solely on the moderating effect of NC on the BDA- FP relationship. However, these studies overlook the dynamic nature of the business environment. This study, possibly the first meta-analysis in this field, is based on the DCV theory. It explores the boundary conditions of this theory from a cultural perspective. Therefore, our study aimed to investigate the moderating effect of NC on the BDA–OA–FP relationship. We obtained scores for national culture from Hofstede’s cultural dimensions and analyzed them using meta-regression. Unlike previous meta-analyses that simply have dichotomized countries into broad categories (e.g., Eastern vs. Western), meta-regression analysis allows for a more nuanced comparison across different countries. Additionally, this study investigated methodological moderators that may exist in the BDA-FP relationship. These methodological moderators include sample size, sampling techniques, time dimensions, and analytical techniques. This exploration sheds light on new potential reasons for the discontinuous nature of the BDA-FP relationship.

This study will significantly contribute to the DCV theory in IS research. Firstly, it establishes OA as a crucial mechanism through which BDA influences FP. Secondly, it explores potential boundary conditions of DCV theory from a cultural perspective. This study also provides valuable insights for firms seeking to realize the investment value of BDA. The rest of this study discusses the theoretical background of variables, followed by hypothesis development, conceptual framework, methodology, findings and discussion. Then, theoretical and practical implications will be further discussed. Finally, the rest presents limitations and future directions.

Theoretical background

Dynamic capabilities view theory

This meta-analysis was grounded in the dynamic capabilities view (DCV). The DCV outlines how firms can respond to swift environmental shifts and generate new business value by integrating, building, and reconfiguring both internal and external resources and knowledge (Teece et al., 1997, 515). Dynamic capabilities are high-level organizational capabilities that contain a set of capabilities; they are (1) sensing new opportunities in the changing environment, (2) seizing the identified opportunities, and (3) transforming to better exploit the new opportunities (Teece, 2007a). The achievement of an organization greatly relies on the competitive advantage derived from enhancing inherent adaptive responses to external environmental changes (Gao and Sarwar, 2022).

Big data analytics as a low-order dynamic capability

BDA is defined as tools and processes often applied to large and dispersed datasets for obtaining meaningful insights (Ghasemaghaei et al., 2015, 1). From the perspective of DCV, the use of BDA can be regarded as an organizational information processing capability that reduces uncertainty and improves the organization’s strategic decision-making capability by stimulating insight and knowledge creation (Chen et al., 2015). The value of BDA hinges on an organization’s ability to use big data’s key dimensions (such as volume, variety, velocity, and veracity) to discern opportunities and threats (sense), allocate resources effectively to capitalize on these opportunities, and mitigate threats (seize), and adapt tangible and intangible assets to ensure ongoing organizational renewal (transform) (Conboy et al., 2020). Therefore, the use of BDA can enhance dynamic capabilities, namely, sense, seize, and transform. Consequently, it is considered a low-order dynamic capability.

Organizational agility as a high-order dynamic capability

OA refers to the capacity of an organization to efficiently and effectively redeploy/redirect its resources to value-creating and value-protecting (and capturing) higher-yield activities as internal and external circumstances warrant (Teece et al., 2016a, 17). From the perspective of DCV, OA is conceptualized as the capacity to sense changes and respond effectively (Overby et al., 2006; Liang et al., 2017). Sensing consists of scanning the environment to detect, capture, and interpret organizational opportunities or threats. Responding represents an organizational ability to mobilize and transform resources to react to the opportunities it senses (Oosterhout et al., 2006). Strong dynamic capabilities are necessary to cultivate the OA required to cope with deep uncertainty (Teece et al., 2016b). Fundamentally, the value of dynamic capabilities can be assessed based on their contribution to supporting OA (Mikalef and Pateli, 2017). Thus, OA can be perceived as a high-order dynamic capability.

National culture in dynamic capabilities view theory

NC refers to the collective programming of the mind distinguishing the members of one group or category of people from others (Geert Hofstede, 2001, 9). Culture is acquired and derived from the societal milieu. Each culture is delineated by patterns typically imperceptible to outsiders, including symbols, norms, beliefs, attitudes, behaviors, and values (Hofstede et al., 2010). In collectivist and restrained cultures, society faces greater supervision and constraints, and sensing, seizing, and reconfiguring activities are seen as departures from established norms due to their unconventional and innovative nature (Taras et al., 2010). In a culture with low power distance and high femininity, nomocracy, corporate board efficacy, effective resource management, free market access policies, and liberal corporate governance mechanisms can provide an enabling environment for firms to configure, deploy, and utilize their dynamic capabilities (Teece, 2007b). Therefore, NC is an important factor that affects dynamic capabilities.

Dynamic capabilities exhibit a hierarchical structure, with high-order dynamic capabilities building upon low-order ones (Schilke, 2014). In this study, BDA represented a low-order dynamic capability, while OA was a high-order dynamic capability. Ultimately, strong dynamic capabilities confer competitive advantages to organizations. Moreover, the NC could affect the BDA–OA–FP relationship. Therefore, DCV was chosen as the underpinning theory for this study.

Measurement of variables in this study

The operational definitions of the variables in this study are presented in Table 2.

Table 2 Operational definitions of variables in this study.
Full size table

Hypothesis development

The big data analytics-firm performance relationship

Big data is an information asset characterized by such a high volume, velocity, and variety to require specific technology and analytical methods for its transformation into value (De Mauro et al., 2016). The true value of big data lies in a firm’s analytical and interpretative capabilities. The integration of artificial intelligence, machine learning, and advanced database technologies facilitates the visualization and analysis of big data, thus yielding actionable insights in real-time. BDA enables firms to fully leverage their data assets, seize new opportunities, and innovate their business models (SAP, 2023). The employment of BDA has emerged as a crucial avenue for firms to surpass their competitors (Shan et al., 2019). A study based on 814 publicly traded U.S. firms indicated that the live BDA assets were linked to an average improvement in firm productivity, yielding an increase ranging from 3 to 7% (Müller et al., 2018). Furthermore, a study based on data from the U.S. Census Bureau discovered that factories extensively implementing predictive analytics of BDA within the manufacturing sector experienced an average productivity increase of 1–3%. This enhancement equals to an approximate rise in sales revenue of $464,000 to $918,000 (Brynjolfsson et al., 2021). In summary, the positive relationship between BDA and FP has been widely validated. Therefore, the following hypotheses were made:

H1: BDA has a positive impact on FP.

The mediating role of organizational agility

BDA enhances an organization’s OA, enabling the identification of opportunities and threats within the environment (the sensing process), the capitalization of these opportunities, and rapid response to threats (the response process) (Barlette and Baillette, 2022). BDA can enhance sensing capabilities to perceive changes in the environment. This is primarily achieved by applying various algorithms to analyze the collected data, thereby identifying different patterns, relationships, and trends within huge datasets (Feng et al., 2019). Furthermore, based on feedback from sensing, BDA augments a firm’s responding capabilities by enabling swift and efficient decision-making. BDA can assist firms in decision-making by identifying actions and assessing their impact on business objectives, requirements, and constraints (Sivarajah et al., 2017). For instance, simulation was often used in BDA to support decision-making. It was particularly beneficial for testing new ideas about business and managerial decisions, as it can demonstrate the ultimate impact of various conditions and alterations on existing processes or systems (Sheng et al., 2021). Therefore, BDA provides technical support for the development of OA. BDA empowers firms to quickly sense threats and opportunities in the environment and assists them in making efficient decisions to seize these opportunities or respond swiftly to threats, thereby helping them enhance performance. Therefore, the following hypotheses were made:

H2: BDA has a positive impact on OA.

H3: OA has a positive impact on FP.

H4a: OA mediates the relationship between BDA and FP.

The mediating role of different levels of organizational agility

BDA can be applied in various domains, such as new product development, process improvement, and strategic planning. This study suggested that these applications can enhance different levels of OA, namely the organizational level (organizational and strategic aspects) and the operational level, ultimately leading to improved FP. A past study found that effective use of BDA enhances customer sensing and responding capabilities, which in turn contributes new product success. The author suggested that BDA assists firms in understanding target customers and identifying demand changes. Based on this feedback, they can rapidly respond to demand changes and provide appropriate services or products for the market (Tseng, 2023). A case study demonstrated that using BDA for new product development can effectively reduce costs and shorten the research and development cycle. By applying BDA, firms can launch a series of new products within a 5-month timeframe, costing $2 million. In contrast, competitors using the traditional approach must invest approximately $20 million over 12 months to achieve a similar product lineup (Zhan et al., 2018).

According to a 2022 survey conducted by the Institute of Management Accountants (IMA) in China, respondents believed that big data technology plays a crucial role in transforming business processes and facilitating interdepartmental collaboration, thereby significantly supporting OA. With the application of digitalization, firms were allowed to gain real-time access to internal data. BDA can transform this data into valuable insights, supporting internal process reengineering or optimization. Consequently, this enhanced operational efficiency and ultimately improves overall performance (Ma and Qi, 2022). An illustrative case demonstrated that firms leveraging predictive analytics of BDA have enhanced supply chain performance and streamlined production processes to achieve agility. Specifically, the supplier on-time delivery rate increased from 92 to 97%, the inventory turnover rate improved from 6 turns per year to 8 turns per year, and the production cycle time decreased from 8 h to 6 h. Additionally, the defect rate declined from 3% to 1.5%. These improvements resulted in cost savings of $750,000 and a 10% increase in sales (Aljohani, 2023).

A study indicated that firms promoting robust BDA usage to drive strategies and provide informed decisions for top management. Insights generated through BDA enabled managers to better sense opportunities and threats while also serving as an anchor for decision-making (Mikalef et al., 2020). This will promote OA and enhance FP. A case study indicated that BDA can assist firms in strategic management. BDA leveraged algorithms to analyze internal and external data (such as financial reports), construct mathematical models, and generate SWOT analysis reports to help firms identify critical issues they face (Pröllochs and Feuerriegel, 2020). As previously mentioned, case studies demonstrate that firms can achieve operational benefits in the short term by optimizing their business processes and shortening the research and development cycle through using BDA. While BDA assists in formulating strategies, the implementation of these strategies is time-consuming. Moreover, in today’s external business environment, uncertainty is high due to various black swan events such as COVID-19 and the Russo-Ukrainian War. These events can impact a firm’s long-term strategy. Unlike altering business processes, adjusting strategy requires complex deliberation and cannot be achieved in the short term. Ultimately, the effects of strategic implementation will only materialize in the future. Therefore, this study posits that BDA is better suited to enhancing process-level agility than organizational-level agility. Based on this, the following hypothesis was made:

H4b: OPE has a stronger mediating effect than ORG in the BDA-FP relationship.

National culture dimensions as contextual moderators

Numerous studies indicate that differences in NC lead to variations in the use and outcomes of IT (Saunders et al., 2015). Similarly, due to big data technologies being used in diverse ways, culture inevitably plays a role in implementing and utilizing BDA (Youssef et al., 2022), thus influencing the outcomes produced. Therefore, this study suggests that the BDA–OA relationship and the OA–FP relationship could be moderated by cultural dimensions (PDI, IDV, MAS, UAI, LTO, and IND).

PDI measures the degree to which the less powerful members of a society accept and expect that power is distributed unequally (Geert Hofstede, 2011). The degree of support or resistance to IT application and usage varies based on the level of PDI (Ribière et al., 2010). An early study found that individuals in high-PDI countries exhibited lower innovation and less trust in technology than those in low-PDI countries (Srite, 2000). Moreover, in low PDI societies, decentralized and flat organizational structures are typical, and employees have the right to participate in decision-making processes. In contrast, in high PDI societies, centralized and authoritarian organizational structures prevail, and decision-making is the responsibility of management (McRay, 2015). Most agility traits require an organizational structure that promotes information sharing, empowerment, and lateral coordination, which typically aligns with decentralization (Adana et al., 2024). Therefore, in low power distance societies, firms have greater trust in big data technology and are more willing to use information obtained from BDA in their management practices. When using information provided by BDA for decision-making, employees can actively participate in the decision-making process and offer additional ideas for management consideration, which will promote OA and ultimately improve FP. Therefore, this study proposed the following hypotheses:

H5a: Low PDI promotes the BDA–OA relationship.

H5b: Low PDI promotes the OA–FP relationship.

IDV is defined as people looking after themselves and their immediate family only versus people belonging to in-groups that look after them in exchange for loyalty (De Mooij and Hofstede, 2010). The study found that IDV is beneficial for promoting innovation at the national level, as individualistic cultures are more inclined to adopt new technologies for innovation (Taylor and Wilson, 2012). Furthermore, past studies showed that IDV provides a more tolerant environment and incentives for future innovation by giving managers and employees more freedom. Additionally, it preferred more flexible and concise communication channels, and small organizations with lower levels of work alienation (Bukowski and Rudnicki, 2019). Therefore, in individualistic societies, firms pursue innovation, are more open-minded about big data technology, and are more willing to try to use it in their management practices. Additionally, within this cultural context, organizational atmospheres are freer, communication is more flexible, and all these factors contribute to the effective use of BDA, thereby promoting agility and enhancing performance. Thus, the study proposed the following hypotheses:

H6a: High IDV promotes the BDA–OA relationship.

H6b: High IDV promotes the OA–FP relationship.

MAS is defined as the dominant values in a masculine society are achievement and success; the dominant values in a feminine society are caring for others and quality of life (De Mooij and Hofstede, 2010). An early study suggested that in a country with higher MAS, if information and communication technology enhance performance, increase opportunities for success, and support competition, there will likely be a positive attitude towards the implementation of these technologies (Kovačić, 2005). The study found that in organizations with a high MAS contest, the implementation of organizational change programs was met with greater resistance because of members’ reluctance to change (Rawski and Workman-Stark, 2018). Conversely, in feminine cultures, members were adapting to changes in goals and objectives. And responding to change, welcoming change was an expression of agile values (Zykov and Singh, 2020). Moreover, a study revealed a negative association between masculine traits and FP. MAS may drive management to make suboptimal decisions, influenced not by rational factors but by unhealthy competition among its members (Martín-Ugedo et al., 2019). In masculine cultures, due to the competitive advantage that BDA can offer, firms are more likely to incorporate it into their management practices. However, when using BDA for certain organizational changes, employees tend to resist the changes in this environment, and sometimes internal negative competition may arise. Even if BDA leads to some degree of OA, substantial performance improvement remains challenging. Therefore, the study proposed the following hypotheses:

H7a: High MAS promotes the BDA–OA relationship.

H7b: High MAS hinders the OA–FP relationship.

UAI is defined as the extent to which people feel threatened by uncertainty and ambiguity and try to avoid these situations (De Mooij and Hofstede, 2010). Low UAI cultures accept higher levels of risk and do not attempt to control uncertainty. Conversely, high UAI cultures seek to develop methods to control future events, thereby reducing uncertainty and risk (Lee et al., 2013). Countries with a high degree of UAI are less likely to adopt new IT, and their perception of project risks is higher than countries with a low degree of UAI (Saunders et al., 2015). Organizations with low UAI typically lack formal constraints and exhibit lower change management capabilities. Conversely, organizations with high UAI tend to be rule-oriented and have established change management processes. Therefore, as UAI increased, control over change became stronger, leading to greater agility in their use of information systems (Woolley and Hobbs, 2008). In low UAI cultures, firms exhibit higher tolerance for risks related to BDA and may adopt BDA early on. However, these firms often lack strong change management capabilities, making agility challenging during certain organizational transformations using BDA. Conversely, in high UAI cultures, although BDA adoption is cautious, firms demonstrate strong control over change during later stages of implementation, leading to improved agility and, ultimately enhanced performance. Therefore, the study proposes the following hypothesis:

H8a: High UAI promotes the BDA–OA relationship.

H8b: High UAI promotes the OA–FP relationship.

LTO is defined as the extent to which a society exhibits a pragmatic future-orientated perspective rather than a conventional historic or short-term point of view (De Mooij and Hofstede, 2010). Short-term orientation cultures pay less attention to tradition and are more open to new ideas. Therefore, in these countries, the adoption rate of new technologies is expected to be higher than in LTO cultures (Erumban and de Jong, 2006). Moreover, the study indicated that short-term-oriented firms focused on achieving quick results for their efforts or strategies and that short-term orientation allowed them to adjust their strategies when deviations occur (Gerlich, 2023). Moreover, the study argued that in a volatile business environment, where products and technologies are rapidly being upgraded, and short-term orientation is conducive to innovation, this helps firms quickly respond to these pressures (Nguyen et al., 2024). However, a study indicated that agile transformation is not a short-term or fleeting trend. When organizations evolve from “doing” agile to “being” agile, they must learn and develop specific capabilities (Gerster et al., 2020). In LTO societies, it was common practice to plan for future life. This tendency may strongly encourage firms to establish a long-term vision and consider long-term returns (Zellweger, 2007). This implies that maintaining agility through BDA over the long term and benefiting from big data investments require LTO. Therefore, the study proposed the following hypothesis:

H9a: LTO hinders the BDA–OA relationship.

H9b: LTO promotes the OA–FP relationship.

IND is defined as indulgence, stands for a tendency to allow relatively free gratification of basic and natural human desires related to enjoying life and having fun; restraint reflects a conviction that such gratification needs to be curbed and regulated by strict social norms (Hofstede et al., 2010). A previous study indicated that firms in high-indulgent cultures exhibit greater readiness to adopt new technologies compared to those in lower-indulgent cultures (Syed and Malik, 2014). The authors posited that indulgent societies may foster innovation to consistently fulfill people’s desires related to enjoyment and life satisfaction. Compared to restrained societies, organizations in indulgent cultures are more tolerant of existing procedural deviations, experience less resistance to change, and exhibit less rigidity (G. Hofstede et al., 2010). In other words, in indulgent societies, firms are more willing to explore new technologies. This is beneficial for the adoption of BDA and using the insights provided by BDA to assist them in management. Additionally, in such an environment, firms demonstrate high flexibility and are more willing to change, which will further promote agility and enhance performance. Therefore, the study proposed the following hypothesis:

H10a: IND promotes the BDA–OA relationship.

H10b: IND promotes the OA–FP relationship.

Methodological factors as moderators

As shown in Table A2, we identified six factors relevant to the research design after coding the samples. Due to the lack of differentiation among specific industries and firm sizes in many studies, it is impossible to perform accurate coding in the subsequent procedure. Therefore, sample size (small vs. large), time dimension (cross-section vs. longitudinal), sampling technique (probability vs. non-probability), and analytical technique (PLS-SEM vs. CB-SEM) will be selected as methodological moderators. Considering sample size is important because studies with small samples may lead to inflated effect sizes due to their homogeneity (Fern and Monroe, 1996). Moreover, a previous meta-analysis indicated that effect sizes can differ based on research design, with cross-sectional studies frequently showing more robust correlations (Hui et al., 2020). A study suggests that if samples of studies are obtained through convenience sampling when these samples are included in a meta-analysis, it may lead to the super-population not being a representative probability sample for that field, thereby affecting the estimation of effect size (Hak et al., 2016). Regarding analytical techniques, scholars typically choose between PLS-SEM or CB-SEM based on research needs. Although both methods have their advantages and disadvantages, currently, no meta-analysis has demonstrated an impact on effect size. Therefore, further exploration is necessary. Thus, the study proposed the following hypothesis:

H11a-d: Methodological moderators (sample sizea, cross-section vs. longitudinalb, probability vs. non-probability samplingc, and PLS-SEM vs. CB-SEMd) can moderate the BDA–OA relationship.

Conceptual framework

Based on the DCV theory, we constructed a conceptual framework, as depicted in Fig. 1.

Fig. 1
figure 1

Conceptual framework.

Full size image

Methodology

Study search and selection

Meta-analysis is the statistical synthesis of results from a series of studies (Borenstein, 2009). Meta-analyses bolster statistical power and generalizability compared to individual studies by deriving effect size estimates from a substantial body of research conducted over an extended period (Chang et al., 2016). To maximize the power, reliability, and generalizability of meta-analysis, researchers should make efforts to collect as many relevant study results as possible for meaningful comparisons, while avoiding selection bias (Eisend and Tarrahi, 2014). Moreover, as BDA belongs to interdisciplinary research, it involves management science, computer science, and other disciplines. It emphasizes the need for maximum coverage and inclusivity in sample selection. To achieve this, the study adopted a comprehensive search strategy (Veloutsou et al., 2022; Veloutsou and Liao, 2023).

This study conducted searches across multiple databases in various languages, including Web of Science, Scopus, China National Knowledge Infrastructure (CNKI), Korea Citation Index (KCI), Scientific Electronic Library Online (SciELO), Cairn.info, and Airiti Library (Airiti). Given that the DCV theory was proposed by Teece in 1996, and research related to IT-OA–FP began after 2000 (e.g., Sambamurthy et al., 2003), the search start date has been set from 1990 to 2024. Moreover, we also conducted a reference-by-reference search based on previous meta-analyses such as those (Bogdan and Borza, 2019; Almeida and Low-Choy, 2021; Anton et al., 2021; Oesterreich et al., 2022; Ansari and Ghasemaghaei, 2023). Finally, after completing database and reference-by-reference searches, Google Scholar was used to fill gaps and achieve a maximal collection of relevant papers. The search terms employed referred to previous meta-analyses included keywords such as “big data/big data analytics/big data analytics capability/business analytics” and their combination with “agility” and “performance/value/growth/benefits. Additionally, when using CNKI and Airiti, corresponding Chinese keywords should be employed.

After completing the literature search, two of the authors refined the papers’ selection using the following criteria. First, they must align with this study’s definition of big data, including three key zero-order relationships: BDA-FP, BDA–OA, and OA–FP (or similar concepts) in each paper. Moreover, all papers must be empirical studies employing quantitative methods. Therefore, qualitative studies, case analyses, and reviews were excluded. Throughout this process, individual scale items underwent a rigorous evaluation to ascertain that the authors of the papers used measurements aligned with the constructs of interest in this study. Second, they must provide sample size and correlation coefficients, or other metrics can be converted into correlation. Third, the unit of analysis focused on the organizational level in a specific country. Thus, regional or global studies were excluded. Finally, given that this study involved papers in multiple languages, it was imperative to ensure the absence of duplicate publications, whereby the same article was published in different journals using different languages while maintaining theoretical frameworks, samples, and measurements.

This meta-analysis adhered to the PRISMA guidelines (Page et al., 2021), and the specific article selection process also referred to (Maseeh et al., 2021; Ashaduzzaman et al., 2023), which was illustrated in Fig. 2. Finally, a total of 34 studies with 42 independent samples were coded for the subsequent meta-analysis. The minimum required sample size for MASEM is at least 10 studies, while 30 studies suffice for obtaining adequate parameter estimates and standard errors. However, the specific sample size should be determined based on the complexity of the research model (Jak et al., 2021). Consequently, the sample size in our study aligns with this requirement. Among these, 29 articles were in English from multiple databases, 3 in Chinese from CNKI, and 2 in Korean from KCI. For additional details, please refer to Table A2.

Fig. 2
figure 2

PRISMA workflow.

Full size image

Coding procedure

In this study, a coding sheet was created using Microsoft Excel, which comprises author(s)/year, sample size, r-value (correlation coefficient), Cronbach’s α, performance type, agility type, country, and scores of Hofstede’s cultural dimensions. If more than 75% of items within each construct closely align with our defined criteria, we categorized that construct as a relevant dimension of performance type (Schmidt and Hunter, 2004). However, if the FP scale is unidimensional and combines financial and non-financial indicators without a set of indicators dominating by more than 75%, it is coded as MXP.

In cases where second-order or multidimensional variables are used in research, we calculate the average correlation coefficient and reliability of the sub-constructs to obtain zero-order correlations and reliability (Borenstein, 2009), thereby mitigating bias. Moreover, if the authors did not provide zero-order correlations, we followed the solution offered by (Singhal et al., 2020) to employ the partial correlation coefficient between two variables from results in structural equation modeling (SEM). Additionally, when studies used linear regression, we followed the recommendation made by (Peterson and Brown, 2005) to convert the regression coefficient to the correlation coefficient. Finally, average weighted reliability was used as a substitute for studies that did not report reliability. Studies lacking these crucial statistical parameters has been annotated in Table A2.

Regarding the cultural moderators, we used Hofstede’s cultural dimensions developed by Geert Hofstede, containing six dimensions (PDI, UAI, IDV, MAS, LTD, and IND) for 109 countries. Each dimension is scored from 0 to 100. This study identified the source country from the data collection section of each paper, and cultural dimensions scores for the respective countries were obtained using the “COUNTRY COMPARISON TOOL” provided by Hofstede-insight.com. Furthermore, after reviewing, all countries of each study can be found on the Hofstede website. While methodological moderators are all categorical variables. The sample size is based on the median of all research sample sizes, which is 244. If the sample size exceeds this value, it is categorized as large (L); otherwise, it is categorized as small (S). The time dimension is divided into cross-sectional study (CS) and longitudinal study (LO). Sampling techniques are categorized as probability sampling (P) and non-probability sampling (NP). Analytical techniques encompass partial least squares structural equation modeling (PLS-SEM) and covariance-based structural equation modeling (CB-SEM).

During the coding process, to mitigate the impact of individual cognitive biases, one author coded the articles, and another author conducted a review. In cases of discrepancies in coding, both authors revisited the original texts and engaged in discussions to reach a consensus, subsequently correcting the coding. After the second author’s examination, an initial consistency level of 92% was achieved, and following discussions, consensus was reached, resulting in a final coding consistency of 100%. The coding result is shown in Tables A2, A3.

Meta-analytic procedures

This study used SPSS V.29 to perform basic meta-analysis. Meta-analysis models, including fixed-effect and random-effect models, are widely used in published meta-analyses (Hunter and Schmidt, 2000). The fixed-effect model assumes that all studies in the analysis share a common true effect size, and any observed differences in effect sizes are solely due to sampling error. In contrast, the random-effect model assumes that the true effect size varies across studies, with observed effect sizes differing not only due to within-study variance but also due to between-study variance (Borenstein et al., 2010). The fixed-effect model is typically applied for conditional inference, meaning that conclusions are not generalizable beyond the included studies. Conversely, the random-effect model is used for unconditional inference, allowing conclusions to be generalized beyond the analyzed studies. Therefore, if all studies estimate a common effect size parameter, the fixed-effect model is appropriate. However, if there is evidence of heterogeneity among the effect sizes estimated by different studies, the random-effect model should be employed (Hedges and Vevea, 1998). When there is heterogeneity among studies, a possible explanation is the presence of study-level moderators that account for this heterogeneity. These variables are related to studies characteristics such as the methods, procedures, and samples, etc (Viechtbauer, 2007). Give that the analysis reveals heterogeneity as shown in Table 7, we chose the random-effects model. Moreover, the random-effects model is appropriate for this study because the synthesized studies were conducted in different countries, periods, and with diverse study designs (Oduro and De Nisco, 2023).

Before conducting the analysis, correlation coefficients for each study were corrected based on the reliability (Schmidt and Hunter, 2004). When the reliability coefficient of a certain study was missing, the weighted average reliability was used for substitution (Lipsey and Wilson, 2001). Subsequently, the adjusted correlations were transformed into Fisher’s z values (z) using the FISHER () function in Excel. The variance of Fisher’s z values (vz) also needed to be computed in Excel (Sen and Yildirim, 2022).

Additionally, Egger’s regression test and trim-and-fill methods are used to examine publication bias in SPSS (Sen and Yildirim, 2022). In Egger’s test, assessing whether the intercept is significantly different from zero served as a criterion for determining the presence of publication bias (Egger et al., 1997). If publication bias exists, it becomes imperative to employ trim-and-fill methods to assess the stability of the effect size. The concept underlying the trim-and-fill method initially removes studies contributing to a funnel plot’s asymmetry. This ensures that the overall effect estimate generated by the remaining studies was minimally influenced by publication bias. Subsequently, the method imputed missing studies in the funnel plot based on the bias-corrected overall estimate (Duval and Tweedie, 2000).

This study mainly used Q statistic and I2 to assess the heterogeneity of effect sizes. The Q statistic is based on the null hypothesis that the true effects of all studies are homogeneous. If the Q value is statistically significant and exceeds the degrees of freedom (df), the null hypothesis is rejected, indicating variability between studies. In such cases, a random-effect model must be employed. At the same time, I2 represents the proportion of true variation in effect sizes. When the I2 value exceeds 75%, it suggests a high heterogeneity among studies (Huedo-Medina et al., 2006). If both indicators meet the criterion, moderation analysis should be conducted.

Moderation and mediation analysis procedures

In this study, the moderator NC comprises six dimensions as continuous variables, and weighted least squares (WLS) regression is recommended for its estimation (Schmidt and Hunter, 2004). WLS regression is employed to explicate and synthesize the estimated regression coefficients, as it accommodated both heteroscedasticity and excessed heterogeneity (Stanley and Doucouliagos, 2017). The WLS approach necessitates computing the inverse variance weight for each effect size. This study followed the steps from (Shan et al., 2019). WLS was performed using the SPSS. Moreover, methodological moderators are categorical variables more suitable for subgroup analysis. In meta-analysis, studies can be stratified into subgroups based on specific characteristics, enabling the comparison of the magnitude of effect sizes under different conditions (Borenstein and Higgins, 2013).

This study used correlation-based meta-analytic structural equation modeling (MASEM) for the mediation analysis. By employing MASEM, researchers can aggregate information from multiple studies to analyze a single model depicting relationships among multiple variables. Alternatively, it allowed for comparing several models supported by different studies or theories to derive the optimal model (Viswesvaran and Ones, 1995). Initially, a correlation matrix was computed. The mean effect size estimates from the SPSS report were transformed into correlation coefficients using the FISHERINV() function in Excel. Subsequently, the correlation matrix was input into Mplus 8. Compared to other software, Mplus can be used for analyzing univariate, multivariate, and three-level meta-analyses. Both fixed-effect and random-effect models were applicable to the analysis (Cheung, 2015). Given that our study was based on a random-effect model and included a mediator, Mplus was suitable. Moreover, the harmonic mean of the total sample size was used as the input for the sample size in Mplus 8 (Viswesvaran and Ones, 1995). Additionally, to compare the mediating effects of different models, we adopted an approach suggested by (Steinmetz and Block, 2022) that involved grouping by OA and conducting separate MASEM analyses, and finally compared the results using descriptive methods or statistical tests. All the formulas and functions are shown in Table 3.

Table 3 Descriptions of formulas/functions used in this study.
Full size table

Publication bias and outliers

Publication bias tests using two methods are shown in Tables 4, 5. The results of Egger’s regression were all significant for BDA-FP, BDA–OA, and OA–FP relationships (p < 0.001), revealing publication bias in this study. However, the trim-and-fill analysis yielded a different result. It revealed that, after imputing ten samples into the BDA-FP relationship, the adjusted effect size increased to 0.627 (p < 0.001, 95% CI: 0.530, 0.723). The effect size after imputation did not remarkably increase compared to the pre-correction effect size (0.535), with a marginal difference of 0.092, and there was no reversal in significance. Similarly, after imputing 11 samples into the BDA–OA relationship, the adjusted effect size increased to 0.87 (p < 0.001, 95% CI: 0.753, 0.988). The effect size after imputation did not remarkably increase compared to the pre-correction effect size (0.743), with a marginal difference of 0.127, and there was no reversal in significance. Moreover, after imputing 12 samples into the OA–FP relationship, the adjusted effect size increased to 0.705 (p < 0.01, 95% CI: 0.591, 0.819). The effect size after imputation did not remarkably increase compared to the pre-correction effect size (0.575), with a marginal difference of 0.13, and there was no reversal in significance. These results suggested the combined effect sizes were robust, and there was no publication bias.

Table 4 Egger’s regression-based test.
Full size table
Table 5 Effect size estimates for trim-and-fill analysis.
Full size table

Additionally, we conducted a sensitivity analysis following the approach outlined by Kepes and Thomas (2018). We removed a total of 10 studies with large effect sizes (greater than 1) and studies with negative effect sizes, then reanalyzed the data for comparison. As shown in Table 6, the results indicate that after excluding these studies, the overall effect size reduction remains within 20%, and the impact of outliers on the findings can be negligible. Therefore, the results of this meta-analysis exhibit robustness.

Table 6 Sensitivity analysis.
Full size table

Results

Heterogeneity

This study assessed the direct associations between BDA, OA, and FP using the dataset in Table A1. The homogeneity test is presented in Table 7. First, for the BDA-FP relationship (Q = 1170.367, p < 0.001). The Q value was greater than the df value (41). Moreover, the value of I2 was 96.2%, which exceeded 75%. Then, for the BDA–OA relationship (Q = 1589.514, p < 0.001). The Q value was greater than the df value (41). Moreover, the value of I2 was 97.3%, which exceeded 75%. Finally, for OA–FP relationship (Q = 1504.098, p < 0.001). The Q value was greater than the df value (41). Moreover, the value of I2 was 97.3%, which exceeded 75%. These results indicated high heterogeneity between studies; thus, adopting a random-effect model was appropriate (Hedges and Vevea, 1998). The random-effect model avoids the significant Type I bias often exhibited by the fixed-effect model in the significance testing of mean effect sizes and moderators. Additionally, the confidence intervals for mean effect sizes produced by the fixed-effect model are narrower than their nominal width, thereby exaggerating the precision of the meta-analytic findings (Hunter and Schmidt, 2000). This greatly enhances the quality of the results in this study. Furthermore, given the high heterogeneity between the BDA-FP, BDA–OA and OA–FP relationships, this supports the assumptions of the existence of moderators. Therefore, it is necessary to conduct further analysis of the moderating effects.

Table 7 Summary of homogeneity test and main effect.
Full size table

Main effects

As shown in Table 7, the effect size between BDA and FP was 0.535, and the 95% confidence interval (CI) ranged from 0.436 to 0.634 without 0, which is statistically significant (p < 0.001). The effect size between BDA and OA was 0.743 (95% CI: 0.436, 0.634), which is statistically significant (p < 0.001). The effect size between OA and FP was 0.575 (95% CI: 0.456, 0.693), which is statistically significant (p < 0.001). Further, using FISHERINV() in Excel, the effect sizes for the aforementioned relationships were transformed to 0.489 (medium correlation), 0.631 (high correlation), and 0.519 (medium correlation). Moreover, the main effect analysis has been repeated to obtain the effect size of different OA groups for transforming correlations.

Mediating effects analyses

Based on the results of the main effects, this study has compiled the correlation matrix for mediation tests, as presented in Table 8. Further, to compare the mediating effects of different levels of OA, the original dataset was manually divided into two groups: organizational-level (ORG) and process-level (PRO). The ORG group comprised 19 datasets, while the PRO group encompassed 23 datasets. Additionally, the harmonic means for each SEM model were 11062, 4642, and 6240. The Mplus input parameters for each model were set to standard deviation = 0 and mean = 1. Since this study was a saturated model based on the correlation matrix, it shows a perfect fit in all MASEM analyses. Hence, there is no need to report its fit indices (Yu et al., 2016).

Table 8 Correlation matrix.
Full size table

As shown in Table 9, BDA had a strong and positive impact on OA (β = 0.631, p < 0.001), which enforced the findings in the main effect analysis that BDA was highly correlated with OA. OA had a positive impact on FP (β = 0.350, p < 0.001), and BDA had a positive impact on FP (β = 0.268, p < 0.001). For the mediating effect of OA, the indirect effect was 0.221 (95%CI: 0.209, 0.234) and statistically significant (p < 0.001), which meant OA can partially mediate the relationship between BDA and FP. Therefore, H4a was supported.

Table 9 Results of MASEM analysis.
Full size table

According to the results in Table 9, the mediating effects of different levels of OA remain significant and allow for further comparison as shown in Table 10. In the ORG model, the indirect effect was 0.191, accounting for only 37% of the total effect. In contrast, the indirect effect was 0.243 in the PRO model, constituting over half of the total effect at 52%. Moreover, in the original model, the indirect effect was 0.221, representing 45% of the total effect. Consequently, it was evident that PRO served as a more effective mediator in the relationship between BDA and FP than ORG. Therefore, H4b was supported.

Table 10 Comparison of mediating effects.
Full size table

Moderating effects analyses

Following the random-effect model setting, the WLS meta-regression results are shown in Table 11. While NC as a whole, did not moderate the BDA–OA relationship or the OA–FP relationship, it was worth noting that the sub-dimensions IDV (β = −0.005, p < 0.175) and IND (β = 0.006, p < 0.175) separately moderated the BDA–OA relationship at the marginal significance level. Similarly, the sub-dimensions UAI (β = 0.012, p < 0.175) and LTO (β = 0.004, p < 0.175) also respectively moderated the OA–FP relationship at the marginal significance level. The concept of marginal significance has gained acceptance among an increasing number of scholars in social science research in past decades, p-values range from 0.06 to 0.175 were reported by many studies (Pritschet et al., 2016). Given that this study represented a preliminary exploration of the NC moderating the BDA–OA and OA–FP relationships, achieving marginally significant moderation effects was considered acceptable. Therefore, H6a, H8b, H9b, and H10a were supported to a limited extent.

Table 11 WLS meta-regression analysis of national culture.
Full size table

The subgroup analysis results are shown in Table 12. The results indicated that different analytical techniques (CB-SEM vs. PLS-SEM) moderated the relationship with BDA-FP (Qbetween = 21.088, p < 0.001). Although the effect size of PLS-SEM (ES = 0.557) was slightly larger than CB-SEM (ES = 0.515), the difference was minimal. Additionally, different time dimensions (CS vs. LO) moderated the relationship with BDA-FP (Qbetween = 0.367, p < 0.01). The effect size of CS (ES = 0.548) was significantly greater than LO (ES = 0.367), but LO’s Qwithin value was not significant. Additionally, the summary of all hypotheses is presented in Table 13.

Table 12 Subgroup analysis of methodological moderators.
Full size table
Table 13 Summary of hypotheses in this study.
Full size table

Discussion

Direct effects between big data analytics and firm performance

The results of this meta-analysis on 34 studies indicated that BDA exhibited a positive impact on FP at a significant level, aligning with findings from previous meta-analyses (Oesterreich et al., 2022; Ansari and Ghasemaghaei, 2023). A survey found that in B2B firms, high-performing BDA users typically invest more in data and skills than basic BDA users, enabling them to achieve revenue growth of 10–20% (Dierks et al., 2023). A Previous meta-analysis has found that the effect size of BDA on overall FP surpasses the average level, albeit lower than that of artificial intelligence (Oduro et al., 2023). This observation underscores that BDA is an effective digital technology for enhancing FP in firms. However, this study found that the link between BDA and FP is not very strong, which further confirms the proposition of mediators between them.

The mediating effect of organizational agility

First, BDA exerted a robust positive influence on OA, verified by main effect analysis and SEM analysis. This outcome once again substantiated the pivotal role of BDA as a significant technology enabler of OA (Xie et al., 2022). Meanwhile, OA also has a positive impact on FP. This finding was extensively validated by the existing body of literature (Walter, 2021). This finding was also substantiated by practical evidence. According to McKinsey’s survey in 2021, 65% of respondents who achieved highly successful agile transformations believe that agility positively impacts financial performance. They further asserted that agility has increased consumer satisfaction and operational performance by 30% (Aghina et al., 2021).

This study further discerned that OA can partially mediate the relationship between BDA and FP. This outcome addressed the argument posed by Sambamurthy et al. (2003) about whether agility served as the key mechanism through which information technology influenced FP. Specifically, this study revealed that, compared to ORG, PRO can more effectively mediate the BDA-FP relationship. In this study, ORG primarily focused on the strategic and organizational levels. BDA plays a crucial role in supporting strategies. By analyzing historical data and evaluating the robustness of strategic options using available business data, BDA helps uncover potential blind spots in strategies (Kunc and O’Brien, 2019). Although big data predictive analytics is more accurate than subjective expert opinions in forecasting future events, large-scale, disruptive events can still impact prediction accuracy (Napoli et al., 2024). It is evident that BDA exhibits limitations when assisting firms in strategic decision-making. In today’s complex and highly uncertain business environment, sudden events can significantly impact corporate strategies. However, the ultimate realization of strategic goals remains uncertain for the future. Therefore, ORG explained only 37% of the mediating effect.

On the other hand, concerning strategic management, BDA finds extensive application in areas such as operations management and service innovation. According to a report by the IMA, among 260 surveyed Chinese firms, the primary benefit of BDA application is the promotion of standardized and efficient business processes. This significantly enhances agility (Ma and Qi, 2022). Furthermore, BDA’s impact on internal processes, financial performance, and marketing can be quantitatively measured through straightforward methods and visualized for management (Willetts and Atkins, 2023). A study suggests that in firms that benefit from successful BDA investments, the adoption of BDA to generate significant impact within the organization typically requires 18–24 months (Dierks et al., 2023). While the time from BDA implementation to effectiveness is not absolute, it may be influenced by various factors in practice. However, relative to firms using BDA for strategic management, the efficiency improvement effects of BDA on internal processes are visible in the short term. Consequently, compared to ORG, the enhancement of PRO through BDA application is evident, highlighting PRO’s substantial mediating role.

The moderating effect of national culture

This study revealed that IDV and IND can moderate the BDA–OA relationship, while UAI and LTO can moderate the OA–FP relationship, all reaching marginal significance levels. This suggested that with an increase in relevant research, these relationships may become significant in the future. Moreover, this result also partially supported the assumptions that agile as a cultural entity requires low levels of PDI, UAI, IDV, and MAS, as well as high levels of LTO and IND (Zakrzewska et al., 2022).

As shown in Fig. 3, the moderating effect of IDV on the BDA–OA relationship exhibited an inverted U-shaped pattern. This revealed that higher IDV levels do not necessarily lead to a stronger BDA–OA relationship. Instead, after reaching a certain inflection point (approximately an IDV score of 40), the BDA–OA relationship began to weaken. A previous study has found a negative correlation between high levels of IDV culture and the average level of digitalization in firms (Rubino et al., 2020), which partially supported the findings of this study. A study suggests that in collectivist countries, when there is less intense competition in achieving individual goals, team performance tends to be relatively better. And agile principles also implicitly imply priority group goals, which affirm the quality of collectivism (Zykov and Singh, 2020). Individualists are more inclined to work with people who have similar knowledge levels to their own. However, this preference made it challenging for employees to collaborate within the organization, as group members come from different departments, levels, and qualifications (Boroş et al., 2010). This indicated that in highly individualistic societies, although firms actively embrace BDA, during its usage—especially in collaboration and teamwork scenarios—employees with a strong individualistic orientation are less inclined to cooperate proactively with others, thereby hindering agility. This may explain why a medium level of IND could promote BDA–OA relationship.

Fig. 3: Moderating effect of IDV on BDA-OA relationship.
figure 3

The blue circle represents studies, the black solid line represents the meta-regression prediction line, and the gray area represents the 95% confidence interval.

Full size image

As shown in Fig. 4, high IND can positively moderate the BDA–OA relationship. A previous study suggested that greater IND positively influences a country’s creativity, as well as the willingness and ability of employees in that country’s firms to innovate and adopt new technologies (Jovanović et al., 2018). A past study highlighted that BDA usage supports organizational innovation processes (Capurro et al., 2022). By harnessing valuable information provided by BDA, firms can drive process innovation and product development. Ultimately, innovation enabled them to swiftly respond to market changes and maintain agility (Al-Omoush et al., 2024). These indicated that firms operating within high IND cultures exhibited a greater propensity to adopt BDA and leverage it to promote innovation, which will facilitate OA. This may explain why the BDA–OA relationship was stronger in high-indulgent cultures.

Fig. 4: Moderating effect of IND on BDA-OA relationship.
figure 4

The blue circle represents studies, the black solid line represents the meta-regression prediction line, and the gray area represents the 95% confidence interval.

Full size image

As shown in Fig. 5, high UAI can positively moderate the OA–FP relationship. UAI cultures are relevant to managerial decisions at all levels. In high UAI cultures, firms tend to employ more formalization and strict control in decision-making and its implementation. Additionally, compared to firms with low UAI, firms with high UAI are less inclined to make high-risk decisions. Instead, they prioritize conservative and stable approaches to maximize profits, resulting in reduced exposure to crises and financial risks (Bate, 2022). This indicated that in firms operating within low UAI cultures, there was a tendency to opt for less risky and more predictable decisions. While this approach may limit agility, it can effectively minimize performance losses, which proves beneficial in the long term. This may explain why the OA–FP relationship was stronger in high UAI cultures.

Fig. 5: Moderating effect of UAI on OA-FP relationship.
figure 5

The blue circle represents studies, the black solid line represents the meta-regression prediction line, and the gray area represents the 95% confidence interval.

Full size image

As shown in Fig. 6, high LTO can positively moderate the OA–FP relationship. Firms with LTO are perceived as more sustainable and better prepared for unexpected events due to their detailed long-term scenario planning (Ortiz‐de‐Mandojana and Bansal, 2016). For example, long-term-oriented firms were more inclined to innovate, even though it required a substantial investment of resources and did not yield immediate returns. However, in the long run, the output from innovation assisted them in enhancing their technological capabilities and competitive advantages, thereby achieving sustainable development (Fang et al., 2024). This indicated that firms within LTO cultures will use BDA to make decisions that align with their long-term interests. For instance, they might leverage BDA to accelerate the development of new products and further drive digital transformation within the firm, which enables them to remain agile, and bring them competitive advantages. This may explain why the OA–FP relationship was stronger in high LTO cultures.

Fig. 6: Moderating effect of LTO on OA-FP relationship.
figure 6

The blue circle represents studies, the black solid line represents the meta-regression prediction line, and the gray area represents the 95% confidence interval.

Full size image

The moderating effect of methodological moderators

This study also revealed that CS vs. LO and CB-SEM vs. PLS SEM can moderate the BDA–OA relationship. This study found that the effect size of CS (ES = 0.548) is significantly larger than that of LO (ES = 0.367), which aligns with previous meta-analyses (Oduro et al., 2023; Oduro and De Nisco, 2023). Notably, the LO group had only three samples, meeting only the minimum requirements for software to run analysis, resulting in a non-significant Qwithin value. Additionally, this study observed that the effect size of PLS-SEM (ES = 0.557) is slightly greater than CB-SEM (ES = 0.515), although the difference is not substantial. Among all samples, scholars tend to use PLS-SEM for handling small sample sizes, as it offers higher statistical power compared to CB-SEM (Rigdon et al., 2017).

Theoretical implications

Primarily, this study synthesized findings from a diverse range of databases over the past five years, including 34 articles related to the nexus of BDA–OA–FP. This study revealed that a positive relationship between BDA and FP. Although the impact of BDA on FP was relatively weak, this finding resolved the controversy surrounding the inconsistent relationship between BDA and FP. This further enriched the current body of knowledge on BDA. The study also furnished reliable evidence to substantiate subsequent studies focused on the BDA-FP relationship.

Secondly, this study contributed to the DCV theory. The study solidified the understanding that OA is a key mechanism through which BDA influences FP. By categorizing OA into two levels, namely ORG and PRO, this study distinguished the different mediating effects played by ORG and PRO in the BDA-FP relationship. Specifically, by comparing the percentage of indirect effects between ORG and PRO, it was observed that PRO serves as a more prominent mediator in the BDA-FP relationship. This finding, which was not previously established in prior studies, may be considered one of the most valuable contributions of this study.

Finally, this study also explored the contextual boundaries of DCV theory from a cultural perspective, an aspect that has often been overlooked in previous studies. Employing Hofstede’s cultural dimensions, the study empirically tested the moderating effect of NC in terms of different dimensions on the BDA–OA and OA–FP relationships at the macro level for the first time. Regrettably, we found that only IDV and IND moderate the BDA-OP relationship, while UAI and LTO moderate the OA–FP relationship at the marginal significance level. This may be the first discovery in BDA-related studies, so we cannot dismiss the theoretical contribution of marginal significance. It provides a potential direction for future studies to investigate how social and cultural factors influence the BDA–OA–FP relationship.

Practical implications

The results of this study also hold relevance for firms’ managers. It is essential to leverage BDA to achieve agility and gain competitive advantages in dynamic business environments. This study suggests that to quickly profit from BDA investments, firms should prioritize applying BDA to internal processes, thereby enhancing process agility and improving performance. Common applications include using BDA in manufacturing to identify real-time production process failures, predict machine maintenance cycles, and make real-time scheduling and cost-benefit decisions within manufacturing execution systems (Cui et al., 2020). To ensure successful BDA implementation, firms should upgrade their IT infrastructure to meet the requirements for collecting, storing, and analyzing large data throughout the entire process. Furthermore, firms must periodically update and even invest additional resources in BDA software to acquire more advanced BDA technologies and access new features. This contributes to maintaining agility. For small and medium-sized enterprises (SMEs), considering third-party cloud-based Big Data services can help reduce implementation costs. Furthermore, employee-specific training is essential to equip staff with BDA-related knowledge and skills.

Based on the findings of this study, the various dimensions of NC, such as medium IDV, high IND, high UAI, and LTO, can collectively facilitate the BDA–OA–FP relationships. The contextual view posits that organizational culture is largely influenced by the environment, emphasizing institutional forces such as national culture (Gerhart, 2009). Therefore, this study recommends that firms need to change their organizational culture to enhance the utilization of BDA for improving OA and ultimately enhancing FP. We propose the following recommendations:

To mitigate the negative impact of high IDV on the BDA–OA relationship, this study suggests that the management should embrace the cooperative and team-oriented aspects of collectivist culture. First of all, the firm needs to build a strong leadership that is purely goal-oriented, and the achievement of organizational goals is a manifestation of effective leadership (Campion and Wang, 2019). Given that the introduction of digital technologies alters workflow processes, enhancing collaboration and communication is essential for successful digital transformation (Trenerry et al., 2021). According to the author’s recommendations, firms should introduce new digital tools in the workplace and enhance team capabilities on virtual platforms. These measures can promote information sharing, improve communication efficiency, and enhance teamwork.

To promote the positive impact of high IND on the BDA–OA relationship, this study recommends that firms should implement flexible work arrangements to foster an indulgent organizational culture. The research suggests that in low levels of IND, employees may feel guilty when given more self-care, autonomy, and independent flexibility, which can compel them to work more and contribute further (Himawan et al., 2022). Some commonly used practices that firms can adopt include work-from-home arrangements and flexible working schedules.

To promote the impact of high UAI on OA and FP, this study recommends that firms should enhance their risk management capabilities. When using BDA, organizations often face various risks, such as security and privacy concerns (Raguseo, 2018). A study suggests that organizations need to implement risk management when using BDA to enhance strategic agility, thereby improving organizational performance (Khaw and Teoh, 2023). Therefore, this study suggests that firms need to establish a risk management framework tailored for digital transformation, aiming to prevent and mitigate potential risks associated with the use of BDA and other digital tools. By doing so, firms can maintain agility in dynamic business environments and achieve competitive advantages.

To promote the impact of LTO on OA and FP, this study recommends that firms should prioritize future-oriented strategies and investments (Saether et al., 2021). However, this poses a significant challenge for corporate management. Given the uncertainties of the future, leaders must be acutely aware of how their industry should evolve to address challenges related to new technologies, market forces, regulations, and more. Management needs a clear direction for industry development (Webb, 2019). Additionally, firms should allocate resources to research and development (R&D) within budget constraints to maintain competitiveness.

Limitations and future directions

This meta-analysis is based on Pearson correlations, requiring the collected complete correlations in each study. However, many studies did not provide correlations or other parameters that could be converted into correlations, further limiting the richness of the sample. Additionally, the samples collected in this study are all publicly published papers, excluding unpublished findings. Furthermore, most studies used self-reported questionnaires filled out by managers or employees, with a small portion using secondary data, which may introduce subjective-objective confusion. Lastly, as some studies span multiple industries and include firms of various sizes, this study did not differentiate based on industry or firm size.

Currently, cross-sectional studies remain predominant. Considering that realizing returns from BDA adoption by firms takes time, especially in terms of strategic management, short-term effects are not readily observable. Therefore, this study advocates for more longitudinal studies in the future to validate the impact of BDA on strategic agility and strategic performance. Additionally, to gain further insights into BDA’s strategic value for organizations, future studies could adopt a qualitative approach, such as conducting in-depth interviews with the management to understand their perceptions of BDA’s strategic value. Furthermore, this study only included BDA. In future meta-analyses, consideration could be given to incorporating other digital technologies, such as cloud computing and blockchain into the current research framework to further strengthen the DCV theory.

Additionally, due to the limited sample size in this study, the moderation effects were only marginally significant. As research volume increases, future meta-analyses should include samples from more countries to enhance diversity. Furthermore, using alternative methods like two-stages MASEM can further validate the hypotheses proposed in this study. Moreover, this study explored NC’s moderating effect from a macro perspective. A past study revealed that organizational culture positively moderates the relationship between different forms of BDA usage and agility (Hyun et al., 2020). Thus, future studies could investigate the impact of organizational culture on the relationship between BDA–OA–FP. Lastly, this study recommends that future studies pay more attention to the application of BDA in SMEs.