Opening the black box of big data sustainable value creation: the mediating role of supply chain management capabilities and circular economy practices

Big data analytics capabilities

BDAC address the ability of organizations to “provide insights using data management, infrastructure, and talent to transform business into a competitive force” (Mikalef et al., 2018, p. 557). Thus, BDAC are identified as a set of tangible and intangible capabilities (AlNuaimi et al., 2021). BDAC require IT capabilities to leverage the volumes of data from sources of structured data (such as enterprise information systems), as well as unstructured data from various field devices (e.g. sensors, RFID tags) (Arunachalam et al., 2018). Additionally, BDAC encompass the IT infrastructure that supports analytics, connecting internal and external elements, such as that seen with customer relationship management data (Kim et al., 2012). With this infrastructure, BDAC also require essential technical and managerial skills related to big data to create value (AlNuaimi et al., 2021). Moreover, BDAC influence strategies such as pricing and inventory levels by exploiting data-driven predictive and optimization models (Barton and Court, 2012).

The relationship between big data analytics capabilities and sustainable performance

A 'three-pillar' concept of sustainable performance (SP) dominates the literature as described by the triple bottom line performance dimensions: economic, environmental and social (Purvis et al., 2019). Economic performance refers to the maximization of the firm's economic value, as reflected in its profit or financial results (Andersson et al., 2022). Environmental performance recognizes positive environmental achievements from the firm's operations, while minimizing negative impacts in terms of intakes and outflows (Nutsugah et al., 2021). Finally, social performance refers to the firm's actions that benefit human capital and society in terms of, for example, community welfare or employee health (Nursimloo et al., 2020).

Previous literature highlights a relevant gap regarding how BDAC affect companies' SP (Raut et al., 2019). In a seminal work, Bharadwaj (2000) posited that firms could leverage organizational capabilities, such as those which fall within IT, to achieve superior firm performance. Some authors have extended this approach to BDAC, arguing that their deployment can lead to better performance (Akter et al., 2016). In the present study, we examine how BDAC impact SP.

Economically, data-driven intelligence from both internal and external sources provides vital insight for management decisions and helps the firm meet customer needs, increase sales and revenue, create new offerings and expand into new markets. These actions result in increases in productivity and financial performance (Akter et al., 2016).

Regarding environmental performance, AlNuaimi et al. (2021) highlight three fundamental forms of big data exploitation: (1) data processing to provide evidence of regulatory compliance; (2) big data analysis to address environmental challenges; and, (3) modeling and testing different production transformations and resource usage to improve environmental impact.

Finally, considering social performance, a data-skilled workforce can utilize advanced analytics tools with big data to manage social challenges, such as human safety, welfare and community development (Shafiq et al., 2020).

Therefore, we propose the following (see Figure 1):

The mediating role of supply chain management capabilities

SCMC identify the planning and management of activities involved in sourcing and procurement, including the coordination and collaboration with channel partners (Torasa and Mekhum, 2020). A construct of four dimensions conceptualizes these capabilities: (1) information exchange, which enables shared knowledge among SC partners; (2) coordination, allowing the firm to coordinate operations with partners; (3) integration, incorporating activities and technological tools; and, (4) responsiveness, acting on changes in the business environment (Wu et al., 2006).

Despite BDAC facilitation of SCMC being a contemporary topic, researchers have failed to reach a consensus on the specific mechanisms involved (Yu et al., 2021). Therefore, our theoretical development addresses how BDAC enable the four SCMC dimensions. First, BDAC facilitate information exchange between SC partners, using systems that collect, manage and share voluminous data from a wide variety of sources, such as ERP systems, orders and shipment logistics, and other data-driven technologies (Govindan and Hasanagic, 2018). Second, data-driven information from internal and external sources enables SC coordination, as observed in inter-organizational SC processes, such as reverse logistics and manufacturing flows (Koot et al., 2021). Third, the effect of BDAC on SCMC activity integration is illustrated by Yu et al. (2021). They show how hospitals gather, analyze and manage electronic health data through different healthcare information systems, increasing cross-functional, customer and supplier integration. Fourth, responsiveness is seen in Wang et al. (2016) in the deployment of data-based tools (e.g. statistical analysis, modeling and complex systems simulation), which uncover hidden data connections and turn them into critical insights related to product design and development, demand planning, sourcing and inventory. Additionally, these tools allow rapid reconfiguration of resources and capabilities in response to changing environments. From all the support mentioned above, sufficient evidence can be found regarding the impact of BDAC on SCMC.

The four dimensions of SCMC also contribute to SP (Lee et al., 2016), which this paper represents using the triple bottom line. As an example, Philips Healthcare Refurbished Systems leverages information and knowledge exchange with its customers, to return reconditioned medical devices to the market, thus reducing manufacturing costs (economic impact), decreasing material consumption, extending the useful life of the products (environmental impact) and improving community access to medical devices (social impact) (Jensen et al., 2019). SC coordination also significantly decreases SC vulnerabilities, which reduces costs and positively impacts economic performance (Munir et al., 2020). Coordination, likewise, results in less waste generation, lower energy consumption and the development of environmental-friendly processes and products (Iranmanesh et al., 2019). An example of how coordinated SC actions impact social performance can be seen in Apple's programs to improve working conditions with partners (Biswas et al., 2018). In other instances, SC integration reduces supervision costs and produces more economical solutions through collective problem-solving (economic impact); partners optimize environmental results by sharing objectives and strategies (environmental impact); and, stakeholder needs are satisfied as a result of information sharing (social impact) (Han and Huo, 2020). SC responsiveness strengthens the market connection, enabling rapid responses and sales increases with a positive economic impact (Swafford et al., 2008). Such responsiveness enables SC to detect environmental problems and adapt to legislative changes, positively affecting environmental performance (Ji et al., 2020). It also facilitates awareness of potential SC disruptions, thus preventing negative social consequences, such as lower wages and unemployment (Cui et al., 2022).

Therefore, and inferring from the support described, the deployment of BDAC in an organization enhances SCMC, and this advancement results in improved SP. Thus, the following hypothesis can be posited (see Figure 1):

The mediating role of circular economy practices

CE, which aims to transform linear production and consumption systems into circular models, is operationalized through specific actions and practices (Schroeder et al., 2019). These CEP are commonly conceptualized in “R” frameworks, such as 3R (reduce, reuse and recycle) (Cui et al., 2021), 4R (reduce, reuse, recycle and recover) (Gebhardt et al., 2022), or even, 9R (Potting et al., 2017).

Govindan and Hasanagic (2018) provide a more holistic approach, which we subscribe to. They propose six clusters for internal CE practices (our CEP dimensions) based on similarities and context: (1) governance initiatives, such as CE policies and performance indicators; (2) economic initiatives that decouple economic growth from environmental impact; (3) cleaner production to increase eco-efficiency; (4) product development pursuing durable design or reuse; (5) management support or CE endorsement from top management; and, (6) knowledge, referred to as CE education, training and creativity.

There is a general call for a deeper understanding of how BDAC facilitate CE models (Modgil et al., 2021). Specific mechanisms of this relationship can be seen in each of the CEP dimensions. (1) Governance initiatives: Big data sharing enables the development and monitoring of reliable CE indicators from all stakeholders, ensuring transparency across organizational boundaries (Kristoffersen et al., 2021b). (2) Economic initiatives: this dimension considers CEP profitable business opportunities. For illustration, the Brazilian company, eStock, develops a profitable reverse logistics business of damaged electronic products, leveraging cloud applications and big data analytics to sort the products and decide their final use (recycle, resale, or repair) (Modgil et al., 2021). (3) Cleaner production: Big data analytics enable real-time and predictive decision-making on clean practices, such as scheduled maintenance or optimizing material and energy consumption (Kristoffersen et al., 2020). (4) Product development: the closed-loop model, advocated by CE, requires sharing massively traceable and trustworthy data on product life cycles among all stakeholders (Chiappetta Jabbour et al., 2019). (5) Management support: data-driven insights on CEP outcomes provide top managers with tools to understand past and present trends and predict future ones (Awan et al., 2021). (6) Knowledge: BDAC are critical enablers for implementing CE training and educational actions. For example, Alpha (a European household appliance retailer) exploits big data from sensors on their machines to offer tailored advice to their clients to improve resource efficiency (Bressanelli et al., 2018). These examples provide sufficient theoretical and empirical evidence of the relationship between BDAC and CEP.

Meanwhile, adopting CEP can guide companies to a more sustainable business (Barros et al., 2021). Conceptually, cleaner production and circular product development lead to economic benefits, by reducing costs and opening new revenue sources, thus improving competitive advantage (Rosa et al., 2019). However, the environmental effects of CEP are most visible in practice. For example, recycling and remanufacturing activities involved in cleaner production and circular product development, result in more sustainable consumption of natural resources, while reducing pollutants and hazardous substances (Khan et al., 2022). Focusing on social effects, CEP can promote new forms of business cooperation, between small and medium-sized enterprises, to exploit by-products, fostering knowledge co-creation and social welfare (Howard et al., 2022).

Following this theoretical development, we posit that the deployment of BDAC to support CEP will increase SP (economic, environmental and social). Thus, we hypothesize the following (see Figure 1):

Sequential mediation of supply chain management capabilities and circular economy practices

Finally, SCMC can facilitate CEP in various forms. First, information exchange and coordination (reflected in training and collaborative work) allow firms and suppliers to produce shared knowledge, thus enhancing CEP (Stekelorum et al., 2021). Other authors note that SC coordination, enabled by partner collaboration, contributes to CE governance initiatives, such as establishing CE standards within the SC (Dubey et al., 2019). Likewise, SC coordination facilitates participation in cleaner production practices along the SC, such as using materials or products derived from recovered materials (Barros et al., 2021). Focusing on SC integration, Calzolari et al. (2021) demonstrated that this integration directly contributes to successful CEP implementation among partners. Seuring and Müller (2008) also point to integration and coordination as means to create partnerships for developing new products with circular life cycles. In addressing SC responsiveness, Bag and Rahman, 2023 prove that SC flexibility enables firms to reconfigure their resources and processes to embrace CEP, resulting in significantly cleaner fabrication and product development.

According to this support, BDAC relate to SP first through SCMC and then via CEP. Thus, we hypothesize the following (see Figure 1):

Sample

Our study population comprises a selection of industries, identified by the European Commission in its new Circular Economy Action Plan, as critical sectors in their potential for environmental impact and circularity (European Commission, 2020). Considering companies within these sectors with at least fifty employees, we identified a population of 3,572 companies. Then, we used a random stratified sampling procedure to generate a representative sample of companies in terms of dimensional parameters and industry.

Since the analysis is performed at the organizational level, the questionnaire is completed by a single individual from each company. Therefore, data collection involved only top managers. Fieldwork was conducted between September 2021 and October 2021 using a questionnaire, following a previous contact by telephone. As a result, a total of 210 surveys were completed. The demographic data of the final sample can be consulted in Table 1.

Measures

Our study uses validated scales for most constructs, and all scales, except controls, were measured on a seven-point Likert scale, ranging from 1 (strongly disagree) to 7 (strongly agree). This work measured BDAC with eight indicators by applying the scale developed by Raut et al. (2021). SCMC were measured with the Wu et al. (2006) scale, in a second-order structure, consisting of four dimensions: information exchange, coordination, integration and responsiveness. Finally, all elements used to measure CEP, whose implementation depends on firm policies that correspond to internal application levels, originate in the research developed by Govindan and Hasanagic (2018). Thus, these practices were grouped into six dimensions that form a higher-order construct (i.e. CEP): governance initiatives; economic initiatives; cleaner production; product development; management support; and, knowledge. Finally, the SP scale has been adopted from Çankaya and Sezen (2019). SP is a multidimensional construct composed of three critical dimensions: economic, environmental and social performance. The questionnaire can be accessed in Table S1 of the Supplementary tables. Additionally, the study controls the industry sector (categorical variable), age (number of years since its founding) and firm size (number of employees).

Given that our primary variables correspond to capabilities, practices and indexes, they can be described as forged concepts (Henseler, 2021), conceptual variables resulting from theoretical thought and composed of a mixture of elements (Henseler, 2017). Consequently, we model our constructs as composites integrated with more specific elements, such as dimensions or facts (Henseler, 2021).

Data analysis

We have selected and applied partial least squares structural equation modeling (PLS-SEM) (Hair et al., 2022; Ciavolino et al., 2022) based on multiple factors. First, the primary constructs of our research model are consistent with a composite measurement model (Henseler, 2021). Second, we use component scores to model higher-order constructs (i.e. SCMC, CEP and SP) and apply the disjoint two-stage approach (Sarstedt et al., 2019; Becker et al., 2023). Finally, one of the purposes of the study is to evaluate the research model in terms of prediction, a task fundamentally incompatible with factor-based methods (Rigdon, 2012).

Thus, PLS-SEM enables us to address the causal-predictive goals of the study (Hair et al., 2019). This study utilizes the Smart-PLS 3.3.9 software (Ringle et al., 2015).

Common method bias

The potential bias introduced by common method bias (CMB) can jeopardize findings, due to systematic errors. We attempted to prevent CMB during the research design phase by adopting the procedural remedies of MacKenzie and Podsakoff (2012). Subsequently, we applied two statistical procedures developed for PLS-SEM to detect various sources of CMB. First, we performed a full collinearity test using variance inflation factors (VIFs) (Kock and Lynn, 2012). A VIF greater than 3.3 indicates pathological collinearity, implying that CMB contaminates the model. However, our model appears free of CMB as indicated by the highest VIF of 2.584 (Table S2–Supplementary tables). Second, we employed the measured latent marker variable (MLMV) technique (Chin et al., 2014). The questionnaire included six items from the Chin et al. (2014) proposal of an observed latent marker variable (Table S1–Supplementary tables). The paths from the MLMV to the rest of the research model's constructs were nonsignificant. In addition, the hypothesized path coefficients were consistent with the initial estimates, with no essential variations (Table S3–Supplementary tables). This implies that CMB was not a significant concern, and hence we present the results of the original research model.

Measurement model

Henseler (2017) argues that, because our primary constructs represent artifacts (forged concepts), indicators of the composites are likely to be correlated. As a result, we used correlation weights to estimate these components in Mode A (Rigdon, 2016).

Following Hair et al. (2022), we observed that both indicators and dimensions (lower-order components) generally had loadings greater than 0.7 (Table 2). Then, we evaluated the internal consistency reliability of the constructs using composite reliability (CR). Since lower- and higher-order constructs showed CR values greater than 0.7, they had good reliability levels (Table 2). The average variance extracted (AVE) was then used to determine the convergent validity (Table 2), and all constructs achieved convergent validity with AVE values greater than 0.5. Similarly, discriminant validity was achieved for all primary constructs. This condition was achieved by adopting both the Fornell–Larcker criterion and the toughest Heterotrait–Monotrait ratio (HTMT) standard of 0.85 (Table 3).

In the end, we performed a confirmatory composite analysis applying bootstrap-based saturated model fit tests (SRMR, d_ULS and d_G), which assessed the external validity of primary constructs (Henseler, 2021). The three measures of the discrepancy between the empirical correlation matrix and the model-implied were all less than the corresponding values of HI95 or HI99 of their saturated model (Table 4), indicating that the discrepancy was not significant. As a result, we found support for the proposed composite model.

Structural model

First, the VIF values for the antecedent variables of the endogenous constructs ranged from 1.032 to 1.880, indicating that multicollinearity was not a concern in our structural model (Hair et al., 2022). Next, we examined the structural model for potential nonlinearities. Interaction terms were included to represent the quadratic effects of each antecedent variable on each dependent construct (Sarstedt et al., 2020). According to the bootstrapping results (10,000 samples), neither of the quadratic effects was significant. As a result, we concluded that the linear effects model was robust.

Figure 2 shows the path coefficients' signs, magnitude, significance and R² values for the dependent variables. We used bootstrapping (10,000 samples) to generate t-statistics and confidence intervals. While five of the six direct effects were significant (Figure 2) (Table S4–Supplementary tables), we did not find a significant relationship in the direct effect (c′) of BDAC on SP (H1). In terms of in-sample predictive power, the dependent variables achieved a satisfactory level as the lowest coefficient of determination (R²) achieved is 0.339 for SCMC (Figure 2). At the same time, SP reached an R-square of 0.585, which can be considered a moderate degree of explanatory power.

From a confirmatory perspective (Hair et al., 2019), we first evaluated the existence of a potential endogeneity problem by applying the Gaussian copula approach (Hult et al., 2018). We focused on the most complex regression of the hypothesized model, SP regressed to BDAC, SCMC and CEP. As a first step, the Gaussian copula technique demands that the antecedent variables, which may be affected by endogeneity issues, have a nonnormal distribution. Both the Shapiro–Wilk test and the Anderson–Darling test (Becker et al., 2022) showed that the distributions of BDAC, SCMC and CEP were not normal (p < 0.05; Table 5). Next, we performed a Gaussian copula analysis in PLS-SEM using the REndo package of the statistical software R (Gui et al., 2017). None of the copula terms was statistically significant at the 5% probability of error level. Table 5 shows the results for the most complex model with the three copulas. As a result, we conclude that the PLS-SEM results are robust and significantly unaffected by possible endogeneity issues, mainly due to omitted constructs that could correlate with one or more predictor constructs and SP.

Subsequently, we also evaluated the overall fit of the estimated model, through several measures of the overall goodness of fit available for PLS-SEM (Henseler, 2021), to obtain empirical evidence for the research model. First, the standardized root mean square residual (SRMR) evaluation offered a fair value of 0.052 (Table 4), below the usual cut-off of 0.08. Next, we conducted several bootstrap-based tests of the overall model fit (SRMR, d_ULS, d_G). Our results showed that all discrepancy values were less than the 95th percentile of their corresponding reference distribution (HI95). This suggests that the estimated model was not rejected at the 5% significance level (Henseler, 2021). Consequently, the postulated model cannot be rejected, as it is likely true.

Once we obtained evidence on the causality of the model, we tested the mediation hypotheses (H2–H4), following Nitzl et al. (2016). We examined the total, direct and indirect effects of BDAC on SP, the latter being controlled by sector, age and size variables. We executed a bootstrapping routine (10,000 samples) using percentile confidence intervals to test the indirect effects. As Table 6 shows, BDAC had a significant total effect on SP. However, after introducing both mediating variables, BDAC no longer had a substantial direct influence on SP (H1: c′) (Figure 2). Consequently, H1 was not supported.

Furthermore, all indirect impacts of BDAC on SP were significant. This indicated that H2 through H4 were supported. Therefore, we found evidence that SCMC positively mediated the relationship between BDAC and SP (H2: a₁b₁). The results also indicated that CEP mediated the relationship between BDAC and SP (H3: a₂b₂). Lastly, we found that BDAC were positively associated with greater SCMC and CEP, which were related to a higher level of SP (H4: a₁a₃b₂). Considering that the direct effect (c′) between BDAC and SP was not significant, we identified a full mediation model, meaning that the effect of BDAC on SP was transmitted entirely with the help of SCMC and CEP. This scenario was also supported by analyzing the variance accounted for (VAF) (Henseler, 2021) (Table 6, Figure 3) for the total indirect effect. As we could observe, 97.5% of the total effect of BDAC on SP was attributable to the sum of indirect effects.

Next, we performed a novel analysis that allowed us to know the additive contribution of the proposed mediators to the model's predictive accuracy. For this purpose, we used the predictive contribution of the mediator (PCM) metric proposed by Danks (2021). We generated three models, according to the mediating hypotheses, that allow us to assess the predictive contribution of SCMC, CEP and SCMC combined with CEP (as sequential mediators), respectively (Figure 4). The SEMinR package (Ray et al., 2022) was used to estimate the PCM metrics for each indicator representing the three dimensions of SP.

We analyzed the mediated effect of BDAC through SCMC on SP (Table 7). PCM estimates were above zero but below 0.05 for the three indicators. This result provided weak but confirmatory predictive evidence for the mediation effect of SCMC. However, in the context of the mediated path of BDAC through CEP on SP, the three PCM values for each dimension ranged from close to moderate (0.05), to strong (above 0.10) (Table 7). Therefore, CEP contributed substantially and provided strong confirmatory predictive evidence for the mediation effect. Ultimately, the sequential indirect effect through SCMC and CEP obtained moderate to very high PCM values (Table 7). This result represented a solid contribution to the predictive power of the two mediators.

In conclusion, the results of the PCM metric attained for the three mediated effects provided additional post hoc evidence to support the generalizability of the proposed mediators. Furthermore, our findings justified the added complexity of the inclusion of SCMC and CEP as mediators in our research model, as both constructs yielded an improved predictive accuracy of SP (Danks, 2021).

Evaluation of the predictive power of the model

We performed the PLS_predict procedure (Shmueli et al., 2019) to assess the out-of-sample predictive power of the model. We began by selecting SP as the target construct. Next, we performed k-fold cross-validation, setting k = 7 subgroups to meet the requirement of reaching a minimum sample of 30 cases in each holdout sample. Then we selected to repeat the algorithm ten times. Afterward, we interpreted the results (Table 8) by completing the following stages (Shmueli et al., 2019):

1. The indicators representing each dimension of SP showed Q²_predict > 0.

2. Since prediction errors are symmetrically distributed, we have used the root mean squared error (RMSE) to assess the degree of prediction error (Table 8). Subsequently, the RMSE statistics from the PLS-SEM analysis were compared with the naive values provided by a linear regression model (LM). Indicators representing economic, social and environmental performance produced smaller prediction errors in the PLS-SEM analysis than LM, showing a high predictive power. In conclusion, these results indicate that the research model could provide generalizable findings for other datasets and potentially equivalent contexts.

Theoretical implications

Considering the results obtained, the first contribution of this research is that BDAC do not directly influence SP. This finding deviates from previous studies which establish that BDAC positively influence SP (Kamble et al., 2020). Still, little is known about the process that leverages big data analytics investments toward firm performance, either directly or indirectly – for example, how a company using IT infrastructure to develop higher-order capabilities of SC processes generates significant and sustainable gains (Rai et al., 2006). Our results demonstrate that SCMC and CEP play a central role in mediating this relationship. Adopting BDAC to achieve SP is complex and requires complementary resources to help organizations realize their full potential. In the hierarchy of capabilities and from the perspective of IT-enabled organizational capabilities (Benitez et al., 2018), lower-order capabilities require higher-order capabilities to affect business outcomes. This research contributes to the literature on IT business value by demonstrating how one IT capability develops other capabilities that influence SP.

The second contribution of this research highlights the mediating role of SCMC. The IT infrastructure for big data analytics can assist in SC functions, such as procurement, warehousing, manufacturing, demand management and logistics (Govindan et al., 2018). To achieve SC innovation and sustainable SC performance, BDAC are considered essential (Akter et al., 2016). Therefore, BDAC and SCMC are fundamental capabilities which provide vital theoretical insights for big data and SC management. SCMC include information exchange, integration, coordination and responsiveness which help firms at distinct stages of SC operations and are key performance indicators for businesses (Shen et al., 2019). The results of previous studies on the impact of SCMC on financial performance show that the result is inconclusive (Yu et al., 2022). However, our results have revealed a significant impact of SCMC on SP. Given this, we attempt to clarify the elements which contribute to determining SCMC and how these capabilities are related to influence SP (Gani et al., 2022).

A third contribution points to the mediating role of CEP in the impact of BDAC on SP. Therefore, adopting digital technologies is consistent with the concepts of CE (Khan et al., 2022). Our contribution underscores the importance of CEP in performance sustainability. The successful implementation of a CE model facilitates the growth of all three aspects of SP, and CEP help to achieve sustainability goals (Schroeder et al., 2019).

Our fourth theoretical insight reveals that SCMC are associated with CEP, and both constructs sequentially mediate the relationship between BDAC and SP. The demonstrated relationship between SCMC and CEP confirms the argument that embracing complex environmental or social initiatives, such as CE models, will inevitably fail without the commitment of SC partners (Silvestre et al., 2018). CE is an innovation in closed-loop SC, since it affects design, implementation and managing activities that combine the upstream and downstream stages of the SC, increasing value generation throughout the entire life cycle of a product (Chowdhury et al., 2022). Finally, the entire mediation sequence emphasizes the importance of digitally transforming key organizational capabilities to achieve sustainable development, as evidenced by BDAC enhancing SCMC, accelerating the adoption of CEP, and thus increasing SP.

Implications for practitioners and policymakers

The positive results achieved in our predictive analyses imply the model's generalizability to potentially similar contexts. Therefore, if companies follow the recommendations based on our model, they could achieve equivalent results, attaining benefits in four primary areas.

First, an organization with appropriate BDAC can develop SCMC to exchange and coordinate information with SC partners, leading to improved integration and responsiveness. The current competitive environment encourages companies to adopt big data analytics techniques. The framework provided can help implement these capabilities to enhance SP.

The knowledge provided by big data analytics and SC capabilities is based on tacit resources. These resources are not visible but can be achieved through learning and practice. From them, it is possible to develop CEP, which can establish a competitive advantage for the company. Additionally, top managers and executives can see how CEP can help them improve their company's performance and what role BDAC and SCMC play as CEP enablers.

Production and SC managers should know that implementing BDAC does not directly improve sustainability, especially at the firm level. However, from a managerial perspective, BDAC have emerged as a critical component that can help companies implement CE models, optimize operations and generate long-term solutions.

Therefore, managers must pay close attention to the various CEP to reap benefits from a sustainability performance perspective. Government policymakers can also justify increasing incentives for CE projects that lead to sustainable community benefits.

Limitations and future research

This study has some limitations. First, the authors evaluated the model in a single country under unique conditions; therefore, the findings should be replicated in different situations. Second, the study relied on measurements based on the participants' perceptions. As such, it must account for the risk that the respondents' impressions do not accurately reflect reality. Finally, the research was carried out cross-sectionally rather than longitudinally.

Future studies could focus on including additional variables into the model proposed in our research, thus increasing the understanding of the mechanisms which drive the creation of sustainable value from BDAC. First, we could focus on establishing interdependence relationships between the different SCMC and the different CEP, determining a sequence of implementation, thus identifying those practices that constitute the base on which the other activities can be successfully implemented. Second, we could analyze certain contingent elements that could affect the contribution of the variables in the model to SP (e.g. environmental uncertainty, industrial sector), facilitating the generation of conditional mediation analyses (Cheah et al., 2021). Third, some variables may affect firm behavior and could increase our understanding of the phenomenon. Examples include the company's strategic orientation driven by big data, environmental self-awareness and stakeholder pressure. Finally, to improve the robustness of the findings, future research might use longitudinal data to assess intertemporal impacts.