This study explores how selected digital technologies and organisational characteristics were associated with environmental practices among European SMEs in the pre-Green Deal period. Using cross-sectional data from Flash Eurobarometer 486 (2020), it estimates a multivariate probit model to jointly analyse five ecological outcomes — eco-innovation, recycling, resource reduction, energy saving, and sustainable product development. The model includes six binary indicators of digital adoption (AI, big data, cloud computing, high-speed internet, robotics, smart devices), as well as firm-level controls for ownership, financial capacity, skill shortages, location, size, and sector.
The results reveal a patterned but non-uniform alignment between digital infrastructure and environmental behaviour. Cloud computing and robotics exhibit the most consistent associations across outcomes, while AI and big data show more selective effects. Among organisational factors, family ownership and financial capacity appear positively associated with green engagement, whereas skill shortages are linked to reduced adoption of energy- and resource-saving practices.
By documenting these differentiated associations, the study provides a structured empirical snapshot of how digital and organisational conditions related to environmental practices among SMEs at a key institutional juncture—before policy acceleration rendered sustainability a more central strategic imperative.
The transition toward a greener and more digital economy has become a defining challenge for the European Union. In response to climate change, resource scarcity, technological disruption, and long-term strategic goals, the European Union has placed the twin transition as foundational principles of the future European economy (European Commission, 2020). This dual transformation requires the active involvement of all sectors of society, including businesses, governments, and civil society. Among economic actors, small and medium-sized enterprises are considered a central pillar of the European productive system and are expected to play a crucial role in translating policy objectives into concrete operational practices. These expectations can be interpreted as institutional pressures that shape not only the strategic priorities of firms, but also the symbolic alignment with broader societal goals. Yet, the capacity of SMEs to pursue both digital and environmental strategies remains highly uneven, depending on structural conditions, organisational capabilities, and available incentives. Although some firms have started to integrate digital and green strategies, the degree of integration varies significantly across the SME landscape. Despite the strategic prioritisation of the twin transition at EU level, many SMEs face structural and cognitive barriers that hinder the effective implementation of both green and digital practices (Rupeika-Apoga & Petrovska, 2022). The literature has examined this relationship from multiple perspectives, including eco-innovation and corporate social responsibility (CSR) (del Río et al., 2016; Horbach, 2024; Triguero et al., 2013). Digital technologies are increasingly viewed not only as tools for improving efficiency, but also as catalysts for systemic change. These technologies enable real-time monitoring and tracking, support life-cycle assessment and predictive simulations, and facilitate the virtualisation of activities - thereby reducing emissions linked to production, logistics, and mobility (Joint Research Centre European Commission, 2023).
Following Di Maggio and Powell’s (DiMaggio & Powell, 1983) theory of institutional isomorphism, the diffusion of digital and green practices among firms may not solely reflect instrumental adaptation to market or environmental challenges, but also a process of conformity to institutionalised expectations. As the EU consolidates a normative agenda linking innovation and sustainability, organisations may adopt digital and environmental routines to gain legitimacy within their fields, regardless of short-term efficiency gains. From this perspective, the alignment between digitalisation and sustainability is not merely a matter of capabilities or incentives, but also a response to the symbolic pressures embedded in the European policy discourse. This lens is particularly relevant to our study, which tests whether SMEs reporting digital adoption are also more likely to engage in ecological practices. If isomorphic pressures are at play, one would expect a pattern of relational alignment between digitalisation and environmental engagement - even in the absence of formal “twin transition” programmes. Rather than indicating a generalised twin transition across the SME landscape, the results reveal differentiated alignments between specific digital technologies and environmental outcomes, suggesting that institutional signals alone - without policy shifts and financial supports - were insufficient to generate systematic twin transition behaviours among SMEs. Indeed, the relationship between digitalisation and sustainability is not deterministic. Recent studies have pointed out that, although SMEs are central to the EU’s twin transition strategy, the actual implementation of integrated digital and green strategies remains limited (Burinskienė & Nalivaikė, 2024). Many firms adopt either digital or sustainable practices but struggle to pursue both simultaneously. Organisational factors such as firm size, internal competences, and strategic orientation play a key role in shaping whether and how these two dimensions are pursued jointly, especially within SMEs (Horbach, 2024).
This study investigates whether the adoption of specific digital technologies is associated with greater engagement in environmental practices among European SMEs. While we do not aim to make causal claims, our objective is to assess the strength and robustness of observed associations. Using data from the Flash Eurobarometer 486 (European Commission, Brussels, 2020) - which provides cross-sectional insights into firm behaviour across EU member states along with several non-EU countries - it examines how patterns of digital uptake relate to eco-innovation, energy saving, recycling, resource efficiency and sustainable product development. Since the survey predates both the adoption of the European Green Deal and major post-pandemic investments, it offers a unique opportunity to explore how firms were positioning themselves in the ecological dimension of the twin transition before policy acceleration. This setting enables a descriptive assessment of organisational conditions and digital infrastructures associated with early-stage sustainability behaviours.
Building on Horbach’s operational framework (Horbach, 2024), which identifies key firm-level indicators of digitalisation and sustainability. This approach is informed by a broader body of research on the uneven diffusion of sustainability innovations (del Río et al., 2016; Triguero et al., 2013), and the temporal lags that often exist between policy design and behavioural change at the firm level (OECD, 2019). The purpose of this analysis is to contribute to a deeper understanding of how the twin transition was implemented in practice - and to identify the organisational conditions that might have supported its implementation within the European SME landscape.
To guide the selection of variables, this study follows the literature-based framework adopted by the aforementioned study by Horbach (Horbach, 2024), who identifies relevant indicators of digitalisation and environmental engagement on the same dataset, by drawing on previous research on eco-innovation and digital transformation in SMEs. Digitalisation is captured through the presence of specific technological applications following the DESI index (European Commission, n.d., 2022) provided by the European Commission - such as AI, cloud computing, big data analytics, and automation - while environmental engagement is assessed through firm-level practices including energy efficiency, resource savings, recycling, and the development of sustainable products. The “human capital” dimension of the DESI index is not included among the digital predictors, as the analysis deliberately focuses on technological infrastructures rather than skills availability. This decision aims to isolate the technological core of digitalisation. Similarly, in order to ensure conceptual clarity and model consistency, the scope of the dependent variables is also intentionally confined to environmental practices - rather than adopting a broader CSR perspective. By restricting these domains, the study provides to fill the gap of previous analysis, enlightening the role of digital activities for environmental sustainability and digital technology only. Concerning the dependent variable, this modelling choice is particularly relevant in the context of a multivariate probit approach adopted in this study, which assumes underlying correlations among them. Focusing only on ecological indicators - such as energy saving, recycling, and resource efficiency - allows the model to isolate environmental behaviour as a specific domain of action, analytically distinct from broader social or reputational concerns.
To account for potential confounding factors and ensure that the observed associations between digitalisation and environmental engagement are not driven by structural heterogeneity among firms, the model includes a set of control variables related to firm organisation and local context. These include ownership structure, financial capacity, location in industrial or urban areas, perceived skill shortages, sector (NACE classification), firm size, and countries. Their inclusion is theoretically grounded in the literature on eco-innovation determinants, which suggests that both internal characteristics (such as firm size or ownership) and external conditions (such as regional infrastructure or labour market composition) can influence a firm’s propensity to adopt digital or sustainable practices (Horbach, 2024; Rennings, 2000). By including these controls, the analysis aims to better isolate the specific contribution of digital technologies, net of these broader structural influences.
The findings are consistent with Horbach’s insights, showing that the adoption of digital technologies is not uniformly associated with green practices. While cloud computing, big data, and robotic automation exhibit consistent and positive associations with various environmental outcomes, other technologies - such as artificial intelligence and smart devices - appear less systematically integrated with ecological strategies. Additionally, firms reporting skill shortages are less likely to engage in energy-saving and resource-reduction activities, suggesting that human capital constraints may hinder the alignment of digital and green objectives within SMEs. These results point to an uneven and selective pattern of digital–green integration across European firms, offering a grounded understanding of how environmental goals were approached in the absence of strong policy incentives. The study thus contributes to trace how digital infrastructures and organisational constraints intersected with emerging ecological practices before policy shifts, offering a contribution to ongoing debates on the uneven social embedding of sustainability transitions.
This study employs cross-sectional data from the Flash Eurobarometer 486 (European Commission, Brussels, 2020), a large-scale survey conducted in 2020 among small and medium-sized enterprises (SMEs) across all EU Member States, along with several non-EU countries such as Bosnia and Herzegovina, Brazil, Canada and United States of America. The dataset includes 16,365 firms and enables a joint analysis of digitalisation and environmental engagement at firm level. The timing of the survey, immediately preceding the institutional launch of the Green Deal and major post-pandemic recovery plans, offers a rare opportunity to observe a valuable snapshot of SMEs in a pre-policy acceleration phase, largely unfiltered by recent institutional incentives.
Data
The five dependent variables used in this study (Table 3.1) are binary indicators derived from question Q23 of the Eurobarometer survey; each coded as 1 if the firm reports engaging in a given environmental activity. This operationalisation builds on Horbach’s framework (Horbach, 2024), which draws on OECD guidelines and earlier classifications of eco-innovation strategies. Specifically, Horbach distinguishes between process-related eco-innovations (e.g., energy and resource efficiency) and product-related innovations (e.g., sustainable products, recycling), echoing early work by Rennings (Rennings, 2000), who conceptualised eco-innovation as a multidimensional phenomenon affecting both production processes and market outputs. This distinction is particularly useful for understanding how ecological objectives are embedded at the firm level, as it captures both internal transformations and externally visible innovation. While Horbach includes both environmental and CSR-related items, this study restricts its scope to ecological behaviours only in order to provide a more analytically coherent contribution to the literature on green innovation. The rationale is both theoretical and methodological: green practices represent a more clearly defined policy domain, and their joint modelling avoids the conceptual heterogeneity that would result from incorporating CSR measures with different motivational and organisational logics. Including heterogeneous outcomes - such as CSR or social responsibility indicators - would conflate distinct behavioural domains and potentially distort the estimated associations, requiring robustness checks with other models that address heterogeneity. Indeed, in the aforementioned paper that addresses similar hypotheses using the same dataset, Horbach also includes a negative binomial model as a robustness check – a specification that may be more appropriate when broader constructs like CSR are involved but less pressing in this case due to the restricted analytical domain. This modelling choice enhances internal validity by minimizing conceptual noise and aligning dependent variables around a consistent environmental logic.
| Eco-sustainable activities | In % of all questioned firms |
|---|---|
| Eco-innovation | 23.8 |
| Recycling or reuse of materials | 63.5 |
| Reduction of material use and resources | 53.6 |
| Energy saving measures or renewables | 54.2 |
| Development of sustainable products/services | 35.2 |
The main explanatory variables in the model (Table 3.2) capture firm-level digital maturity across six technological domains: artificial intelligence (AI), big data analytics, cloud computing, high-speed connectivity, robotics, and smart devices or sensors. Each variable is represented by a binary indicator derived from item Q15 of the Flash Eurobarometer 486 (2020), taking the value 1 if the firm reports adoption of the respective technology. This disaggregated approach allows for a more granular test of the core hypothesis: that some digital adoptions may be systematically associated with environmental engagement. The use of separate indicators, rather than a composite index, aligns with the structure of the original dataset and ensures compatibility with the multivariate probit specification, which estimates the marginal contribution of each technological domain across five environmental outcomes. To translate this conceptualisation into measurable firm-level indicators, the selection of digital variables is guided by the structure of the Digital Economy and Society Index (DESI) developed by the European Commission (European Commission, n.d., 2022). DESI identifies four pillars of digital development - connectivity, human capital, digital public services, and the integration of digital technology - of which the latter two are particularly relevant to enterprise behaviour. The six technologies included in this study reflect core components of DESI’s “integration of digital technology” dimension and serve as empirical proxies for the firm’s digital capacity to collect, process, and act upon information. Their inclusion in the model enables the identification of differentiated effects across technological domains, providing insight into which aspects of digitalisation are more tightly linked to specific environmental practices. As previously reported in the introduction session, the ‘human capital’ component of the DESI index is excluded from the main digital predictors, as it could represent a structural constraint rather than a technological tool, and is modelled separately as a control variable (skill shortages) to narrow the domain to digital technology only.
| Digitalisation activities | In % of all questioned firms |
|---|---|
| AI, machine learning, pattern recognition | 8.7 |
| Big data analytics | 15.8 |
| Cloud computing | 52.4 |
| High-speed infrastructure | 36.3 |
| Use of robots, automation of processes | 9.8 |
| Smart devices, intelligent sensors, thermostats | 29.7 |
Control variables reported in Table 3.3 reflect structural and organisational firm characteristics shown in the literature to influence environmental strategies: family ownership (Beise & Rennings, 2005), financial capability (Cainelli et al., 2020), high-growth orientation, location in industrial or urban areas, skill shortages – as perceived barriers to green innovation, firm size – measured as the natural logarithm of the number of employees to account for skewed distribution and diminishing returns in scale effects, and sector of activity – collapsed into macro-categories using NACE codes. The inclusion of financial and organisational barriers is particularly relevant given the evidence from Rupeika-Apoga and Petrovska on the constraints SMEs face in simultaneously implementing digital and green innovations (Rupeika-Apoga & Petrovska, 2022); indeed, the variable on skill shortages is particularly salient, as the aforementioned study suggests that human capital constraints are a critical bottleneck in SMEs’ ability to engage in twin transition strategies. This variable reflects perceived – not objectively measured – constraints, potentially capturing managerial awareness or strategic prioritisation, rather than actual workforce composition. However, as reported in the discussion session, this variable could function both as a control and a potential moderator, highlighting organizational capacity as a pathway of influence.
| Characteristic of the firm | In % of all questioned firms |
|---|---|
| Family owned | 22.7 |
| High financial capability | 16.9 |
| Located in an industrial area | 14.5 |
| Located in a big city | 47.6 |
| Skills barrier | 21.7 |
Multivariate Probit model
The analysis relies on a multivariate probit specification, jointly estimating the five binary outcomes to capture the potential interdependence of green behaviours. This modelling choice is motivated by the theoretical expectation that firms may adopt multiple environmental practices in correlated ways — due either to common organisational drivers or complementary resource configurations. The use of a joint estimation framework allows for the modelling of latent correlations in the error terms across equations, which would be lost in separate binomial regressions. The model is estimated using the mvProbit function in R, and average marginal effects (AMEs) are computed to ease interpretation. AMEs offer an intuitive summary of the marginal impact of each explanatory variable on the probability of each outcome, averaged over the distribution of observed covariates. This is particularly useful in models where coefficients themselves are not directly interpretable in probability terms. Indeed, in this case, marginal effects represent the change in the predicted probability of each outcome with respect to a one-unit change in the explanatory variables, evaluated at the mean of the covariates.
Although the specification does not control for unobserved firm-level characteristics — such as internal culture or strategic orientation — the focus still remains descriptive. Both country (isocntry) and sector (nace_a) variables are included in the model to adjust for contextual and sectoral variation. However, as neither variable was treated as a factor, they do not operate as full fixed effects. Their inclusion provides only a first-order adjustment for differences across countries and sectors, rather than absorbing all unobserved heterogeneity. The cross-sectional nature of the data and the absence of suitable instruments prevent causal inference, but the model still offers a valid approximation of associations under observed conditions. Moreover, standard errors are not cluster-adjusted, which may lead to a mild underestimation of uncertainty if residual correlations exist within country or sectoral groups; firms in the same country or sector may face similar unobserved influences such as national policy or industry norms, causing their error terms to be correlated.
The results, summarised in Table 4.1, suggest a differentiated yet coherent pattern of association between digital adoption and environmental practices among European SMEs.
Cloud computing emerges as the most consistent predictor across all five environmental outcomes. It shows statistically significant and positive marginal effects on eco-innovation (AME = 0.024, p < 0.05), recycling (0.078, p < 0.001), resource reduction (0.084, p < 0.001), energy saving (0.055, p < 0.001), and sustainable product development (0.067, p < 0.001). These results reinforce the view that cloud infrastructures not only support dematerialisation and operational flexibility but also facilitate scalable data storage and real-time processing capabilities that can enhance environmental monitoring and optimisation processes - as also noted in policy-oriented assessments such as the Joint Research Report of 2022 (Joint Research Centre European Commission, 2023) on digital transition and its environmental potential.
Robotic automation also displays consistent and statistically significant associations across all environmental outcomes, with marginal effects ranging from 0.042 to 0.074 (all p < 0.05). These results align with Horbach’s framework (Horbach, 2024), which highlights automation as a key enabler of resource-efficient and cleaner production processes. The results are broadly in line also with recent evidence by Almuaythir et al. (Almuaythir et al., 2024), who highlight the potential of robotics to enhance sustainable development through improved productivity, innovation, and energy efficiency in several sectors.
Big data analytics similarly demonstrates significant predictive power, particularly for eco-innovation (0.066, p < 0.001), sustainable products (0.065, p < 0.001), and resource reduction (0.038, p < 0.05). The ability of big data systems to support predictive modelling, traceability, and lifecycle analysis likely contributes to these outcomes, providing firms with the analytical capacities necessary to improve efficiency and reduce waste, as discussed in empirical study such as Shajari and David (Shajari & David, 2025) and Angelopoulos (Angelopoulos & Kontakou, 2020) which highlight the role of data infrastructures in enabling proactive environmental management and optimization of resources - especially in the industrial sector. However, its association with energy saving is not statistically significant, which may reflect a limited integration of big data analytics into real-time energy management systems among SMEs. While previous studies have shown that big data can significantly contribute to energy efficiency through demand forecasting and process optimisation (Angelopoulos & Kontakou, 2020), such applications may require complementary investments in IoT or automation that are less accessible to smaller firms. Moreover, given that the data analysed here refer to 2019, the apparent discrepancy might also be attributed to temporal gaps in the diffusion of such technologies.
AI and machine learning display a more selective pattern of significance. While not associated with recycling (AME = 0.01, p = 0.648), AI adoption shows robust positive effects on sustainable product development (0.075, p < 0.001), resource reduction (0.058, p < 0.01), and energy saving (0.053, p < 0.05). This pattern is in line with evidence from Burinskiene and Nalivaike (Burinskienė & Nalivaikė, 2024) and Shajari and David (Shajari & David, 2025), who highlight how AI technologies are typically deployed in contexts requiring high computational precision—such as product design, predictive maintenance, and advanced optimisation—rather than in routine or labour-intensive processes. Nevertheless, the lack of association with recycling in our results contrasts with findings from Fang (Fang et al., 2023), who demonstrates that AI-driven classification and sorting systems can significantly enhance recycling efficiency. This discrepancy may reflect sectoral heterogeneity, gaps in adoption across European SMEs, or temporal factors, considering that the data precede the post-2020 acceleration in AI integration within waste management systems.
Some technologies such as high-speed broadband and smart devices do not exhibit statistically significant associations with environmental practices in this study.Understanding whether, when, and how such infrastructural components contribute to ecological strategies remains an open question for empirical investigation. Future studies may benefit from disaggregating these technologies by use-case, sector, or intensity of integration, in order to clarify if there are conditions under which their environmental relevance becomes more salient.
Turning to the control variables, family ownership shows significant and positive associations across all green outcomes, most strongly with recycling (0.053, p < 0.001) and resource reduction (0.05, p < 0.001). This supports prior findings that family-owned firms are more inclined toward long-term strategic behaviour and reputation-sensitive investments, including environmental innovation (Miroshnychenko et al., 2025). High financial capability is also correlated with sustainable product engagement (0.039, p < 0.05), though its role in other areas appears more limited. This is supported by studies such as Angelopoulos (2020) (Angelopoulos & Kontakou, 2020), who shows that financial endowment facilitates complex green investments, especially in the domain of product development where upfront investments tend to be higher. Importantly, perceived skill shortages — used as a proxy for human capital constraints — emerge as a negative predictor for recycling and resource reduction. This aligns with Rupeika-Apoga and Bassi et. Al, studies (Bassi & Guidolin, 2021; Rupeika-Apoga & Petrovska, 2022), both of whom highlight the centrality of skilled labour availability in enabling SMEs to pursue complex innovation goals. Indeed, the negative marginal effects observed suggest that workforce gaps may directly inhibit the capacity to engage in ecological practices, particularly those requiring technical or operational changes.
Overall, the findings point to heterogeneous but interpretable relationships between digital capabilities, organisational characteristics, and environmental practices. Rather than predicting behavioural change per se, the findings offer a structured interpretation of how digital infrastructures and organisational conditions — such as governance, finance, and human capital — were correlated to ecological engagement among European SMEs in the pre-Green Deal landscape. As mentioned in the previouses sections, this snapshot reflects a baseline configuration prior to the policy acceleration that followed in the early 2020s.
| Variable | Eco-Innovation | Recycling | Resource Reduction | Energy Saving | Sustainable Products |
|---|---|---|---|---|---|
| AI, machine learning, pattern recognition | 0.04*** (0.009) | 0.01 (0.648) | 0.058*** (0.01) | 0.053** (0.014) | 0.075*** (0) |
| Big data analytics | 0.066*** (0) | 0.035** (0.028) | 0.038** (0.024) | 0.02 (0.208) | 0.065*** (0) |
| cloud Computing | 0.024** (0.014) | 0.078*** (0) | 0.084*** (0) | 0.055*** (0) | 0.067*** (0) |
| family owned | 0.048*** (0) | 0.053*** (0) | 0.05*** (0) | 0.048*** (0) | 0.042*** (0.001) |
| high finance capability | 0.031** (0.022) | 0.003 (0.878) | 0.018 (0.311) | 0.024 (0.177) | 0.039** (0.014) |
| high speed infrastructure | 0.032*** (0.001) | 0.064*** (0) | 0.091*** (0) | 0.075*** (0) | 0.063*** (0) |
| industrial area | 0.025** (0.042) | 0.04** (0.013) | 0.001 (0.954) | 0.008 (0.612) | 0.033** (0.026) |
| Size | 0.01*** (0.007) | 0.009** (0.036) | 0.013*** (0.005) | 0.028*** (0) | 0.002 (0.585) |
| Use of robots, automation of processes | 0.042*** (0.003) | 0.044** (0.025) | 0.059*** (0.004) | 0.073*** (0) | 0.074*** (0) |
| skills barrier | -0.089*** (0) | -0.015 (0.249) | -0.07*** (0) | -0.082*** (0) | -0.103*** (0) |
| Smart devices, intelligent sensors, thermostats | 0.142*** (0) | 0.105*** (0) | 0.139*** (0) | 0.143*** (0) | 0.107*** (0) |
| urban area | -0.02** (0.028) | -0.023** (0.04) | -0.021* (0.07) | -0.052*** (0) | -0.003 (0.797) |
Robustness
To assess the robustness of the findings, the analysis relies on the estimation of average marginal effects (AMEs), computed at the mean of the explanatory variables. This approach allows for a meaningful interpretation of the non-linear effects associated with each covariate on the five binary outcomes, offering a descriptive understanding of how different forms of digital adoption relate to environmental practices. While the specification does not explicitly account for endogeneity or unobserved firm-level heterogeneity — due to the cross-sectional nature of the data and the lack of appropriate instruments — the use of AMEs provides a structured approximation of how digital and organisational factors are aligned with ecological engagement under observed conditions. Standard errors are not cluster-adjusted at the country level, and this limitation should be considered when interpreting statistical significance. No formal tests for multicollinearity or specification redundancy were conducted, and the model does not evaluate joint model stability through variable exclusion or alternative estimation techniques. Despite these constraints, the consistency of patterns across the five outcomes — combined with the theoretical justification for treating them as a coherent analytical domain — supports the descriptive validity of the model. The focus on environmental practices, as opposed to broader sustainability indicators that include social or CSR dimensions, strengthens the comparability of effects and reduces the risk of conflating distinct behavioural logics. As such, the results should be interpreted as robust descriptive associations within the ecological dimension of the twin transition, rather than as causal effects.
Interpretation
An important consideration when interpreting these results is the distinction between statistical and substantive significance. Several coefficients are statistically significant at conventional thresholds, yet their practical relevance cannot be assumed. In this study, average marginal effects for key predictors — particularly cloud computing and automation — range from approximately 5% to 8%, and are consistently observed across multiple environmental outcomes. In the behavioural and organisational sciences, such effect sizes are sometimes regarded as moderate but potentially meaningful, especially when they reflect stable associations across different models or subsamples (Gelman & Hill, 2006). However, in the absence of clear benchmarks or outcome-based valuation, we cannot make strong claims about their substantive impact. Statistical significance does not automatically translate into economic or policy relevance. What matters, therefore, is not only whether an effect is unlikely to have occurred by chance, but whether its magnitude can plausibly shape firm-level behaviour or inform policy design. This distinction is particularly relevant in cross-sectional research, where the goal is to uncover structural regularities rather than estimate causal impacts. In this sense, the value of the findings of this study lies less in the magnitude of individual effects and more in the consistency and directionality of the observed associations across environmental, digital and organisational dimension of the twin transition landscape.
This study explored the associations between digital adoption, organisational characteristics, and environmental engagement among European SMEs, based on data collected in 2019. The results reveal a differentiated yet coherent pattern, in which certain digital technologies — most notably cloud computing — are consistently linked to a wide range of environmental practices, while others, such as big data analytics and AI, exhibit more selective associations. These findings contribute to the growing body of research on the twin transition, offering a micro-level perspective that complements the predominantly macro-oriented literature.
Cloud computing shows the most consistent and statistically robust associations across all five environmental outcomes. Its technical features — scalability, dematerialisation, real-time data access — may help explain why firms that adopt cloud computing also tend to report higher levels of resource monitoring and operational efficiency, as also noted in policy-oriented assessments such as the JRC report (Joint Research Centre European Commission, 2023). Big data analytics shows strong associations with eco-innovation and sustainable product development, practices that typically require traceability, forecasting, and lifecycle analysis (Angelopoulos & Kontakou, 2020; Shajari & David, 2025). Its non-significance in the energy domain, despite expectations from existing literature, may reflect limited integration of data analytics with energy systems among SMEs, or a temporal lag in adoption at the time of data collection.
Another consistent finding concerns robotic automation, whose positive association with all five environmental indicators suggests a pattern consistent with the idea that automation can support environmental goals. These results are conceptually aligned with the perspective offered by Almuaythir et al.(Almuaythir et al., 2024), who frame digital automation as enabler of sustainability transitions through improvements in resource efficiency, emissions reduction, and cleaner production processes. Their review identifies a set of mechanisms through which automation can facilitate responsible consumption and climate action, particularly in the manufacturing and industrial sectors. While our analysis focuses on SMEs rather than large-scale systems, the observed association between automation and the adoption of environmental measures aligns with this conceptual framework.
AI and machine learning display a similar pattern: they are significantly associated with product-oriented and resource-saving behaviours, but not with recycling. While this is partly consistent with literature highlighting AI’s use in advanced optimisation and product design (Burinskienė & Nalivaikė, 2024; Shajari & David, 2025), it contrasts with more recent evidence showing successful AI applications in waste sorting and classification (Fang et al., 2023). This divergence could reflect sectoral differences, SME-specific barriers, or possibly the timing of the survey, which predates the acceleration in AI uptake in circular economy applications.
Turning to organisational factors, family ownership is positively associated with all environmental outcomes, a pattern consistent with the idea that long-term orientation and reputation-sensitive governance may foster sustainability-oriented behaviours (Miroshnychenko et al., 2025). Financial capability shows a significant effect only for sustainable product development, suggesting that while resource availability matters, it is not a sufficient condition for broader environmental engagement (Angelopoulos & Kontakou, 2020). In contrast, skill shortages — used here as a proxy for human capital constraints — display significant negative effects, particularly for recycling and resource reduction. These findings are consistent with the view that technical and digital competences play a critical role in enabling SMEs to engage in complex ecological practices (Bassi & Guidolin, 2021; Rupeika-Apoga & Petrovska, 2022).
An important methodological reflection concerns the distinction between statistical and substantive significance. While several effects are statistically robust, their practical relevance cannot be assumed. In this study, average marginal effects for key predictors typically range from 5% to 8%, and are consistently observed across multiple outcomes. In behavioural and organisational research, such effect sizes are not uncommon, and can be considered meaningful when supported by coherent and theoretically grounded patterns (Gelman & Hill, 2006). However, in the absence of external benchmarks or direct outcome validation, claims about impact should remain cautious. The value of these findings lies not in the magnitude of individual coefficients, but in the structured associations they reveal between digital infrastructures, organisational conditions, and environmental engagement.
It is also essential to consider the temporal and epistemic context of the data. Collected prior to the implementation of the European Green Deal, these results offer a baseline perspective on SME behaviour in a pre-transition landscape—before the onset of regulatory and financial incentives that have since reshaped the sustainability agenda. Furthermore, the data are self-reported and cross-sectional, and do not include objective environmental performance indicators. As such, the findings should be interpreted as perceptions and associations rather than as behavioural facts or causal effects.
In sum, this study contributes to a more differentiated understanding of the twin transition by showing how digital and organisational factors were already linked to environmental practices among European SMEs before the recent policy acceleration. While the analysis is limited by the cross-sectional and self-reported nature of the data, it provides a structured account of pre-existing patterns of association. Rather than offering predictive insight, the study documents a set of empirically grounded associations that characterised SME behaviour at a key turning point in the European sustainability agenda.