What works in digital transformation for agro-industry? Evidence from meso-institutional governance in the tomato value chain

Agri-food systems are undergoing profound transformations driven by the dual imperative of efficiency and sustainability, placing both innovation and coordination mechanisms at the core of agri-food competitiveness (Govindan, 2018). Across Europe, agri-food value chains are evolving towards tighter vertical coordination and collaborative governance to meet quality, safety and sustainability demands, although collaboration often remains constrained by trust asymmetries and heterogeneous actor capabilities (Folkerts and Koehorst, 1997; Matopoulos et al., 2007; Rossi et al., 2023).

Within this context EU strategies, including the Green Deal, the Farm to Fork and the reform of the Common Agricultural Policy, have promoted producer aggregation as enablers of innovation and sustainability transitions, facilitating the diffusion of technical and organisational innovations aligned with societal and market demands for environmentally responsible production (Weituschat et al., 2023; Cappella et al., 2025). However, the pace and success of innovation adoption in agriculture, especially digital technologies, remain uneven, revealing the interplay between structural, institutional and behavioural factors.

At the European level, the Common Market Organisation (CMO) has promoted producer cooperation through producer organisations (POs) as collective entities able to coordinate production, stabilise markets and facilitate innovation uptake (Blasi et al., 2017; ISMEA, 2022; Nasso et al., 2024).

Against this backdrop, the Italian processed tomato supply chain (PTSC) provides a paradigmatic case where vertical coordination, producer aggregation and institutional governance intersect to sustain competitiveness and foster innovation. The sector operates under two meso-institutional inter-branch organisations, OI Nord Italia and OI Centro Sud Italia, coordinating production, contractual frameworks and sustainability standards across their respective geographical areas (Mantino and Forcina, 2018; Samoggia et al., 2022; Čechura et al., 2024). Although Italy remains the largest processed tomato producer in Europe and the third worldwide in 2022 and 2023, the sector faces rising competition from low-cost exporters (Čechura et al., 2024), making digitalisation and governance capacity critical for maintaining competitiveness.

In this context, POs act as pivotal intermediary governance nodes combining horizontal coordination among farmers with vertical linkages to downstream processors, facilitating information sharing, traceability and risk–reward alignment across production tiers (Hamprecht et al., 2005). Through technical assistance, decision support systems (DSS) platforms and contractual monitoring mechanisms, POs align individual incentives with collective goals, functioning as active mediators of digital transformation within hybrid agri-food chains.

Within fruit and vegetable production systems, technology acceptance among farmers is shaped by a complex interplay of technological, organisational and social determinants. Studies have shown that farmers’ perceptions of usefulness and institutional support critically affect their willingness to adopt innovations (Aubert et al., 2012), while perceived risk, social influence and trial opportunities may influence resistance or openness to change (Naspetti et al., 2017; Beeke et al., 2024; Califano et al., 2026). Understanding these factors is essential to explain why technology diffusion remains heterogeneous, even within well-structured and competitive supply chains. Therefore, this study investigates not only the individual determinants of adoption but also how supply chain coordination and governance mechanisms shape the diffusion of technological innovations within producer-led contractual systems.

To addresses these challenges, this study adopts the technology acceptance model (TAM) (Davis et al., 1989) as the theoretical framework to analyse the determinants of digital technology adoption among farmers within the Italian PTSC. Existing TAM applications in agri-food contexts, including extensions that incorporate social influence, facilitating conditions and institutional support, predominantly treat adoption as a single-stage phenomenon, failing to distinguish between the cognitive processes governing initial uptake and those sustaining continued use (Karahanna et al., 1999; Bhattacherjee, 2001). In supply chains where digital tools are introduced through intermediary organisations such as POs, this limitation is particularly consequential: the governance mechanisms lowering barriers for prospective adopters may operate through different pathways than those sustaining engagement among experienced users.

The present study addresses this gap through two interrelated research question:

By integrating the analysis of behavioural determinants of technology acceptance (RQ1) with the institutional and relational functions of POs (RQ2), this study examines how vertically coordinated supply chains can enable digital transition, drawing on data from 260 farmers in the Italian PTSC distinguishing DSS adopters from non-adopters.

The empirical contribution rests on two analytically distinct but complementary elements. Firstly, rather than estimating a single pooled TAM model, the study develops and tests two parallel structural equation models, one for non-users capturing adoption intention (AI) and one for current users, capturing continued DSS use. This stage-differentiated design surfaces asymmetric adoption mechanisms that remain concealed in conventional single-group applications (Taylor and Todd, 1995; Benbasat and Barki, 2007).

Secondly, the study combines behavioural modelling with a structured descriptive-comparative analysis of current contractual arrangements and farmers’ governance preferences. Following a sequential explanatory logic, in which the profiles identified through the SEM serve as the interpretive lens for the contractual strand, this design links digital adoption profiles to heterogeneous governance expectations within a coordinated supply chain (Venkatesh et al., 2013; Creswell and Plano Clark, 2023). This approach extends the interpretive framework beyond individual cognition to encompass supply chain governance as a structural determinant of innovation behaviour (Ménard and Valceschini, 2005; Suvanto and Lähdesmäki, 2023) and allows to capture how coordination mechanisms shape innovation outcomes (Azorín and Cameron, 2010). Together, these contributions address a gap that persists in SCM research on digital transformation: existing work has examined how governance structures, contractual, relational and institutional, enable or constrain coordination in agri-food chains (Fischer, 2013; Suvanto and Lähdesmäki, 2023), yet the micro-level behavioural mechanisms through which upstream producers respond to those governance arrangements and why identical coordination structures produce divergent adoption outcomes across heterogeneous actor populations, remain under theorised. By demonstrating that adoption-stage position conditions which governance mechanisms are effective and for whom, this study contributes empirical evidence and a theoretical proposition that advances the integration of behavioural and governance perspectives in SCM, a direction identified as a priority in research on digital supply chain transformation (Keller et al., 2024).

The remainder of the paper is structured as follows. Section 2 develops the theoretical framework and hypotheses; Section 3 describes data and methods; Section 4 presents results; Sections 5 and 6 discuss findings, implications and limitations.

The present study adopts the TAM (Davis et al., 1989) to investigate farmers’ behavioural responses towards digital DSS within the Italian processed-tomato supply chain. TAM posits that the acceptance of a technology is primarily driven by perceived usefulness (PU), the degree to which users believe a system enhances their performance and perceived ease of use (PEU), the degree to which they believe it requires minimal effort. These two cognitions shape an individual’s attitude towards adoption (AA) and, ultimately, their adoption decision (AD) or AI, considering the split of the analysis between users and non-users of technology. TAM’s robustness is well established across technological and sectoral contexts (Venkatesh and Davis, 2000; King and He, 2006), including agricultural settings (Rezaei-Moghaddam and Salehi, 2010; Aubert et al., 2012; Pierpaoli et al., 2013).

While TAM and its subsequent extensions, including the unified theory of acceptance and use of technology, have incorporated social influence, facilitating conditions and institutional context (Venkatesh et al., 2003; Venkatesh et al., 2012), these frameworks predominantly treat adoption as a single-stage phenomenon, conflating pre-adoption intention with post-adoption continued use. Prior research has shown that the cognitive antecedents of initial adoption intention differ from those governing continued use, since post-adoption evaluations are shaped by direct experience and accumulated learning rather than anticipatory beliefs alone (Karahanna et al., 1999; Bhattacherjee, 2001). Subsequent work has corroborated this stage-based logic across a variety of technological and organisational contexts, consistently finding that the strength and directionality of structural relationships between TAM constructs shift significantly once users have accumulated hands-on experience with a technology (Thong et al., 2006; Liao et al., 2009; Venkatesh et al., 2012). In agricultural settings, where digital tools such as DSS are often introduced through intermediary organisations, this distinction becomes particularly relevant: governance mechanisms lowering barriers for prospective adopters operate through different pathways than those sustaining engagement among experienced users. Accordingly, the study estimates two parallel yet structurally comparable SEM models, one for non-users and one for current users, enabling a stage-differentiated analysis of technology acceptance that is theoretically grounded in the distinction between intention formation and continued-use reinforcement. This design is the basis for the hypothesis development, as detailed below.

Technology adoption in agriculture is embedded within social, organisational and institutional environments (Eastwood et al., 2019). In the Italian tomato sector, POs and OIs play a decisive role in promoting innovation by providing agronomic advice, digital infrastructures and economic incentives. Accordingly, this study extends the traditional TAM by incorporating two contextual constructs that reflect the collective dimension of technological change: support quality (SQ) for current users and perceived support (PS) for non-users. These constructs capture farmers’ evaluation of the reliability, accessibility and competence of technical support provided by POs or consultants. Prior evidence indicates that the perceived quality of advisory services significantly increases both PEU and the intention to continue using digital tools (Long et al., 2016; Gupta et al., 2024). Thus, the first hypothesis states that:

Environmental and market conditions also shape farmers’ perceptions of technology performance. The construct environment (ENV) represents external pressures such as policy incentives, climate variability and social norms that may enhance or hinder digitalisation. Empirical studies have shown that favourable environmental conditions, including infrastructural readiness and peer influence, facilitate both PEU and AI (Karahanna et al., 1999; Tey and Brindal, 2012). Hence, it is expected that a supportive environment will exert a positive effect on PEU:

According to TAM when users find a system effortless, they are more likely to perceive it as beneficial (Venkatesh et al., 2003). In agriculture, studies on precision farming and smart irrigation confirm that usability perceptions strongly predict perceived performance benefits (Aubert et al., 2012; Bagheri et al., 2021; Cappella et al., 2026). Therefore:

The subsequent link between usefulness and AA reflects the instrumental motivation of farmers: technologies considered helpful for improving productivity or quality are more likely to generate favourable attitudes and adoption behaviour (Rezaei-Moghaddam and Salehi, 2010). Prior studies also suggest that PEU may directly influence attitude when the technology reduces workload or cognitive complexity (Aubert et al., 2012; Hasler et al., 2017). Hence:

The final stage of the behavioural chain connects cognitive and attitudinal beliefs to the actual or intended adoption behaviour. According to the theory of planned behaviour (Ajzen, 1991) and extended TAM frameworks (Venkatesh and Davis, 2000; Rezaei et al., 2020), attitude is a primary determinant of adoption intention, while PU may also exert a direct motivational effect. Contextual conditions may further shape behaviour, as supportive environment strengthen the translation of intention into action (Naspetti et al., 2017; Wang et al., 2019). Therefore:

Together, these hypotheses are tested through two parallel yet distinct structural models. For non-users, the model includes the constructs PS, PEU, PU, ENV, AA and AI. For users, the model mirrors this structure but replaces PS with SQ, reflecting the evaluation of experienced assistance and substitutes AI with AD, capturing the continuation or reinforcement of DSS use. Consistent with the hybrid governance logic characterising PO-mediated supply chains (Williamson, 1996; Ménard and Valceschini, 2005), this dual-model specification allows governance mechanisms to exert stage-specific effects on the adoption pathway rather than constraining them within a single pooled structure assuming homogeneous adoption processes across users and non-users (Karahanna et al., 1999) (Figure 1).

Before analysing the evolution of governance mechanisms within the case studies, we first reviewed the contractual frameworks regulating the processed tomato sector in Italy and Europe. In line with the CMO (Regulation EU No 1308/2013) and the national interprofessional agreements coordinated by the Ministry of Agriculture (Ministero dell’Agricoltura and della Sovranità Alimentare e delle Foreste (MASAF), 2025), contracts between farmers, POs and processors are formally defined and often negotiated within framework agreements at regional or interprofessional level. This preliminary contextualisation allowed us to interpret the observed governance arrangements, such as contract-framework and cooperative schemes, co-defined by inter-branch organisation (IO) within the existing regulatory setting that structures the European and Italian tomato value chain. The empirical investigation was based on a structured questionnaire designed to capture farmers’ perceptions, attitudes and behavioural intentions towards digital DSS. The instrument was structured into four sections, following established procedures for survey-based technology acceptance research (Davis et al., 1989; Venkatesh and Davis, 2000).

Section 1 collected socio-demographic and structural data. The second explored the current use of digital tools, capturing both the frequency and mode of use (autonomous, with consultant support or fully delegated). Section 3 comprised the psychometric constructs operationalising TAM and its extensions. Multi-item measures were adapted from validated scales in the technology adoption and agri-digitalisation literature and adjusted to the agricultural context. Respondents indicated their level of agreement on a five-point Likert scale (1 = “strongly disagree”, 5= “strongly agree). At this stage, the questionnaire was split into two parallel versions according to respondents’ experience with DSS: one for users, focusing on continued and one for non-users, adapted to measure future adoption intention (an English translation of the constructs is provided in Table A1 in the  Appendix).

Construct measures were adapted from validated scales: PU and PEU from Aubert et al. (2012); AA from Rezaei-Moghaddam and Salehi (2010); AD and AI from Davis et al. (1992) and Naspetti et al. (2017); SQ and PS from Long et al. (2016); ENV from Yeo and Keske (2024) and Wang et al. (2019).

Section 4 addressed contractual relationships and service provision, capturing both the content of current agreements between farmers and buyers or Pos and preferred future contractual arrangements. Following a sequential explanatory logic (Creswell and Plano Clark, 2023; Venkatesh et al., 2013), the behavioural profiles identified through the SEM served as the interpretive framework for the contractual analysis. This strand operates as a descriptive-comparative analysis examining whether the behavioural segmentation emerging from the SEM corresponds to systematically different governance expectations across the two groups, consistent with mixed-method approaches in organisational and supply chain research, in which a quantitative strand informs the interpretive lens of a complementary descriptive component (Edmondson and McManus, 2007; Teddlie and Tashakkori, 2010) and with recent SCM studies integrating behavioural and institutional evidence to examine how coordination mechanisms shape innovation behaviour at the supply chain level (Suvanto and Lähdesmäki, 2023; Ménard and Valceschini, 2005). Its analytical contribution is not to test additional hypotheses, but to translate adoption profiles into governance-level signals relevant for supply chain design.

The questionnaire was translated into Italian following a back-translation procedure (Brislin, 1980) and reviewed by an expert panel of eight researchers, agronomists and PO representatives. A pilot test with 25 farmers confirmed internal consistency and readability.

The data were collected through a nationwide survey conducted between June and July 2025, targeting farmers and producers engaged in the Italian processed tomato supply chain. The survey was implemented as part of the UNIEAT project, coordinated by the University of Tuscia in collaboration with POs and Confagricoltura, the main national farmers’ association. These institutional partners played a pivotal role in ensuring sectoral representativeness and in facilitating access to respondents spread across major production areas in both Northern (OI Nord) and Southern (OI Sud) Italy.

The survey was administered online via Microsoft Forms. Participants received an information sheet detailing the study’s objectives, confidentiality terms and consent procedures in accordance with the EU General Data Protection Regulation (GDPR 2016/679). Participation was voluntary, and responses were anonymised before analysis. After data cleaning, 260 valid responses were retained, exceeding the minimum sample size requirements for SEM based on a 10:1 observation-to-parameter ratio (Kline, 2023). Of these, 118 respondents (45.4%) were current DSS users, while 142 (54.6%) were non-users, providing two distinct yet comparable groups for the dual TAM model.

The respondents covered 27 provinces and reflected the national distribution of processed tomato production reported by ISTAT (2020), with 83% of the sample located in Northern Italy (mainly Emilia-Romagna, Lombardy and Piedmont) and 17% in Southern regions (Campania, Puglia and Basilicata).

Table 1 reports the socio-demographic characteristics of the surveyed farmers. The sample was predominantly male (85.4%), with a mean age of 52 years. DSS users were approximately six years younger than non-users and showed higher tertiary education (33% vs 19%), suggesting education plays a facilitative role in digital engagement. Most respondents farmed full-time (85.4%), with an average of 24 years of experience and 19.7 ha under tomato cultivation.

Table 2 reports the total utilised agricultural area (UAA) and the area under tomato cultivation for both users and non-users of DSS, together with national reference values from the 7th Agricultural Census (ISTAT, 2020). The sample accounts for roughly 6% of the national area dedicated to industrial tomato production, indicating that it represents a substantial share of the country’s specialised production base. Moreover, nearly 60% of the tomato area in the sample is rented land, reflecting the structural characteristics and land-use flexibility typical of specialised intensive cropping systems in the sector.

The adequacy of the measurement model was examined through confirmatory factor analysis (CFA) following the methodological guidelines of Kline (2023). Construct reliability was assessed using AVE, Cronbach’s α with 95% confidence intervals (Trinchera et al., 2018) and the omega coefficient (Trizano-Hermosilla and Alvarado, 2016).

SEM estimation used the maximum likelihood method with robust standard errors and mean–variance adjustment (MLMV) due to significant multivariate non-normality (non-users’ statistic = 6921.361, p < 0.001; users’ statistic = 3073.143, p < 0.001) (Asparouhov and Muthén, 2010; Maydeu-Olivares, 2017).

Model goodness-of-fit was evaluated using the Satorra–Bentler scaled chi-square statistic (SBχ²) (Satorra and Bentler, 2001), comparative fit index (CFI), Tucker–Lewis Index (TLI), root mean square error of approximation (RMSEA) and standardised root mean square residual (SRMR) (Hu and Bentler, 1999).

Discriminant validity was assessed via the HTMT criterion (Henseler et al., 2015) and indirect effects were tested using bootstrap standard errors with 5,000 replications. Measurement invariance was tested on the four shared constructs (PEU, PU, ENV, AA) following Vandenberg and Lance (2000), estimating configural, metric and scalar models using MLMV. Model comparisons used the scaled chi-square difference test alongside ΔCFI ≤ −0.010 and ΔRMSEA ≤ +0.015 criteria (Cheung and Rensvold, 2002; Chen, 2007). Coefficient difference tests for shared structural paths used the z-test procedure of Paternoster et al. (1998). All analyses were conducted in RStudio (version 2025.09.1 + 401) using the Lavaan package (Rosseel, 2012), complemented by semTools (Jorgensen, 2016) and semPower (Moshagen and Bader, 2024) for power and model evaluation routines.

To assess potential common method bias (CMB), we applied the post hoc marker-variable technique (Podsakoff and Organ, 1986; Podsakoff et al., 2003). Following the procedure proposed by Lindell and Whitney (2001) and refined by Malhotra et al. (2006), a theoretically unrelated variable was included to estimate the shared variance attributable to common method effects. After adjusting the covariance matrix using this marker-variable approach, the magnitude and significance of the structural path coefficients remained largely stable, indicating that common method variance did not substantially bias the model estimates. A detailed comparison of pre- and post-CMB-corrected path coefficients is provided in Table A2 in the  Appendix.

Table 3 presents reliability statistics for all latent constructs. For both groups, 95% CI lower bounds for α exceed 0.76, confirming satisfactory internal consistency. The omega coefficients (ω), ranging from 0.766–0.964, further corroborate the internal consistency of the indicators, offering a robust estimate under congeneric measurement conditions (Trizano-Hermosilla and Alvarado, 2016). Similarly, the AVE values are above the 0.50 criterion, demonstrating adequate convergent validity (Hair et al., 2021).

Factor loading point estimates are all above 0.60, supporting a good validity of the measurement model (for more details see Table A3 in the  Appendix). One item of the PS construct (PS4: “To use DSS effectively, I would need continuous support”) was excluded from the final measurement model during CFA-based measurement validation, as its factor loading fell below the commonly recommended levels for indicator reliability (λ = 0.457; Hair et al., 2021). Discriminant validity (Table 4), assessed through the HTMT criterion, also confirmed adequate differentiation among latent variables, with all pairwise correlations below 0.85 (Henseler et al., 2015; Kline, 2023).

Both structural models exhibited a satisfactory overall fit to the data according to robust estimation criteria (Browne and Cudeck, 1992; Hu and Bentler, 1999). For non-users, the fit indices were: SBχ²(141) = 187.960, robust CFI = 0.940, robust TLI = 0.927, SRMR = 0.073 and robust RMSEA = 0.080 (90% CI [0.046, 0.109]). For users, the corresponding indices were: SBχ²(141) = 175.826, robust CFI = 0.996, robust TLI = 0.995, SRMR = 0.052 and robust RMSEA = 0.032 (90% CI [0.012, 0.046]). In both cases, the ratios of chi-square to degrees of freedom (χ²/df = 1.33 for non-users; 1.25 for users) were well below the threshold of 3, indicating a parsimonious and well-fitting model (Schermelleh-Engel et al., 2003).

Table 5 reports standardised path coefficients, standard errors and significance levels for all hypothesised relationships. Overall, the results confirm the robustness of the proposed TAM-based framework, though with some distinctions between non-users and users. Observed differences in the pattern of significant paths are interpreted as within-group findings, consistent with a stage-based conceptualisation of technology adoption (Karahanna et al., 1999). The formal cross-group validity of these differences, assessed through measurement invariance analysis and coefficient difference tests is presented at the end of this section.

For non-users, PS exerts a strong and significant positive influence on PEU (β = 0.745, p < 0.001), supporting H1. The environmental context (ENV) also positively affects PEU (β = 0.124, p < 0.10), confirming H2 and as expected, PEU significantly predicts PU (β = 0.931, p < 0.001), in line with H3. Regarding attitudinal dynamics, neither H4a (PEU → AA) nor H4b (PU → AA) reached statistical significance among non-users. However, at the behavioural level, the ENV and AA emerged as the most influential determinants of AI, with strong positive effects (β = 0.289, p < 0.001 and β = 0.707, p < 0.001, respectively), supporting H5c and H5d. In addition, the path from PU to AI (β = 0.633, p < 0.10) was marginally significant, lending partial support to H5b. Conversely, PEU has no direct effect on AI (H5a rejected). For users, the model revealed a more consolidated cognitive pathway. Both SQ and ENV significantly enhance PEU (β = 0.346, p < 0.001; β = 0.553, p < 0.001), confirming H1 and H2. PEU also exerts a strong positive influence on PU (β = 0.838, p < 0.001), supporting H3, while its effect on AA is particularly pronounced (β = 0.811, p < 0.001), validating H4a. The effect of PU on AA is non-significant, leading to the rejection of H4b.

At the behavioural level, both PEU (β = 0.458, p < 0.05) and attitude of adoption (β = 0.338, p < 0.01) significantly predict the AD, supporting H5a and H5d. The path from PU to AD is significant (β = 0.250, p = 0.049), supporting H5b, while the environmental factor shows no effect, rejecting H5c. Figure 2 shows the graphical results of the two TAM models.

The configural model confirmed equivalent factor structures across groups. Metric invariance was fully supported (ΔCFI = 0.000; ΔRMSEA = −0.003; Δχ²(9) = 8.66, p = 0.469), indicating that factor loadings are statistically equivalent between users and non-users. Full scalar invariance was not supported (ΔCFI = −0.022; ΔRMSEA = +0.018; Δχ²(9) = 65.87, p < 0.001), as four items, ENV_1, ENV_2, ENV_3 and PEU_1, exhibited systematically higher intercepts in the users group, reflecting the structurally elevated perceptions of peer exposure and ease of use associated with direct DSS experience (Byrne et al., 1989). A partial scalar model freeing these four intercepts met the invariance criteria (ΔCFI = −0.006; ΔRMSEA = +0.004; Δχ²(5) = 16.99, p = 0.005), confirming partial scalar invariance (Vandenberg and Lance, 2000). Fit indices for all invariance models are reported in Table 6.

To formally assess whether the observed differences between groups reflect statistically distinct mechanisms rather than within-group sampling variation, coefficient difference tests were conducted for the shared structural paths, following the procedure proposed by Paternoster et al. (1998). As reported in Table 7, only ENV → PEU relationship differed significantly across groups (β∼non-users∼ = 0.124, SE = 0.066; β∼users∼ = 0.553, SE = 0.086; z = −3.957, p = 0.000), confirming that the role of the environmental context in shaping PEU is significantly stronger among users than non-users, a pattern consistent with the partial scalar non-invariance identified for the ENV items. By contrast, the remaining three shared paths, PEU → PU (z = 1.308, p = 0.191), PEU → AA (z = −1.082, p = 0.279) and PU → AA (z = 0.649, p = 0.516), did not differ significantly between groups.

This section examines contractual arrangements and governance preferences across the two behaviourally distinct groups identified through the SEM, assessing whether security- and innovation-oriented profiles correspond to divergent supply chain coordination expectations (Figure 3).

Among current arrangements, the most common contractual clauses concern quantity specifications (25.9% of users and 24.1% of non-users), input-use restrictions (23.8% and 19.8%, respectively), traceability requirements (22% and 20%) and quality-related bonuses (17.5% and 18.1%). Fixed-price clauses are relatively rare, present in only 5% of contracts among users and 6% among non-users. These results indicate that both groups operate under structured yet heterogeneous contracts that include a combination of production and quality parameters.

PO services are accessed by 75.4% of users and 54.9% of non-users, covering consulting (39% vs 34%), input supply (25%–26%) and commercial management (25% vs 20%). This differential embeddedness is consistent with the SEM profiles: among non-users, lower PO access leaves the PS→PEU pathway latent, a structural inhibitor rather than a neutral developmental stage. Among users, 45.8% adopt DSS with consultant support and 31.4% use them daily.

The section on future contractual expectations revealed distinct priorities between the two groups (Figure 4). Non-users prioritise minimum price guarantees (25.9%) and technological investment support (12.6%), consistent with their security-oriented SEM profile: a demand for contractual arrangements that externalise risk and delegate support responsibilities, minimum price guarantees as downside protection and investment support as a substitute for autonomous technological capability, rather that reward.

Conversely, users prioritised quality parameters (19.9%), multi-annual commitments (13.5%) and facilitated credit access (11.6%), mirroring their innovation-oriented SEM profile, where continued adoption is driven by internalised, experience-based evaluations. At the governance level, this translates into a demand for contractual arrangements that reward performance and sustain long-term engagement, rather than providing downside protection against adoption risk. Despite these differences, both groups expressed limited interest in new variety incentives or insurance support clauses.

Comparing current contracts and wishlist preferences, minimum price clauses are rarely present in existing agreements (4.5% users; 9.3% non-users) yet highly requested for future arrangements (22%–26%), while quality parameters and technological investments remain aspirational. Conversely, traceability and input restrictions, already widespread, are less frequently cited as future priorities, suggesting they have become established contractual features.

This study provides new empirical evidence on the behavioural and institutional determinants shaping farmers’ acceptance and resistance towards digital farming technologies within a highly coordinated agri-food supply chain. By combining TAM with the relational and governance dimensions of SCM, the analysis highlights how cognitive, contextual and organisational factors jointly influence technology adoption.

The analysis was guided by two research aims: firstly, to identify the behavioural and contextual drivers shaping farmers’ intention to adopt digital tools; and secondly, to explore how the organisational mechanisms of coordination and support embedded in POs and IOs shape innovation behaviour in the Italian processed tomato supply chain. Taken together, the findings show that digitalisation in hybrid agri-food chains cannot be reduced to individual attitudes or to contractual design alone but emerges from their interaction within specific governance arrangements.

With respect to the first research question, the results demonstrate that the environmental context exerts a direct and significant influence on AI among non-users, while the same effect disappears among users. This suggests that in the early stages of adoption, farmers are particularly sensitive to enabling environments, comprising peer influence, infrastructural readiness and institutional incentives (Eastwood et al., 2019; Tey and Brindal, 2012; Sun et al., 2022). For non-users, perceiving a supportive environment fosters awareness and motivation to adopt digital innovations. This outcome confirms that innovation does not occur in isolation but emerges through favourable ecosystems of trust, learning and demonstration (Klerkx and Begemann, 2020), highlighting the importance of collective learning mechanisms within supply chains, consistent with sustainable SCM studies showing that participatory engagement increases technology legitimacy (Nyaga et al., 2010; Fischer, 2013).

Among users, environment acts as a background condition rather than a differentiating driver, consistent with stage-based adoption theory (Karahanna et al., 1999). For current users, the results validate the core assumptions of TAM: PEU significantly affects both PU and AA, which in turn predict actual adoption behaviour (Davis et al., 1989; Venkatesh and Davis, 2000). However, in the case of non-users, the relationships between PEU and Attitude and between PEU and AI are non-significant, indicating that perceptions of ease of use are not yet internalised. For them, ease of use is externally derived from the perceived quality of technical support and peer influence, rather than from direct experience.

The coefficient difference tests reported in Section 4.2, reveal a nuanced picture of adoption asymmetry. The structurally distinct path ENV → PEU confirms that environmental context exerts a significantly stronger effect on PEU among users than non-users, consistent with the partial scalar non-invariance identified for ENV items and with the experiential grounding of usability perceptions among adopters. Importantly, while ENV shapes PEU more strongly among users, it is among non-users that ENV exerts a direct and significant effect on adoption intention, a path that is non-significant among users. By contrast, the core cognitive chain (PEU → PU → AA) operates equivalently across groups, suggesting that the fundamental TAM mechanism is robust to adoption stage. Observed differences thus reflect divergences in antecedent conditions and outcome variables rather than a structural reconfiguration of the cognitive pathway. This behavioural asymmetry implies that farmers’ decision-making processes are intertwined with the relational and informational structures provided by POs, reinforcing the need to interpret adoption as embedded in supply chain governance rather than as an isolated cognitive act.

Addressing the second research question, the contractual analysis reveals that farmers’ existing agreements are primarily structured around production, quality and traceability clauses, whereas future preferences express a clear demand for minimum price guarantees and technological investment support. This discrepancy reveals a latent tension between transactional safeguards and developmental incentives (Ménard and Valceschini, 2005; Samoggia et al., 2022). The “wishlist” analysis reveals systematic differences: non-users rank minimum guaranteed price as their primary priority, followed by technological investments support, while users favour quality parameters, multi-annual commitments and facilitated credit access. This inversion reveals two distinct behavioural profiles, a security-oriented group seeking downside protection and an innovation-oriented group willing to trade price stability for performance incentives and long-term partnerships, consistent with entrepreneurship research, which shows that more innovative farmers display higher risk-taking and opportunity-seeking orientations (Miller, 2011; Lumpkin and Dess, 1996).

This correspondence demonstrates that the cognitive segmentation identified through the SEM has a concrete institutional counterpart in farmers’ governance expectations. The contractual analysis functions as a second layer of evidence corroborating the SEM findings, translating adoption profiles into governance-level signals relevant for supply chain design. Together, the two strands confirm that behavioural heterogeneity is not merely attitudinal but embedded in distinct governance logics, a finding that could not have emerged from either strand in isolation.

Taken together, the findings carry a direct theoretical implication for SCM governance research. Existing hybrid governance models have theorised coordination primarily in terms of contractual design and transaction cost minimisation (Ménard and Valceschini, 2005), while relational governance frameworks have emphasised trust and commitment as integration drivers (Nyaga et al., 2010; Suvanto and Lähdesmäki, 2023). The present study demonstrates that neither strand fully accounts for digital transition dynamics in agri-food chains, because the effectiveness of both contractual and relational mechanisms is conditioned by the adoption-stage position and behavioural heterogeneity of upstream producers, a structural dimension that existing coordination models do not capture. Security-oriented and innovation-oriented producers respond to identical governance arrangements in systematically different ways, generating divergent adoption outcomes that cannot be explained by contractual design or relational trust alone. Incorporating actor-level behavioural segmentation into supply chain governance models is therefore not merely a theoretical refinement but a necessary condition for explaining and predicting digital transition outcomes across heterogeneous producer populations.

This theoretical proposition has direct practical implications for how hybrid governance is designed in coordinated agri-food chains. Recent SCM scholarship has established that effective governance requires contractual and relational mechanisms to coexist (Williamson, 1996; Suvanto and Lähdesmäki, 2023); the present findings extend this by showing that the composition of those mechanisms must be calibrated to the adoption profile of the producer base rather than applied uniformly. In the Italian PTSC, POs combine horizontal cooperation with vertical integration, facilitating collective bargaining and innovation diffusion, but may also generate new asymmetries if communication and trust are not adequately managed (Čechura et al., 2024). Strengthening these relational interfaces is essential to transform coordination from compliance-oriented to learning-oriented governance. Our findings indicate that this alignment cannot be one-size-fits-all: POs need to calibrate their service and contractual offer to adoption profile, combining minimum price guarantees and risk-sharing instruments for security-oriented farmers with quality-based premiums, multi-annual commitments and credit facilitation for innovation-oriented ones. Designing such tailored hybrid arrangements is a key lever to translate digital tools into effective and inclusive supply chain innovation.

From a managerial standpoint, the findings suggest that digital transition strategies in agri-food supply chains should be approached as collective processes rather than individual technological choices. The study’s theoretical contribution to SCM is threefold: behavioural heterogeneity among upstream producers constitutes a governance variable conditioning the effectiveness of hybrid coordination arrangements; TAM constructs function not merely as individual-level cognitions but as signals of inter-organisational coordination quality; and POs are repositioned as active governance variables mediating digital transition outcomes across the chain rather than as contextual background factors. These contributions ground the following practical implications.

For POs and interbranch institutions, four implications emerge. Firstly, strengthen enabling environments for trust and learning is essential, particularly for non-users’ whose adoption intention is strongly influenced by environmental support. POs and processors should invest in demonstration platforms, peer exchanges and transparent communication to facilitate informal learning communities, strengthening PEU and reducing uncertainty in line with “learning-oriented governance” models (Fischer, 2013; Eastwood et al., 2019). This is directly grounded in the empirical results: ENV exerts a significant direct effect on adoption intention among non-users but not among users, confirming that environmental enablers are most critical at the pre-adoption stage, where they substitute for experiential evaluations. Secondly, institutionalise relational incentives and behavioural segmentation in contract design. Traditional contracts, centred on quantity and compliance, are insufficient to stimulate digital innovation: hybrid contracts combining performance-based rewards (e.g. bonuses for data sharing, quality improvement or environmental compliance) with stability clauses (e.g. multiannual commitments, guaranteed minimum prices) can foster mutual commitment and balance risk (Ménard, 2004; Nilsson et al., 2014). Processors should define incentive schemes aligned with data governance and quality upgrading, anchoring digital tools within broader supply chain objectives. This is grounded in the contractual evidence: non-users ranked minimum price guarantees as their primary priority, while users favoured quality parameters and multiannual commitments, a divergence that mirrors the security-oriented versus innovation-oriented profiles identified in the SEM. Differentiated contractual bundles are therefore an empirically indicated necessity, not merely a governance preference.

Thirdly, enhance advisory and support infrastructures as relational assets. The strong positive effect of PS and SQ on PEU underscores that the relational performance of technical advisors and POs is as critical as technological performance. Investing in digital intermediaries, specialists bridging agronomy, data management and market relations, can improve adoption consistency across heterogeneous farmers (Long et al., 2016; Gupta et al., 2024). The empirical basis is twofold: PS exerts the strongest effect on PEU among non-users, indicating that external mediation substitutes for direct experience at the pre-adoption stage; and the ENV→PEU path differs significantly across groups, confirming that advisory quality operates differently across the adoption trajectory.

Fourth, design digital governance architectures at the supply chain level. Behavioural heterogeneity between users and non-users constitutes a structural signal for downstream actors, processors, branded manufacturers and inter-branch organisations, who must account for their supplier base’s adoption profile when designing digital strategies. Processors sourcing from POs with high non-users proportion face a supply base characterised by security-oriented preferences, low institutional embeddedness and reliance on external mediation: in this context, embedding minimum price guarantees, investment support clauses and structured advisory provisions directly into the processor–PO framework agreement becomes a precondition for activating upstream digital adoption rather than an optional incentive. Conversely, innovation-oriented bases warrant performance-based contracts and data-sharing incentives. This differentiation implies that digital governance in agri-food supply chains requires vertical alignment across contract tiers, such that the clauses negotiated between processors and POs are coherent with and reinforcing of the contractual arrangements between POs and their member farmers. Such multi-level contract coherence is a necessary condition for digital tools to generate value across the full chain rather than remaining confined to individual nodes (Keller et al., 2024). More broadly, these findings suggest that supply chain actors at different tiers face asymmetric digital adoption environments that call for differentiated governance responses, a dynamic that existing SCM literature has begun to document in agri-food contexts but has rarely traced back to the behavioural heterogeneity of upstream producers (Belhadi et al., 2024).

For policymakers, the findings point to a specific and empirically grounded limitation of current agri-digitalisation strategies. The pronounced gap between current contractual arrangements and non-adopters’ governance expectations, particularly the near-absence of minimum price guarantees despite being their primary priority, signals a measurable mismatch between IO-level incentive structures and the risk profiles of the non-adopting segment. Public policy should support revision of interprofessional contract frameworks under the CMO to incorporate differentiated clauses, transforming IOs into active instruments of digital transition governance (Nyaga et al., 2010; Govindan, 2018; Keller et al., 2024).

Several limitations must be acknowledged. Firstly, the single supply chain focus may limit the generalisability of the results to other commodities or institutional contexts. Extending the analysis to other agri-food chains would help assess whether the behavioural and organisational mechanisms identified here are context-specific or universal.

The cross-sectional design captures a snapshot of perceptions without accounting for temporal dynamics. Longitudinal studies could trace how trust, support and usability evaluations evolve as farmers transition across adoption stages.

Full scalar invariance was not achieved for ENV and PEU items, implying that latent mean comparisons across groups should be interpreted with caution. Moreover, the constructs PS and SQ and the outcome variables AI and AD, are conceptually non-equivalent by design and were excluded from formal invariance testing. Cross-group comparisons involving these constructs remain descriptive rather than formally validated and future research using longitudinal designs could track how PS evolves into SQ as farmers transition from non-adoption to adoption, thereby enabling a more rigorous test of stage-based theoretical propositions.

Although the survey instrument mitigated common method bias and achieved high reliability, self-reporting limitations may persist, particularly regarding sensitive issues like contractual satisfaction or institutional trust. Future research could integrate qualitative methods, such as in-depth interviews or participatory observation, to enrich understanding of relational governance and decision-making heuristics.

It should also be acknowledged that the RMSEA upper confidence bound for the non-users structural model (0.109) marginally exceeds the conventional 0.08 threshold and the configural model estimated in the measurement invariance analysis yields an RMSEA of 0.084. While these values are marginal and do not disqualify the models, given that robust CFI and TLI remain within acceptable ranges and the RMSEA point estimate falls below 0.08, they suggest that the non-users measurement structure may benefit from further refinement in future research, particularly in studies where cross-group comparability is of primary analytical interest (Browne and Cudeck, 1992).

Furthermore, PO-level heterogeneity might introduce unobserved institutional variance that could not be fully captured through farmer-level survey data; multilevel models or clustered designs would disentangle individual from organisational effects.

Another limitation concerns potential self-selection bias: farmers who adopt DSS may already possess higher innovation propensity or lower risk aversion, which could also influence their contractual preferences. Addressing this through propensity score matching or instrumental variables would strengthen causal inference.

Finally, causal directions between PS and PEU may be bidirectional, with farmers who experience smooth tool use retrospectively evaluating support as more effective. Longitudinal or mixed-method approaches could clarify these dynamics.

Future research should extend this framework through multi-actor perspectives, incorporating processors, consultants and policymakers and comparative studies across governance models.

Further theoretical development integrating TAM with supply chain relational theory and psychological ownership (Suvanto and Lähdesmäki, 2023), would enriching the behavioural-governance nexus identified here.

This study contributes to the understanding of how behavioural and organisational mechanisms jointly shape digital technology adoption in agri-food supply chains. By integrating TAM with the relational and governance dimensions of SCM, it demonstrates that adoption is not merely a function of PU and PEU, but also of the social environment and institutional support surrounding farmers.

Effective digital transitions require ecosystems of collaboration in which technology, incentives, relational capital and organisational structures co-evolve with intermediary institutions such as Pos acting as active governance variables mediating behavioural heterogeneity across the chain.

The authors are grateful to UNIEAT and Confagricoltura for their valuable support throughout the entire research process and for their assistance during the data collection.

Maria Teresa Cappella: Methodology, Validation, Investigation, Formal analysis, Data Curation, Writing - Original Draft, Writing - Review & Editing; Eleonora Sofia Rossi: Conceptualization, Validation, Investigation, Writing - Original Draft, Visualization; Emanuele Blasi: Conceptualization, Validation, Investigation, Writing - Review & Editing, Supervision, Project administration.

Data will be made available upon request.