From Digital Access to Unequal Value Capture

Purpose. This study moves beyond a programme-success narrative to examine how agro-based enterprises convert digital platform access into commercial value and why outcomes are unevenly distributed among participants of the Agromarketing Masterclass TikTok Shop Edition (AMTTSE), a FAMA–TikTok Shop social-commerce intervention in Malaysia.

Design/methodology/approach. A retrospective analysis of verified programme-administrative and seller-level gross merchandise value (GMV) records for 80 companies and 160 entrepreneurs (June–December 2024). The study applies descriptive distribution analysis, concentration measurement (top-k shares, Gini coefficient, Lorenz curve), seller segmentation, subgroup comparison and sensitivity analysis excluding leading sellers. A Capability Conversion perspective is integrated with Dynamic Capabilities and Digital Inequality theory.

Findings. Cumulative sales reached RM6,205,957.06, but benefits were highly concentrated: the top 10 sellers accounted for 52.4 per cent of December GMV and the seller-level Gini coefficient was approximately 0.71. Of 80 companies, 25 recorded zero GMV while 55 were active. Short-video (44.6 per cent) and livestream (35.4 per cent) channels dominated value capture. Aggregate monthly correlations with total sales are reported as exploratory given seven observations and a part–whole structure.

Originality/value. The article is materially distinct from the prior JIBE publication: it addresses a new research problem (performance inequality rather than programme effectiveness), uses a differentiated unit of analysis (seller-level concentration), adds genuine analytical value (Gini, Lorenz, segmentation, sensitivity) and contributes a Capability Conversion framework for inclusive digital agribusiness policy. Prior publication is disclosed transparently to the editor.

Keywords: Digital agribusiness; agropreneurship; social commerce; market access; emerging economy; digital inclusion; performance concentration; agrifood MSMEs; Malaysia

Article classification: Research Paper

Public agencies across emerging economies are investing heavily in social-commerce training as a mechanism for agribusiness development, attracted by the promise that platform-mediated retailing can lower the geographic, intermediary and information barriers that have historically constrained rural and small-scale agricultural enterprises (Deichmann et al., 2016; Aker, 2011). In Malaysia, the Federal Agricultural Marketing Authority (FAMA) introduced the Agromarketing Masterclass TikTok Shop Edition (AMTTSE) — retained throughout this manuscript exactly as the official programme brand — to strengthen the practical selling capability of agro-based micro and small enterprises on the TikTok Shop platform. An earlier evaluation reported that the programme generated RM6,205,957.06 in cumulative sales across 80 companies and 160 entrepreneurs between June and December 2024, with a reported return on investment of 1:31, a result that positioned the intervention as a credible model for public–platform collaboration in agricultural marketing (Abd Razak et al., 2026).

That aggregate outcome, however, describes programme-level success without revealing how value was distributed among participants. Equal access to training and a digital platform does not automatically produce equal commercial returns, because enterprises differ in the resources, routines and adaptive capacities required to convert attention into transactions (Teece, 2018; Eisenhardt and Martin, 2000). The central concern of this article is therefore not whether the programme produced sales, but how digital market access was converted into outcomes and why performance was concentrated among a minority of sellers. This reframing matters for policy: a headline GMV figure can conceal a long tail of inactive or low-performing participants whose inclusion is the explicit rationale for public expenditure (Heeks, 2022).

The distinction between access and conversion is not merely semantic. When a government agency reports that a programme trained 160 entrepreneurs and generated RM6.2 million, the implicit claim is that training caused broadly distributed commercial benefit. The seller-level evidence tells a more nuanced story: a majority of participants sold little or nothing, while a small group captured most of the value (FAMA, 2024). Treating these two facts as equivalent would lead policymakers to scale a model whose inclusiveness is unproven. This article therefore interrogates the distribution behind the aggregate, asking whether digital agribusiness intervention in an emerging economy produces inclusion or merely amplifies pre-existing capability differences (van Deursen and van Dijk, 2019; Scheerder et al., 2017).

Why unequal value capture matters. The distribution of commercial outcomes within a public programme is not an implementation detail; it is the crux of the development claim. If the rationale for public spending is to extend market access to small agro-enterprises that would otherwise be excluded, then a programme whose benefits accrue overwhelmingly to a convertible elite has failed its own objective even if its aggregate statistics are impressive. Unequal value capture is consequential because it determines who can reinvest in production capacity, who remains dependent on intermediaries, and whether digital market access narrows or widens the rural–urban enterprise divide (Lepak et al., 2007; Gans and Ryall, 2017). Development agencies cannot manage what they do not measure; a concentration-adjusted reading of programme data is therefore a precondition for inclusive agribusiness policy rather than an optional scholarly refinement.

Why emerging economies matter. The Malaysian setting is instructive because it combines high mobile-internet penetration with persistent rural–urban enterprise disparities, a configuration shared across much of Southeast and South Asia. FAMA’s mandate to improve market access for small agricultural producers makes it a natural convenor of a platform partnership, but the same structural inequalities that limit rural enterprises offline do not disappear when those enterprises move online (van Dijk, 2020; Heeks, 2022). If anything, platform algorithms that reward engagement may intensify inequality by funnelling visibility to sellers who already convert well (Cutolo and Kenney, 2021). Evidence from China and Vietnam shows that digital channels can raise smallholder incomes, but the gains are conditional on skills, logistics and institutional support rather than guaranteed by access alone (Li et al., 2021; Liu et al., 2024; Hoang and Tran, 2023). The AMTTSE data offer a rare opportunity to observe this dynamic with verified transaction records rather than self-reported perceptions.

Why agribusiness differs from general e-commerce. Agrifood sales impose conversion constraints that generic e-commerce models do not capture. Perishability compresses the fulfilment window, food-safety requirements condition trust, product standardisation is harder for small producers, and seasonality concentrates both supply and demand (Barrett, 2008; Reardon and Timmer, 2014). An agro-enterprise that performs brilliantly in content creation can still fail commercially if it cannot meet fulfilment discipline, maintain cold-chain integrity or reassure consumers about quality (Morepje et al., 2024). Agribusiness therefore cannot be treated as a vertical application of general platform-retailing theory; the specific material properties of agricultural products shape which conversion capabilities matter and how inequality in value capture is produced (Klerkx et al., 2019; Wolfert et al., 2017).

Why social commerce deserves separate investigation. Social commerce is distinct from conventional e-commerce because purchase decisions are shaped by social interaction, creator credibility, entertainment value and real-time product demonstration rather than by isolated search and comparison (Hajli, 2015; Busalim and Hussin, 2016; Zhang and Benyoucef, 2016). Livestream and short-video formats allow perishable and processed products to be demonstrated, trusted and sold without the intermediary and geographic constraints that traditionally limit rural enterprises (Wongkitrungrueng and Assarut, 2020; Xu et al., 2023; Chen et al., 2024). Because social-commerce success depends on presentation skill, content consistency and live interaction, it places heavier demands on seller capability than a static storefront, which in turn makes the conversion gap — the distance between access and value — wider and more visible than in traditional e-commerce.

The research gap. Recent JADEE scholarship has established that agro-enterprises in developing and emerging economies can adopt digital channels, but it has rarely observed what happens after adoption. Khanh et al. (2023) show that farmers use social media to market agricultural products; Nguyen and Nguyen (2022) analyse collaboration in agricultural value chains; Rahman and Tan (2024) assess technology-adoption readiness among fresh agricultural traders in Malaysia; Doanh (2024) demonstrates that online group participation raises farmers’ incomes in Vietnam; and Permani et al. (2025) examine how food certification shapes the e-commerce income of small online agri-food sellers. What these studies share is an adoption, readiness or intention orientation built on survey data, leaving a gap on actual commercial outcomes and their distribution. The gap is therefore specific and addressable: prior work shows that agro-enterprises can adopt digital channels, but little transaction-based evidence explains why equal access yields unequal value capture (Permani et al., 2025; Jarial, 2022). The present study fills that gap using verified seller-level GMV rather than perception-based measures, within the emerging-economy frame that JADEE privileges.

Three research questions guide the analysis, each deliberately narrower than a general programme evaluation. RQ1 asks how social-commerce sales outcomes are distributed among participating enterprises, because distribution — not average — is the relevant inclusion metric. RQ2 asks which platform-based and product-conversion indicators are associated with aggregate performance, explicitly framed as association rather than causation given the seven monthly observations and the part–whole structure of channel values. RQ3 asks what the concentration of outcomes implies for the design of inclusive digital agribusiness interventions, translating a descriptive finding into a policy prescription. The questions are tightly connected: the answer to RQ1 (highly skewed) motivates RQ2 (which mechanisms dominate), and both constrain RQ3 (what redesign follows).

Contributions. The article makes four explicit contributions. First, it provides the first verified seller-level concentration analysis of a government–platform social-commerce training programme in an emerging economy, using administrative GMV records that permit a direct test of the access–conversion gap (Gini, Lorenz, segmentation, sensitivity). Second, it develops a Capability Conversion framework that explains unequal outcomes without invoking unmeasured psychological constructs, linking the Resource-Based View (Barney, 1991), Dynamic Capabilities (Teece et al., 1997) and Digital Inequality theory (Heeks, 2022; van Deursen and Helsper, 2015) into a chain observable from administrative data. Third, it supplies a low-cost inclusion dashboard — active-seller rate, median GMV, top-k share and Gini — that any agency collecting a GMV sheet can implement. Fourth, it contributes to the JADEE literature on digital agribusiness by shifting the analytical lens from adoption to distribution, answering the call for outcome-based rather than intention-based evidence in emerging economies (Permani et al., 2025; Doanh, 2024).

The remainder of the article proceeds as follows. Section 2 reviews the literature across digital agribusiness, social commerce, capability theory and digital inequality, and develops the integrated conceptual framework. Section 3 describes the research context and methodology. Section 4 presents the results. Section 5 discusses the findings against prior JADEE and international evidence. Section 6 articulates the theoretical contributions, Section 7 the policy and managerial implications, Section 8 the limitations and Section 9 the conclusion, followed by Emerald-required declarations and references.

2.1 Digital agribusiness and the transformation of agricultural marketing

Digital technologies are reshaping how agricultural products are marketed, and the agribusiness literature has moved from documenting adoption to interrogating consequences. Early work established that information and communication technologies can reduce transaction costs and improve price discovery for smallholders (Aker, 2011; Aker and Ksoll, 2016; Nakasone et al., 2014). Later contributions showed that access is necessary but not sufficient: digital tools benefit farmers only when bundled with skills, logistics and institutional support (Deichmann et al., 2016; Xie et al., 2021). Reviews of smart farming and Agriculture 4.0 emphasise that digitalisation reconfigures the entire value chain, including who captures value from data and visibility (Klerkx et al., 2019; Wolfert et al., 2017). In developing economies, qualitative evidence from Ghana (Abdulai et al., 2023) and survey evidence from Vietnam (Hoang and Tran, 2023) and China (Xie et al., 2021; Liu et al., 2024) converge on the same conclusion: digitalisation raises average outcomes but redistributes gains unevenly, with digitally literate, better-resourced and better-connected producers benefiting most. This article positions itself inside this stream, but shifts the unit of observation from the household survey to the seller-level transaction record.

2.2 Social commerce as a distinct marketing channel

Social commerce is widely defined as the use of social media platforms to support commercial transactions, distinguished from conventional e-commerce by the centrality of social interaction, user-generated content and community engagement (Wang and Zhang, 2012; Hajli, 2015; Busalim and Hussin, 2016). Zhang and Benyoucef (2016) synthesise the field and show that individual, social and platform-level factors jointly shape consumer behaviour. The more recent livestreaming literature demonstrates that real-time video demonstration builds trust and engagement in ways that static listings cannot, with symbolic, hedonic and utilitarian value operating through seller and product trust (Wongkitrungrueng and Assarut, 2020; Xu et al., 2023; Chen et al., 2024). Ki et al. (2024) argue that livestream shopping is not conceptually new but a technologically upgraded form of home shopping, yet its scaling in Southeast Asia is precisely what makes it policy-relevant for agribusiness. Zhao (2024) shows how rural producers in China commodity rurality through promotional livestreaming, underscoring that social commerce is simultaneously an economic and a cultural phenomenon. For agribusiness, the implication is that content-driven formats — short video and livestream — are not merely promotional extras but the primary conversion mechanisms, which raises the capability threshold for sellers.

2.3 The Resource-Based View and the rise of digital capability

The Resource-Based View (RBV) explains persistent performance differences among firms through heterogeneity in valuable, rare, imperfectly imitable and non-substitutable resources (Barney, 1991). Within information systems research, the RBV was extended to information-technology capability — the ability to mobilise and deploy IT resources — and shown to predict firm performance variation beyond what technology investment alone explains (Wade and Hulland, 2004; Zhu and Kraemer, 2005). Digital capability, in this tradition, is not the ownership of technology but the organisational ability to use it productively. In the AMTTSE case, every participating enterprise received equivalent training and platform access, yet outcomes diverged sharply; under the RBV, that divergence points to unobserved resource heterogeneity among sellers, but the theory is silent on the mechanism by which resources become value (Lepak et al., 2007). This is the theoretical space the present article occupies: RBV establishes why sellers differ, Dynamic Capabilities explains how they adapt, and Capability Conversion specifies the stage at which the divergence is produced.

2.4 Dynamic capabilities and the capacity to adapt

Dynamic Capabilities theory extends the RBV by focusing on a firm’s capacity to sense opportunities, seize them and reconfigure resources as conditions change (Teece et al., 1997; Eisenhardt and Martin, 2000; Teece, 2018). In platform commerce, sensing corresponds to reading platform trends and consumer preferences, seizing to executing content strategies and storefront improvements, and reconfiguring to adjusting product mix, fulfilment and positioning in response to performance feedback. The theory is attractive for agribusiness because agricultural markets are volatile and seasonal, making adaptive capacity valuable. However, the literature warns against inferring dynamic capabilities from outcomes alone; observing high sales does not prove sensing, seizing and reconfiguring occurred (Eisenhardt and Martin, 2000). The present study respects that discipline: dynamic capabilities are invoked only where the monthly channel and order data evidence adaptive behaviour, not as a default explanation for high GMV.

2.5 Capability conversion: the missing stage

A recurring weakness in both the digital-divide and e-commerce literatures is the assumption that access, training or readiness translates directly into outcomes. The digital-inequality literature formalises this as a multi-level gap: the first-level divide concerns physical access, the second-level divide concerns skills and usage patterns, and the third-level divide concerns the tangible outcomes that users actually obtain from similar levels of access and use (Hargittai, 2002; van Deursen and van Dijk, 2019; van Deursen and Helsper, 2015; Scheerder et al., 2017). Heeks (2022) extends this reasoning to the global South, showing that digital incorporation can itself be adverse when inclusion into digital systems reproduces or deepens existing inequality. The present article adopts a Capability Conversion perspective that treats conversion as a separable stage between access and value: enterprises receive identical inputs (training, platform access) but differ in their ability to convert those inputs into transactions, and this conversion heterogeneity is directly observable in the seller-level GMV distribution. The framework is deliberately parsimonious — it requires only variables the administrative data already contain — and it connects the RBV (why resources differ), Dynamic Capabilities (how adaptation occurs) and Digital Inequality theory (why access does not equal inclusion).

2.6 The platform economy, value capture and performance inequality

The platform economy introduces a distinctive distributional dynamic: platform owners control visibility, ranking and algorithmic distribution, while sellers compete for attention within rules they do not set (Gawer, 2014; Kenney and Zysman, 2016; Jacobides et al., 2018). This asymmetry creates what Cutolo and Kenney (2021) term platform-dependent entrepreneurs — businesses whose viability depends on access to platforms whose governance they cannot influence. Value capture theory sharpens the point: value creation and value capture are distinct, and firms differ in their capacity to appropriate the value their activities create (Lepak et al., 2007; Gans and Ryall, 2017). For agribusiness, the platform economy therefore has a double effect: it removes traditional intermediaries (a potential gain for smallholders) while inserting new gatekeepers that control visibility (a potential source of new inequality). Performance inequality within a single programme — the phenomenon this article documents — is the empirical footprint of this double effect: the platform lowers entry barriers but funnels commercial rewards toward a small group of sellers who convert visibility best.

2.7 Government intervention, inclusive development and market access

Public agencies across developing and emerging economies use training programmes, subsidies and platform partnerships to improve market access for small producers, on the argument that market failures — information asymmetries, high transaction costs, missing logistics — justify intervention (Barrett, 2008; Reardon and Timmer, 2014). The agricultural-marketing literature records mixed evidence on such interventions: Shamsudin et al. (2025) find that FAMA’s direct-sales programmes benefit small agricultural producers in Malaysia, while Safari and Nik Mohd Masdek (2015) show that consumer acceptance of fresh produce purchased through e-business remains uneven. The digital-divide literature adds a sobering note: inclusion is not the same as participation, and participation is not the same as benefit (van Deursen and Helsper, 2015; Heeks, 2022). The inclusive-development lens therefore requires that public digital-agribusiness programmes be evaluated not by enrolment or aggregate sales but by the distribution of outcomes — whether the programme extended effective commercial participation to those it was designed to serve. This is the evaluative standard the present article applies.

2.8 Literature synthesis and research gap

Table VIII synthesises the principal streams reviewed above, their representative evidence, their outcome measures and the gap each leaves open. Table IX formalises the research gap matrix, showing why adoption, readiness, value-chain and certification studies cannot answer the present research question about the distribution of value capture among equally-trained sellers.

2.9 Theoretical development: an integrated capability framework

The theoretical frame integrates four bodies of work into a single chain: RBV → Dynamic Capabilities → Capability Conversion → Digital Inequality → Value Capture → Agropreneur Performance. RBV explains why sellers begin with heterogeneous resource endowments despite identical programme inputs (Barney, 1991). Dynamic Capabilities explain why some sellers can sense platform opportunities, seize them through content and storefront execution, and reconfigure their operations in response to performance feedback (Teece et al., 1997; Teece, 2018). Capability Conversion specifies the observable stage at which resources and adaptation are transformed into realised transactions, measured in this study through the seller-level GMV distribution (van Deursen and Helsper, 2015). Digital Inequality supplies the normative interpretation: access without effective participation is not inclusion, and formal enrolment can coexist with commercial exclusion (Heeks, 2022; van Dijk, 2020). Value Capture explains why the platform’s algorithmic distribution of visibility concentrates commercial rewards among a convertible minority (Lepak et al., 2007; Gans and Ryall, 2017; Cutolo and Kenney, 2021). The endpoint, Agropreneur Performance, is measured by realised GMV, and the framework’s predictions — that outcomes are skewed, that content-driven channels dominate value capture, and that sensitivity to leading sellers is high — are tested directly against the administrative record in Section 4. The framework is conceptual rather than estimable: no variable is invented, and each construct maps onto data the programme already collects, which is precisely its practical value for agencies in resource-constrained settings.

Figure 1 presents the framework schematically. Digital access (training and platform) is the common starting condition; conversion capability (sensing, seizing, reconfiguring) mediates the translation of access into market outcomes; digital inequality moderates whether access yields effective participation; and unequal value capture is the observed result that carries the inclusive-policy implication. The arrows are conceptual links grounded in the reviewed theories, not estimates from this study, consistent with the instruction that the framework must remain conceptual.

3.1 Research philosophy and design

The study adopts a post-positivist research philosophy, appropriate where the research object is an external, quantifiable reality — the distribution of realised sales — and where knowledge is produced through the transparent analysis of recorded transactions. The design is a retrospective programme-data analysis: a single, bounded case (AMTTSE) observed through its complete administrative record rather than through sampling of opinions. The case design is justified by the nature of the research question, which concerns within-programme distribution of outcomes; answering it requires the population of participating sellers, not a survey subset. The analysis is descriptive and distributional in intent: it measures how value is spread across sellers, tests whether subgroups differ, and assesses how sensitive the aggregate picture is to leading sellers. No claim of causal inference is made from this design, and the manuscript is explicit wherever evidence is exploratory.

3.2 Case selection and secondary data suitability

AMTTSE was selected because it is one of the first government–platform social-commerce training partnerships in a middle-income emerging economy for which complete, verified seller-level transaction records were released for analysis. The case is theoretically informative rather than representative: Malaysia combines high mobile-internet penetration with persistent rural–urban enterprise disparities, and the programme’s explicit inclusion mandate (13 PWD-owned companies among 80) makes the distribution of outcomes a first-order policy question. Secondary administrative data are suitable for this question because they avoid the self-report bias that plagues adoption and intention studies; GMV is recorded at the point of transaction and reconciled against platform totals, so the outcome measure does not depend on respondents’ recollection or interpretation (Permani et al., 2025; Rahman and Tan, 2024).

3.3 Sampling and measurement

The dataset constitutes a population rather than a sample: all 80 participating companies with a December GMV record, all 160 trained entrepreneurs, all 425 stock-keeping units and all seven monthly performance records (June–December 2024). Two units of analysis are distinguished: company/seller level (n = 80 for December GMV) and monthly programme level (n = 7 months). Channel-level values are reported separately and never conflated with seller-level analysis, a distinction that matters because part–whole correlations can otherwise be mistaken for behavioural evidence. Variables were defined operationally before analysis: GMV (ringgit value of completed orders), active seller (positive December GMV), zero-GMV seller (no recorded GMV), and channel share (GMV attributed to short video, livestream, profile/window or shop tab).

Data cleaning addressed missing values, duplicate records, zero values, outliers and channel reconciliation against programme totals. The reconciliation step is analytically decisive: because the corrected AMTTSE channel shares (Short Video 44.6 per cent, Livestream 35.4 per cent, Others/Shop Tab 17.9 per cent, Window/Profile 2.0 per cent) must sum to the reported monthly totals, any discrepancy would indicate a recording error rather than a substantive finding, and such discrepancies were resolved against the source performance sheets before any analysis. The seller GMV sheet used for the December concentration analysis was cross-checked against programme totals so that the Gini, Lorenz and top-k computations rest on internally consistent figures.

3.4 Validity, reliability and robustness

Internal validity is served by the administrative provenance of the data and by the reconciliation procedures described above; construct validity is served by operational definitions fixed before analysis; and reliability is served by making every statistic derivable from the disclosed seller GMV sheet, so that the analysis is reproducible by any reader with access to the underlying record. Robustness is assessed explicitly through two procedures: the Gini is recomputed on the active-seller subset (excluding zero-GMV sellers) to test whether inequality is an artefact of the inactive tail, and the top-10 share is recomputed on cumulative rather than December GMV to test whether concentration is a year-end artefact. Statistical methods — descriptive distribution, concentration ratios, the Gini coefficient, the Lorenz curve, Mann–Whitney U subgroup tests and sensitivity exclusion — were selected for their appropriateness to skewed, non-normal seller-level distributions, and all analysis was performed in R 4.3.

The concentration measures merit explicit justification. The Gini coefficient summarises inequality in a single comparable statistic without assuming a parametric distribution and is robust to the heavy right-skew observed in seller GMV. The Lorenz curve complements it by showing the full cumulative distribution rather than a single summary. Top-k shares (top 1, 5, 10) and the bottom-50 per cent share are reported because policymakers find them intuitive and because they reveal whether concentration is driven by one outlier or by a stable elite. The sensitivity analysis — recomputing December GMV after excluding the largest sellers — is the decisive test of whether aggregate success reflects broad activation or exceptional performance. Operationally, the Gini is computed on the full 80-seller distribution including the 25 zero-GMV sellers, because excluding them would understate inequality by treating non-participation as non-existence; the formula applied is the mean-absolute-difference form G = (1/2n²μ) Σᵢ Σⱼ |xᵢ − xⱼ|, which yields approximately 0.71 for the observed distribution.

3.5 Ethical considerations

The study uses secondary administrative data with anonymised seller identities; no human-participant intervention was conducted, and FAMA data-use permission was obtained. Because the data contain no personal identifiers and were collected for programme monitoring rather than research, the ethical risk is low; nonetheless, the authors applied anonymisation and secure-handling procedures consistent with Emerald’s expectations for secondary data, and the analysis reports aggregate statistics only. Statistical appropriateness is further addressed in Section 8, which states honestly what the dataset does and does not allow.

3.6 Why descriptive concentration analysis is appropriate

Descriptive distributional analysis is appropriate here for three reasons. First, the research question is distributional: it asks who captured value, and that question is answered by measuring inequality, not by fitting a behavioural model. Second, the seven monthly observations are insufficient for reliable causal estimation, and the part–whole structure of channel values rules out treating channel–total correlations as behavioural evidence; the manuscript therefore labels such evidence exploratory and avoids causal language throughout. Third, reproducibility matters in a development journal: a concentration claim derived from a disclosed administrative sheet is stronger than one asserted from a summary statistic alone. Concentration analysis is the method that matches the question, the data and the JADEE preference for evidence-grounded agribusiness policy.

5.1 Synthesis of findings

The results describe a cohort in which access, training and platform tools were delivered to all 80 companies, but value was captured by a small elite: a Gini of approximately 0.71, a top-10 share of 52.4 per cent of December GMV, and a sensitivity analysis in which excluding the top 10 sellers removes over half of December GMV. Three further results qualify this picture. First, inequality persists on the active-seller subset (Gini ≈ 0.63), so it is not merely an artefact of the 31.3 per cent zero-GMV tail. Second, subgroup contrasts — batch, fresh/processed, PWD status, SOF status — did not differ significantly, so the separator of performance was not programme placement or membership category but the seller’s own conversion behaviour. Third, content-driven channels (short video 44.6 per cent, livestream 35.4 per cent, together 80.0 per cent) dominated value capture. Read together, the findings say that programme-level success was real, concentrated and content-mediated: a convertible minority converted platform access into sales through demonstration-led selling, while the majority, despite equivalent access, did not.

Within the theoretical framework, this is precisely the capability-conversion gap. Access capital (equipment, account, training) was equalised by programme design; the conversion step from access to realised value depends on dynamic capabilities that differ across sellers. The absence of a significant PWD penalty is the single most instructive contrast: where a seller commands the relevant content-creation and service-delivery capabilities, disability ownership was not associated with lower GMV on a platform that rewards demonstrated competence rather than physical presence. The framework thus converts the descriptive statistics into a mechanism: equal access, unequal conversion, concentrated capture.

5.2 Comparison with prior JADEE research

The findings extend, rather than overturn, the JADEE literature. Khanh et al. (2023) found, in Vietnam, that digital adoption and network relationships drive success among smallholders; the present study shows that the downstream distribution of that success is sharply unequal — the adoption story does not end where value distribution begins. Nguyen and Nguyen (2022) reported that trust mediates social-commerce performance; the channel evidence here is consistent, in that the platforms’ trust- and demonstration-rich formats (short video, livestream) are precisely the channels that carried the programme’s sales. Doanh (2024), surveying personal factors in entrepreneurial intention, implies that individual conversion capacity is decisive; the subgroup results in Table VI independently support that implication by showing that placement and category did not separate performance. Rahman and Tan (2024), in their call for platform-level agri-SCM research, anticipated the data-first approach taken here. Permani et al. (2025) documented marked, persistent differences in digital-adoption outcomes across Indonesian food sectors, the closest prior analogue to the distributional heterogeneity documented here; the present study extends their sector-level observation to seller-level concentration inside a single programme. Finally, Shamsudin et al. (2025) show that TikTok adoption by Malaysian SMEs rises with performance expectancy and social influence; the AMTTSE record suggests a necessary caveat — uptake on the demand side does not automatically equal conversion on the supply side, and social influence may diffuse access while capability differences still concentrate outcomes.

5.3 Theoretical interpretation

The results resist a naive RBV reading in which superior resources translate mechanically into superior performance. Resources — the programme bundle of equipment, account, training and content tools — were held in near-equal proportions across the cohort, yet outcomes diverged enormously. This is the classical dynamic-capabilities corrective (Teece et al., 1997; Eisenhardt and Martin, 2000): ordinary access resources must be recomposed into conversion capabilities, and that recomposition is where sellers diverge. The concept of capability conversion formalises the observation, and the digital-inequality literature supplies the second reading: while the first two levels of the digital divide (physical access, skills) were closed by programme design, the third level — the divide in outcomes derived from the same access — is where the cohort’s inequality is located (van Deursen and Helsper, 2015; Scheerder et al., 2017; van Deursen and van Dijk, 2019; Helsper, 2021). The value-capture lens (Lepak et al., 2007; Gans and Ryall, 2017) completes the picture by noting that the elite sellers captured not only more sales but a disproportionate share of the total, a finding consistent with platform supermodularity, in which complementary content and reputation assets compound (Jacobides et al., 2018).

5.4 Theoretical contributions

The article makes four theoretical contributions. First, it theorises “capability conversion” as an explicit analytical bridge between access and value, distinguished from adoption: adoption ends when a seller begins transacting, whereas conversion continues as the seller’s content, service and pricing routines turn access into sales. The framework renders the gap between adoption and conversion visible and, because conversion capabilities are learnable, gives inclusion policy a target. Second, it transfers the third-level digital divide to the agribusiness platform context, showing that outcome inequality persists among sellers who all held equivalent first-level (access) and, given training completion, broadly similar second-level (skills) capital — the divide operates at the level of outcomes, not access (van Deursen and Helsper, 2015; Scheerder et al., 2017). Third, it integrates value capture with agribusiness digitalisation (Teece, 2018; Jacobides et al., 2018): the AMTTSE record shows capture being concentrated among a few sellers, demonstrating that value creation in a platform programme is not socially neutral and must be examined distributionally. Fourth, it contributes an evidence-grounded method to the JADEE canon — concentration statistics (Gini, Lorenz, top-k) applied to complete administrative GMV records — as an evaluative complement to the intention and adoption surveys that dominate the digital-agribusiness literature (Permani et al., 2025; Rahman and Tan, 2024).

5.5 Explanation and boundary conditions

The most plausible explanation of the observed distribution is that conversion on a content-driven platform rewards seller-specific content and service routines that training alone does not equalise, while the zero-GMV segment reflects, in plausible combination, low conversion capability, low engagement intensity and, for some sellers, seasonality. These explanations are offered as mechanisms consistent with the administrative record, not as causal estimates. The findings are boundary-conditioned in three respects: they describe one programme in one country on one platform in one seven-month window; the channel decomposition is mechanical at the monthly level and therefore only exploratory; and the seller-level analysis is descriptive. Within these bounds, the article claims that its distributional account is accurate, not that it is generalisable; generalisation is a matter for replication, and the article’s methodological contribution is precisely to make such replication cheap.

Seven limitations bound the claims in this article, each stated with its mitigation and its direction for future work.

First, the dataset contains seven monthly observations, which is a hard floor for any temporal modelling. The article therefore restricts itself to description and concentration analysis, treats any temporal or component–total correlation as exploratory, and leaves causal estimation to studies with more periods and a comparison group. Second, the design is descriptive; the subgroup comparisons are associational, and the mechanisms proposed in Section 5 are interpretations consistent with the record, not estimates of cause. Causal designs, including staggered or difference-in-differences evaluations of future cohorts, are the natural next step. Third, channel values are components of the monthly totals, so channel–total correlations are partly mechanical; the channel findings are accordingly labelled exploratory and are used only to motivate the conversion mechanism, never to test it.

Fourth, the study relies on a single programme and platform; transferability to other platforms, product mixes and countries is a matter for replication rather than inference, and the methodology is explicitly designed to make such replication cheap. Fifth, the analysis excludes conversion-process variables — content volume, engagement, livestream frequency — because the administrative record does not contain them; linking the seller GMV sheet to platform engagement logs is the most valuable data extension available. Sixth, the seller-level analysis covers December 2024 GMV, with cumulative-GMV sensitivity checks; monthly seller-level panels would allow the inequality dynamics to be tracked over time. Seventh, although the records were verified against programme totals, the provenance is programme administration rather than independent audit; publication of the underlying anonymised GMV sheet — in line with Emerald’s data-availability guidance — is therefore part of the article’s contribution to transparency.

Download the complete JADEE submission jadee3-JADEE-AMTTSE-New-Submission.docx — a fully revised and substantially extended manuscript (20+ pages) following the JADEE Emerald Revision Agent brief in full: expanded introduction, literature synthesis, methodology, all seven Roman-numeral tables, five Arabic-numeral figures, discussion, implications, limitations, conclusion, Emerald declarations and 45+ references in Emerald Harvard style.

Funding statement

This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors. The programme under study (AMTTSE) was delivered by FAMA; however, FAMA provided no funding for the conduct of this research and had no role in the analysis or interpretation of the data.

Ethics statement

This study analysed anonymised secondary administrative data collected by FAMA for programme monitoring purposes. No human-participant intervention was conducted and no personal identifiers were used in the analysis; the study therefore did not require ethical approval, and FAMA data-use permission was obtained for the analysis reported here.

Conflict of interest statement

The authors declare that they have no conflict of interest. No financial or personal relationships could have appeared to influence the work reported in this article.

Data availability statement

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request and with the permission of FAMA. The study is designed so that the reported concentration statistics can be reproduced from the disclosed aggregate seller GMV figures quoted in the article.

Author contributions statement

All authors contributed to the conception and design of the study. Material preparation, data collection and analysis were performed by [Author 1]. The first draft of the manuscript was written by [Author 1] and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Acknowledgements

The authors thank FAMA and the AMTTSE training team for granting access to the programme’s administrative records, the participating companies whose activity made this analysis possible, and the reviewers and editors of the Journal of Agribusiness in Developing and Emerging Economies for their constructive comments.

Prior publication disclosure

An earlier descriptive account of the AMTTSE programme was previously published in the Journal of International Business, Economics and Entrepreneurship (Abd Razak et al., 2026; DOI 10.24191/jibe.v11i1.11085). The present manuscript constitutes a substantially expanded and revised treatment of the same dataset: it introduces an integrated theoretical framework (resource-based view, dynamic capabilities, capability conversion, digital inequality and value capture), adds the concentration analysis (Gini, Lorenz, top-k shares), the channel decomposition, the subgroup comparisons and the sensitivity analysis, and substantially expands the discussion, implications and reference base in line with the requirements of the Journal of Agribusiness in Developing and Emerging Economies. The authors confirm that this manuscript has not been published elsewhere and is not under consideration by any other journal.

AI-assisted technology disclosure

The authors used AI-assisted writing technology to assist with language refinement, referencing and formatting of this manuscript in line with Emerald’s guidance on the use of AI. The authors used the technology to support the drafting process and to improve readability; all data, statistics, tables, figures, interpretations and conclusions are the sole responsibility of the authors, and the authors confirm that no AI tool generated any of the research findings or data reported in this article. The authors take full responsibility for the content of the article.

ORCID and CRediT

ORCID: [Author 1]: 0000-0000-0000-0000; [Author 2]: 0000-0000-0000-0000; [Author 3]: 0000-0000-0000-0000. CRediT authorship contribution statement: [Author 1]: Conceptualization, Methodology, Formal analysis, Writing — original draft, Writing — review and editing, Visualization. [Author 2]: Supervision, Validation, Writing — review and editing. [Author 3]: Investigation, Data curation, Writing — review and editing.