Enhancing Supply Chain Resilience in Textile SMEs: A Human-Centric Customer-to-Manufacturer Framework Using Public E-Commerce Data

Upstream textile small and medium-sized enterprises (SMEs) frequently exhibit constrained supply chain resilience owing to persistent information latency and structural dependence on downstream orders. To address these challenges, this study develops and validates a customer-to-manufacturer (C2M) intelligence framework that enables data-driven production planning using publicly available e-commerce data. The framework incorporates ethically compliant acquisition of consumer demand signals, semantic translation of unstructured market data into textile engineering attributes, machine-learning-based demand forecasting, and human-centric decision support. Utilizing 3.87 million consumer comments from 127,846 product listings, a Neural Boosted Tree model with entity embeddings for textile attributes was constructed. This model achieved a mean R2 of 0.921 in cross-validation, surpassing benchmark methods. Consumer comment volume was validated as a proxy for sales activity, facilitating demand estimation. Forecasts were translated into production guidance using Monte Carlo simulation and a decision dashboard. In a 12-month field study at a Taiwanese dyeing SME, implementation resulted in a 28% reduction in inventory value, a 31% decrease in dye lot changeovers, and a 16% increase in capacity utilization. This research extends the C2M paradigm from downstream retail contexts to upstream textile SMEs, proposes an integrated and operationally feasible intelligence framework for resource-constrained manufacturers, and demonstrates how digital intelligence can enhance supply chain resilience while supporting, rather than replacing, human decision-making. The results indicate that upstream textile SMEs can leverage publicly visible e-commerce signals to enhance production planning responsiveness, minimize inventory exposure and dye-lot disruptions, and strengthen resilience to demand uncertainty through planner-centered digital decision support.

Summary

Main Finding

An ethically sourced, data-driven customer-to-manufacturer (C2M) intelligence framework using publicly available e‑commerce signals can materially improve production planning and supply-chain resilience for upstream textile SMEs. In a 12‑month field deployment at a Taiwanese dyeing SME, the framework reduced inventory value by 28%, cut dye‑lot changeovers by 31%, and raised capacity utilization by 16%. A Neural Boosted Tree model with entity embeddings for textile attributes produced highly accurate demand forecasts (mean R2 = 0.921), enabling planner-centered decision support via Monte Carlo–based production guidance.

Key Points

Problem: Upstream textile SMEs face low supply‑chain resilience due to information latency and dependence on downstream orders.
Intervention: A C2M intelligence pipeline that (a) ethically collects public e‑commerce consumer signals, (b) semantically maps unstructured market text to textile engineering attributes, (c) forecasts demand with machine learning, and (d) provides human‑centric decision support (dashboard + Monte Carlo simulations).
Data scale: 3.87 million consumer comments across 127,846 product listings were used.
Modeling: A Neural Boosted Tree model that incorporated entity embeddings for textile attributes outperformed benchmarks in cross‑validation (mean R2 = 0.921).
Proxy validation: Consumer comment volume was validated as a usable proxy for sales activity, enabling demand estimation where sales data are unavailable.
Operationalization: Forecasts were converted into production guidance using Monte Carlo simulation and presented through a planner-facing dashboard to support—not replace—human decisions.
Field outcomes (12 months, single SME): −28% inventory value, −31% dye‑lot changeovers, +16% capacity utilization.
Contribution: Extends C2M concepts from retail to upstream manufacturing and demonstrates an integrated, feasible intelligence approach for resource‑constrained firms.

Data & Methods

Data sources: Publicly visible e‑commerce data — 3.87M consumer comments from 127,846 product listings; ethically compliant data acquisition procedures were followed.
Preprocessing/semantic translation: Unstructured market text (comments, reviews) was semantically translated into structured textile engineering attributes (e.g., fabric type, color, finish) to make market signals actionable for manufacturing decisions.
Forecasting model:
- Architecture: Neural Boosted Tree (gradient‑boosted trees enhanced with neural embeddings), using entity embeddings to represent categorical textile attributes.
- Evaluation: Cross‑validation with mean R2 = 0.921; benchmark methods used for comparison (details not specified in summary).
Demand proxy validation: Empirical tests showed comment volume correlates with sales activity sufficiently to be used as a demand proxy.
Decision support: Forecast outputs fed into Monte Carlo simulation to quantify demand uncertainty and generate production planning scenarios; results presented via a human‑centric dashboard to guide planners.
Field validation: 12‑month deployment at a Taiwanese dyeing SME; operational metrics tracked pre/post implementation to quantify impacts.

Implications for AI Economics

Information frictions and bullwhip reduction: Public e‑commerce signals can reduce information latency upstream, diminishing demand uncertainty and bullwhip effects, which improves allocative efficiency across tiers.
Cost and productivity effects: Measurable reductions in inventory holdings and changeovers and higher capacity utilization imply lower holding and setup costs and higher throughput—potentially raising firm-level profitability and lowering lead times.
Competitive dynamics and diffusion: Firms that adopt such low‑cost external data strategies may gain a short‑term competitive edge; widespread adoption could compress margins but raise sectoral efficiency. Barriers include analytics capability and access to clean, representative public data.
Labor and skill complementarity: The human‑centric design indicates AI augments planner decision-making rather than displacing it—shifting demand toward analytical and decision‑support skills in SMEs.
Privacy, ethics, and governance: Using public consumer signals mitigates direct privacy risks, but ethical data collection practices and platform‑specific biases must be managed. Policymakers and firms should consider standards for transparent, equitable use of platform data.
Policy and industrial policy: Support for analytics adoption (training, subsidized tools) could be a high‑leverage intervention to raise sector resilience among resource‑constrained upstream firms.
Research and robustness needs:
- External validity: Results are from one SME and one country; replication across geographies, product types, and multi‑tier supply chains is needed.
- Causal mechanisms: More rigorous causal identification (e.g., randomized rollouts, synthetic controls) would strengthen claims about effect pathways.
- Data bias and platform dependence: Further work should test robustness to platform composition, seasonal effects, and manipulation of public signals.
- Integration opportunities: Combining public signals with limited private order data, price dynamics, and lead‑time models could enhance forecasts and welfare impacts.
Broader economic effects: If generalized, such approaches can improve small‑firm resilience, reduce upstream volatility, and lower systemic supply‑chain fragility—important as manufacturing networks incorporate more digital demand signals.

Assessment

Paper Typequasi_experimental Evidence Strengthmedium — The study provides strong within-sample predictive performance and large operational improvements after deployment, supporting plausibility of a causal effect; however, causal attribution is limited by a single-site, non-randomized pre-post design, potential confounders (concurrent managerial/process changes, demand shifts), and lack of external or placebo controls. Methods Rigormedium — Methodologically strong components include large-scale data collection, careful semantic feature engineering, an advanced Neural Boosted Tree with entity embeddings, cross-validation, and Monte Carlo translation to operational decisions; weaknesses include lack of randomized evaluation or control firms, limited transparency on hyperparameter tuning/overfitting checks, and sparse reporting of robustness checks and time-series confounder adjustment. SampleModeling dataset: 3.87 million consumer comments from 127,846 e-commerce product listings (publicly available platform comments), with semantic extraction to textile engineering attributes; Forecasting model evaluated via cross-validation reporting mean R2=0.921; Implementation sample: one Taiwanese upstream textile dyeing SME where the C2M decision-support system was deployed and evaluated over a 12-month operational period (pre- and post-implementation operational metrics reported). Themesproductivity human_ai_collab IdentificationNon-randomized before-after field deployment at a single Taiwanese dyeing SME supplemented by model validation on observational e-commerce data: (1) ML demand-forecasting model trained on 3.87 million consumer comments across 127,846 product listings with cross-validation (mean R2=0.921) and validation that comment volume proxies sales; (2) forecasts converted to production guidance via Monte Carlo simulation and delivered through a decision dashboard; (3) operational outcomes compared pre- and post-implementation over a 12-month field study (inventory value, dye-lot changeovers, capacity utilization). No randomized control or external comparator was reported. GeneralizabilitySingle-firm field study in one country (Taiwan) limits transferability to other firms, sectors, or national contexts, Results pertain to upstream textile dyeing SMEs and may not generalize to large manufacturers, other industries, or other supply-chain structures, Semantic translation models may be language- and platform-specific; consumer comment usefulness as a sales proxy may vary by market and product category, Effect sizes could depend on the SME’s prior digitization level, managerial capacity to act on forecasts, and specific implementation support received, Temporal effects, seasonality, or concurrent demand shocks during the 12-month period may limit applicability to other timeframes

Claims (13)

Claim	Direction	Confidence	Outcome	Details
Upstream textile SMEs frequently exhibit constrained supply chain resilience owing to persistent information latency and structural dependence on downstream orders. Organizational Efficiency	negative	high	supply chain resilience (constrained due to information latency and downstream order dependence)	0.08
This study develops and validates a customer-to-manufacturer (C2M) intelligence framework that enables data-driven production planning using publicly available e-commerce data. Adoption Rate	positive	high	feasibility and validation of a C2M intelligence framework for production planning	n=3870000 0.48
The framework incorporates ethically compliant acquisition of consumer demand signals, semantic translation of unstructured market data into textile engineering attributes, machine-learning-based demand forecasting, and human-centric decision support. Organizational Efficiency	positive	high	presence of specified framework components (ethical data acquisition, semantic translation, ML forecasting, human-centric decision support)	0.48
The study utilized 3.87 million consumer comments from 127,846 product listings to build and validate models. Other	null_result	high	dataset size (number of consumer comments and product listings)	n=3870000 127,846 product listings 0.8
A Neural Boosted Tree model with entity embeddings for textile attributes was constructed and achieved a mean R2 of 0.921 in cross-validation, surpassing benchmark methods. Decision Quality	positive	high	forecasting accuracy (mean R2)	n=3870000 mean R2 of 0.921 0.48
Consumer comment volume was validated as a proxy for sales activity, facilitating demand estimation. Other	positive	high	validity of consumer comment volume as proxy for sales activity	n=3870000 0.48
Forecasts were translated into production guidance using Monte Carlo simulation and a decision dashboard. Organizational Efficiency	positive	high	operational production guidance derived from forecasts (method implementation)	0.48
In a 12-month field study at a Taiwanese dyeing SME, implementation resulted in a 28% reduction in inventory value. Organizational Efficiency	positive	high	inventory value	n=1 28% reduction in inventory value 0.48
In the same 12-month field study, implementation resulted in a 31% decrease in dye lot changeovers. Organizational Efficiency	positive	high	number of dye lot changeovers	n=1 31% decrease in dye lot changeovers 0.48
In the same 12-month field study, implementation resulted in a 16% increase in capacity utilization. Firm Productivity	positive	high	capacity utilization	n=1 16% increase in capacity utilization 0.48
This research extends the C2M paradigm from downstream retail contexts to upstream textile SMEs and proposes an integrated and operationally feasible intelligence framework for resource-constrained manufacturers. Adoption Rate	positive	high	extension and operational feasibility of C2M paradigm for upstream textile SMEs	n=1 0.48
The results indicate that upstream textile SMEs can leverage publicly visible e-commerce signals to enhance production planning responsiveness, minimize inventory exposure and dye-lot disruptions, and strengthen resilience to demand uncertainty through planner-centered digital decision support. Organizational Efficiency	positive	high	production planning responsiveness, inventory exposure, dye-lot disruptions, resilience to demand uncertainty	n=3870000 0.48
The framework demonstrates how digital intelligence can enhance supply chain resilience while supporting, rather than replacing, human decision-making (human-centric/planner-centered decision support). Decision Quality	positive	high	human-centric support vs. automation replacing planners	n=1 0.24

E‑commerce comments used as demand signals cut inventory and changeovers at a Taiwanese dyeing SME: a C2M AI system reduced inventory value by 28%, lowered dye-lot changeovers by 31% and raised capacity utilization by 16% in a 12‑month field deployment.