IoT and machine learning drive supply-chain gains in Batam manufacturing by enabling end-to-end automation; robotics and computer vision deliver smaller, more context-dependent benefits, especially for smaller firms.

Main Finding

IoT and Machine Learning are the primary drivers of supply chain efficiency in Batam’s manufacturing sector; their impact is amplified when embedded within end-to-end Supply Chain Automation Processes (SCAP). Robotics & Automation and Computer Vision have weaker or context-dependent effects (largely due to capital/integration barriers). Overall, AI + SCAP jointly explain a substantial share of variance in Supply Chain Efficiency (SCE), with inventory management, warehouse automation, logistics optimization, and predictive maintenance identified as the key automation domains that translate AI capabilities into performance gains.

Key Points

• Context: Study of 320 purposively selected manufacturing firms in Batam (electronics, shipbuilding, plastics, automotive, consumer goods) operating in FTZ/SEZ.
• AI disaggregated into five sub-technologies: Machine Learning (ML), Internet of Things (IoT), Robotics & Automation (R&A), Natural Language Processing & Chatbots (NLP&C), and Computer Vision (COMV).
• SCAP (mediator) comprises Inventory Management Automation, Logistics & Transportation Automation, Procurement & Supplier Management Automation, Warehouse Operations Automation, and Predictive Maintenance.
• Main empirical results:
  • IoT and ML show highest adoption and strongest positive effects on SCE (operational efficiency, accuracy/reliability, customer satisfaction).
  • SCAP mediates the relationship between AI technologies and SCE—AI yields greatest benefits when deployed through automation processes.
  • R&A and COMV display moderate/heterogeneous impacts, constrained by capital intensity and integration difficulty (especially for SMEs).
  • NLP&C moderate adoption; contributes to communication/response time gains.
• Measurement and robustness: survey items adapted from prior literature; constructs show good reliability and convergent validity (Cronbach’s alpha, CR, AVE thresholds met). Common method bias checks (Harman’s single-factor, VIFs <3.3, marker-variable) reported.

Data & Methods

• Population/sample: 1,309 registered manufacturing firms in Batam; purposive sampling to obtain 320 respondent firms (managers/supervisors). Sector and size mix: electronics 34%, shipbuilding 20%, plastics 18%, machinery/automotive 15%; firm sizes—small 20%, medium 45%, large 35%; ownership—55% foreign, 45% domestic.
• Instrument: Structured questionnaire measuring AI sub-technologies, SCAP domains, and SCE dimensions (operational efficiency, speed/responsiveness, accuracy/reliability, resilience/risk mitigation, customer satisfaction). Items adapted from established sources (Kamble, Wamba, Marinov, Papadopoulos, etc.).
• Analytical approach: Partial Least Squares Structural Equation Modeling (PLS-SEM) to estimate direct and mediated effects. Reliability/validity (factor loadings, Cronbach’s alpha, Composite Reliability, AVE) all satisfactory. Procedural and statistical steps taken to limit common method bias.
• Limitations noted by authors (implied/typical): cross-sectional self-reported survey, purposive sample limits generalizability beyond Batam, causality inference limited by design.

Implications for AI Economics

• Technology-specific returns: Disaggregating “AI” matters. IoT and analytics (ML/predictive analytics) generate higher immediate returns to supply chain efficiency than capital-intensive robotics or vision systems—relevant for cost-benefit prioritization and investment sequencing.
• Diffusion & scale: SMEs face steeper adoption barriers for R&A and COMV due to upfront capital and integration complexity. Policies that subsidize capex, provide shared automation facilities, or promote leasing models could accelerate uptake and equalize productivity gains.
• Productivity and competitiveness: IoT+ML embedded in SCAP improves real-time visibility, demand forecasting, inventory turns, and service quality—mechanisms through which AI raises firm-level productivity and the competitiveness of FTZ/SEZ clusters.
• Labor and distributional effects: The relative advantage of data-driven vs. capital-intensive automation suggests heterogeneous labor impacts—IoT/analytics may complement managerial/technical labor (skills upgrading), while R&A could displace routine roles. Economic analyses should model distributional outcomes (wages, employment composition).
• Network and trade effects: In export-oriented FTZ clusters like Batam, upstream/downstream supplier integration (procurement automation, logistics orchestration) magnifies gains; cross-firm coordination frictions imply potential for positive spillovers but also winners/losers across supply-chain positions.
• Policy & investment sequencing: For constrained budgets, prioritize IoT infrastructure and analytics platforms (lower marginal cost, faster ROI) and concurrent investment in workforce digital skills and data governance to capture scale benefits from later adoption of robotics/vision.
• Research agenda for AI economics: quantify ROI and R&D-like spillovers of different AI technologies; estimate heterogeneous productivity elasticities by firm size/sector; model general equilibrium implications of automation sequencing in export-oriented clusters; use longitudinal or quasi-experimental designs to identify causal effects and labor-market impacts.

Notes and suggested follow-ups: - The paper reports “substantial” explained variance but does not present exact R² values in the excerpt — checking full results (R² for SCAP and SCE) would help quantify aggregate economic impact. - Future empirical work should complement self-reports with operational metrics (throughput, lead time, stockouts, downtime) and examine dynamic adoption paths (phased investments, learning-by-doing).