AI speeds up procurement negotiations but trust is conditional: explainability matters most, and human oversight and governance matter too; moderate, well-governed AI use with human-in-the-loop design preserves trust, with EU practitioners emphasizing XAI while ASEAN practitioners place more weight on human oversight.

Main Finding

AI-augmented procurement negotiation systems raise perceived operational efficiency but can erode professionals’ trust in the AI when automation displaces relational cues and explanations are weak. Perceived explainability (XAI) has the strongest positive association with AI system trust; visible ethical governance and human‑in‑the‑loop (HIL) relational design also raise trust. XAI and HIL together statistically account for the observed association between AI use and trust, and a small inverse‑U pattern suggests moderate, well‑governed automation yields higher trust than very low or very high automation intensity. Regional differences: XAI–trust links are stronger in the EU; HIL–trust links are stronger in ASEAN. Evidence is associative (exploratory, non‑probability sample) and not causal.

Key Points

• Scope: AI-augmented procurement negotiation systems = ML/generative tools that recommend prices/concessions, draft negotiation artifacts, or automate limited interaction steps while humans retain final decision authority.
• Trust construct: Focus on AI system trust (willingness to rely on the AI; beliefs about competence, fairness, responsibility). Explicitly distinguished from dyadic buyer–supplier trust (not measured).
• Trust‑by‑design levers:
  • XAI (transparency/explainability): intelligibility and traceability of recommendations.
  • Ethical governance visibility: auditability, accountability, redress mechanisms.
  • Human‑in‑the‑loop relational design (HIL/RD): review gates, override authority, human contact channels to preserve relational cues.
• Main hypotheses tested:
  • H1: AI use → higher perceived efficiency but lower AI trust when transparency/HIL weak.
  • H2–H4: XAI, governance visibility, and HIL each positively associated with AI system trust.
• Empirical patterns:
  • AI use positively associated with perceived negotiation efficiency.
  • XAI shows the largest positive association with AI trust; governance and HIL add further positive associations.
  • XAI and HIL mediate the AI use → trust association in augmented models.
  • Small inverse‑U relationship: moderate, well‑governed automation → higher trust than extremes.
  • Cross‑region differences: EU emphasizes XAI and auditability; ASEAN emphasizes HIL and relational legitimacy.
• Contributions: clarifies trust boundary (AI system vs interfirm trust), documents procurement‑specific mechanisms (confidentiality, auditability, negotiation tacticity) that can invert the efficiency→trust relationship, and proposes an evidence‑consistent roadmap for trustworthy negotiation systems while noting limits on generalization.

Data & Methods

• Design: Exploratory sequential mixed‑methods (QUAL → QUAN).
  • Qualitative phase: 6 multi‑sector case studies across EU and ASEAN; 36 semi‑structured interviews used to refine constructs and survey items.
  • Quantitative phase: Cross‑sectional online survey of purposively recruited AI‑exposed procurement professionals (N = 238).
• Analysis:
  • Measurement development via qualitative grounding and confirmatory factor analysis for discriminant validity.
  • Structural equation modeling (SEM) to estimate associations among AI use, perceived efficiency, XAI, governance visibility, HIL, and AI system trust.
  • Augmented SEM tested mediation (XAI and HIL accounting for AI use–trust link).
  • Multi‑group SEM comparing EU vs ASEAN samples.
• Key limitations:
  • Non‑probability, cross‑sectional, single‑source sample → associations not causal; not population‑representative.
  • Dyadic buyer–supplier trust and negotiation outcomes (prices, concessions, relationship quality) were not measured; spillovers cannot be estimated here.
  • Possible unmeasured confounders; interactions among the three levers were not modeled in the main specification.

Implications for AI Economics

• Adoption vs trust trade‑off: AI increases operational efficiency in procurement but can reduce users’ willingness to rely on AI unless explainability and governance mechanisms are in place. Economic models of AI diffusion and adoption should include non‑monotonic effects of automation intensity on effective uptake mediated by trust.
• Investment priorities and returns:
  • Explainability (XAI) yields the largest association with trust — investing in XAI is likely to improve acceptance and effective use, potentially increasing realized efficiency gains.
  • HIL design and governance visibility also raise trust; their value may vary by institutional context (higher marginal value of HIL in relational contexts such as ASEAN; higher marginal value of XAI/auditability in regulated contexts such as the EU).
  • Firms should consider the cost‑benefit of added XAI/governance layers vs incremental efficiency; the inverse‑U pattern implies there is an optimal automation intensity balancing efficiency gains and trust‑dependent usage.
• Market structure and bargaining outcomes:
  • Changes in trust and reliance on AI can alter bargaining power, transaction costs, and information asymmetries in procurement markets. For example, low trust may suppress use of AI recommendations, preserving existing human bargaining rents; high trust supported by XAI/governance may shift surplus extraction by enabling faster, more consistent offers.
  • Regulatory compliance costs (e.g., EU AI Act requirements for high‑risk systems) will affect the economics of procurement AI deployment — smaller suppliers/buyers may face higher relative compliance burdens, potentially affecting market concentration.
• Policy implications:
  • Regulators and standard‑setters that promote explainability, audit trails, and human‑oversight mechanisms can reduce negative externalities from opaque automation and encourage socially beneficial uptake.
  • Differential regional regulatory regimes imply heterogeneous adoption strategies and welfare impacts; cross‑jurisdictional procurement and global supply chains must account for these frictions.
• Research directions for AI economics:
  • Causal and structural work to quantify welfare and distributional effects: estimate how investments in XAI/governance/HIL change negotiation outcomes (prices, contract terms), search and matching frictions, and bargaining surplus.
  • Dyadic, panel, or experiment designs to identify causal pathways from AI design features → trust → enacted behavior → economic outcomes.
  • Cost‑effectiveness analyses comparing returns to spending on XAI, governance, and HIL versus the marginal efficiency gains from further automation.
  • Models of equilibrium automation intensity in procurement platforms that internalize trust dynamics (endogenous uptake) and regulatory compliance costs.
  • Investigate heterogeneity across firm size, sector, and institutional context to predict differential diffusion and market‑structure consequences.
• Practical takeaway for economists advising firms/policymakers: encourage moderate, well‑governed automation with transparent explanations and preserved human oversight to capture efficiency gains while maintaining the trust needed for those gains to materialize in contract and price outcomes.