Supply-chain AI often underdelivers not because models are weak but because frontline workers, vendors, and processes won't adopt them; treating internal AI as a product and applying product-marketing/CVM plus Lean Six Sigma can close the 'Adoption Gap' and unlock realized value.

Main Finding

The paper diagnoses a persistent “Adoption Gap” as the primary reason enterprise supply‑chain AI investments fail to deliver operational value. The gap is a structural disconnect between engineering‑centric development (accuracy, models) and frontline / multi‑firm operational realities (perceived usefulness, ease of use, incentives, data sharing). The authors argue that closing this gap requires importing product‑marketing and Customer Value Management (CVM) practices into internal enterprise software deployment and pairing those practices with a Lean Six Sigma operational blueprint so predictive systems become steerable, trusted, and adopted at scale.

Key Points

• Adoption Gap defined: predictive artifacts function technically but do not change how work is done because operators, suppliers, and managers do not adopt or act on them.
• Not primarily an algorithmic failure: model accuracy matters but is inert without adoption; common failure mode is stalled pilots rather than broken models.
• Pilot trap: pilots run under favorable conditions by enthusiasts do not test adoptability in ordinary production settings; scaling requires designing for production/ adoption from day one.
• Enterprise software precedent: parallels to ERP-era failures — value requires reorganizing processes, incentives, and workflows around the system, not merely installing it.
• Theoretical synthesis: draws on technology‑acceptance, diffusion of innovation, trust in automation, and value management literatures to explain adoption dynamics (perceived usefulness, ease of use, social influence, facilitating conditions).
• Proposed remedy: apply product marketing and CVM internally and across supplier tiers to translate predictive capability into captured operational value:
  • Create operator‑centered personas and internal value propositions.
  • Reframe predictive metadata into high‑incentive, legible, steerable workflows (design for ease of use and trust).
  • Build an internal “go‑to‑market” (launch) motion with continuous feedback loops and acceptance metrics.
  • Extend CVM outward to align supplier incentives and improve vendor data integrity.
• Operationalization: a Lean Six Sigma blueprint that phases adoption (Define, Measure, Analyze, Improve, Control) with concrete adoption metrics (e.g., override rate, percent of decisions influenced, supplier participation, data completeness, time‑to‑value).
• Limits: conceptual synthesis without empirical testing; calls for field studies to validate the framework and measure effect sizes.

Data & Methods

• Nature: conceptual synthesis and theory development (not empirical).
• Inputs: systematic and practitioner literatures on AI in supply chains, technology adoption (TAM, UTAUT), diffusion, trust in automation, resilience, ERP/enterprise software history, product marketing, and CVM.
• Method: interdisciplinary integration — diagnose the adoption problem, map causal mechanisms from the literature, translate product‑marketing/CVM methods into enterprise and inter‑firm contexts, and derive an operational Lean Six Sigma blueprint to implement and measure adoption.
• Outputs: conceptual framework and operational blueprint (phased adoption plan and suggested metrics). The paper explicitly identifies the need for empirical validation and sets an agenda for future field studies.

Implications for AI Economics

• Returns to AI depend critically on complementary investments: economic value of supply‑chain AI accrues only when firms invest in organizational redesign, frontline training, UX/productization, and supplier incentives — thus standard ROI assessments that focus on model performance will misstate expected returns.
• Reallocation of R&D spending: economics of AI deployment should account for tradeoffs between further model accuracy and investments in adoption (product managers, change management, incentive design); marginal returns likely higher for adoption investments in many cases.
• Measurement & evaluation: policymakers, investors, and firms should include adoption metrics (override rates, decision influence, supplier participation) alongside technical benchmarks when valuing AI projects and setting funding/priorities.
• Market design and contracting: multi‑tier supply chains create externalities (data sharing, coordination). Optimal contracting, pricing of data, and procurement practices must be designed to internalize benefits to suppliers so vendor data integrity and participation improve.
• Labor and incentives: economics of labor in instrumented operations must consider how automation complements (not merely replaces) work — value capture depends on aligning operator incentives and perceptions of usefulness; misalignment can suppress productivity gains.
• Productivity paradox reframed: the lag between AI investment and measured productivity arises from missing complementary organizational capital; macroeconomic assessments of AI diffusion should model these complementarities and adjustment lags.
• Policy and resilience: supporting standards for data interoperability, incentive‑compatible data‑sharing mechanisms, and training subsidies could accelerate socially valuable adoption and industrial resilience.
• Research agenda for AI economics: quantify effect sizes of adoption interventions, estimate social returns to complementary investments, compare welfare outcomes under different contracting/incentive regimes in multi‑firm supply networks, and measure how adoption improvements change the marginal value of prediction accuracy.

If you want, I can (a) extract a concise checklist of operational adoption metrics and how to measure them, or (b) translate the Lean Six Sigma adoption blueprint into a one‑page implementation plan for supply‑chain teams. Which would be most useful?