← Papers

Most enterprise AI projects fail not because models are inadequate but because organizational and architectural friction prevents production; the paper offers a six-stage 'Deployment Wall' and a reproducible 'Seam Index' to diagnose and cost those frictions so buyers can compare platforms on architecture rather than benchmarks.

The Deployment Wall: A Diagnostic Framework and Instrument for Enterprise AI in the Deployment Era

Fabricio F. Costa · July 31, 2026

arxiv descriptive low evidence 7/10 relevance Full text usable extracted full text Source PDF

Structured author observations

Linked only from stored provider relations; the raw author line above is never matched by name.

Arxiv

Latest observation: August 4, 2026

Fabricio F. Costa unresolved corpus identity

Semantic Scholar

Latest observation: August 25, 2026

Fabricio F Costa provider ID

Enterprise AI value is limited not by model capability but by deployment frictions, so the paper proposes the 'Deployment Wall' (six-stage value-leak model), a 0–12 'Seam Index' diagnostic, and the 'Deployment Debt' construct to make deployment barriers measurable for platform decisions.

Citation observations

Cumulative provider counts captured on specific dates; providers are never combined.

0 cumulative citations

OpenAlex · Observed August 7, 2026

View corpus context

0 cumulative citations

Semantic Scholar · Observed August 25, 2026

View corpus context

Enterprise investment in generative artificial intelligence (AI) tripled in a single year to roughly US$37 billion, yet independent field research finds that about 95% of enterprise generative-AI pilots deliver no measurable profit-and-loss impact. We argue that the dominant explanation--that models are not yet capable enough--is mistaken, and that enterprise AI has entered a Deployment Era in which advantage derives not from model intelligence but from the removal of the organizational and architectural friction that prevents a capable model from reaching production. Building on the software-engineering literature on technical debt and machine-learning deployment, and on a structured synthesis of independent field studies, we make the diagnosis operational. We introduce three linked constructs and one measurement instrument: the Deployment Wall, a six-stage value-leak model that mechanically reproduces observed survival rates; the Seam Index, a reproducible 0-12 diagnostic that scores any platform by how many of six recurring friction "seams" it removes natively rather than leaving to the adopter; and Deployment Debt, a construct that reframes unresolved friction as a compounding, quantifiable liability. We specify a scoring protocol with evidence anchors so the instrument can be applied consistently, illustrate it on a worked platform-selection example, and derive six falsifiable propositions with a research agenda for validation. The framework converts an eight-figure platform decision from a benchmark comparison into an architecture comparison.

Summary

Main Finding

Enterprise generative-AI value is no longer gated primarily by model intelligence but by deployment friction. The paper proposes a diagnostic framework and instrument showing that most pilots fail to produce P&L impact not because models are inadequate but because organizational and architectural seams prevent capable models from reaching durable production. Building deployment capability now compounds a durable competitive advantage; misdiagnosing the problem as “more intelligence” favors costly inaction.

Key Points

Stylized empirical facts motivating the framework
- Enterprise generative-AI spending rose sharply (≈US$37B) even as independent field research (MIT Project NANDA et al.) finds ≈95% of pilots deliver no measurable P&L impact.
- Frontier models are converging on benchmarks; marginal competitive value from model selection is falling (intelligence commoditizing).
Core constructs introduced
- Deployment Wall: a six-stage, cumulative value‑leak / survival funnel that mechanically reproduces observed low survival-to-value rates. Stages: 1) model selection, 2) integration, 3) governance, 4) workflow redesign, 5) enterprise adoption, 6) realized business value.
- Six recurring seams (where value leaks): Data; Identity & Access; Security & Compliance; Governance (monitoring/auditability/accountability); Change Management (workflow redesign, skills, incentives); Cost (total cost of ownership: integration, maintenance, inference).
- Seam Index: a reproducible 0–12 diagnostic instrument that scores a platform by how many of the six seams it removes natively (scoring protocol with evidence anchors provided).
- Deployment Debt: reframes unresolved seams/friction as a compounding, quantifiable liability (analogous to technical debt) that accrues operational, governance, and competitive cost over time.
Managerial and strategic implications highlighted
- Model selection consumes visible attention but is a small part of the work that determines outcomes; most effort and risk lies in crossing seams.
- The durable moat shifts from the model layer to seam-layer assets (complements). Hyperscalers and platform providers that natively absorb seams create distributional competitive advantages.
- Platform choice becomes an architecture/operations decision (how many seams the vendor removes) rather than a pure benchmark/model selection decision.
- Misdiagnosis (waiting for “better models”) is costly because deployment capability compounds advantage and is not easily purchasable later.

Data & Methods

Methodological approach
- Conceptual, design-oriented paper (not a randomized empirical study). Constructs and instrument developed inductively from:
  - Author’s participant-observation (professional experience deploying frontier models across organizations of various sizes and regulated sectors).
  - Structured synthesis of independent field studies, surveys, and market analyses (examples: MIT Project NANDA 2025, RAND 2024, McKinsey 2025, BCG 2025, S&P Global 2025, Menlo Ventures 2025). Table 1 in the paper lists the studies used.
- Reproducibility: Seam Index includes an explicit scoring protocol with evidence anchors so independent evaluators can apply it consistently. The Deployment Wall is an illustrative survival model calibrated to observed field survival rates; propositions are stated in falsifiable form for empirical testing.
Evidence types and boundaries
- Participant-observation: rich, longitudinal, non-random; used to generate hypotheses and instrument design.
- Structured synthesis of public studies: triangulates field observations; used to ground the survival/leak patterns and seam importance.
- Boundary conditions: framework most directly applies to medium-to-large, regulated or data-sensitive organizations. Very small/simple firms may still be gated by model quality. The framework is calibrated to the current deployment era; seam heights may shift as platforms absorb more seams.

Implications for AI Economics

Where economic rents will accrue
- As model intelligence commoditizes, rents move to complementary, co-specialized assets — here, the seams (data plumbing, identity integration, governance tooling, operationalized workflows, cost control).
- Providers that natively remove seams (higher Seam Index) create platform-level moats; this explains strategic moves by frontier labs and cloud providers to bundle deployment services and certified partners.
- Valuation and buyer choice should price not only model capability but also the degree to which a vendor reduces buyers’ deployment friction (i.e., how much Deployment Debt it prevents).
Investment and strategy consequences for firms
- Firms should treat deployment capability as a strategic asset to be built early. Early investment compounds advantage; waiting for marginally better models risks falling behind.
- Capital-allocation decisions (e.g., eight-figure platform buys) should be converted from benchmark comparison to architecture comparison: evaluate platforms by Seam Index and the expected reduction in Deployment Debt.
- Procurement and governance processes need metrics (Seam Index, quantified Deployment Debt) to compare vendors and to surface long-run TCO and governance liabilities.
Measurement, markets, and policy implications
- Deployment Debt provides a framework to quantify future operational liability — useful for boards, auditors, and regulatory risk assessments.
- Market pricing of platform services may shift from per-token or model-performance metrics to bundled payments for seam removal (data connectors, enterprise identity integration, compliance-ready monitoring).
- Regulators and firms should recognize that value and risk are concentrated not only in model internals but in the seams where models touch enterprises.
Research agenda (summary)
- Validate that higher Seam Index correlates with higher survival to realized value and greater ROI.
- Quantify Deployment Debt and its compounding dynamics; test how seam absorption by platforms lowers the Deployment Wall.
- Track how market structure evolves as hyperscalers and labs internalize seams (impacts on competition, prices, entry).
Limitations and caveats
- Framework is conceptual and needs empirical validation; evidence base is participant-observation plus synthesis of public studies.
- Best suited to medium-to-large organizations and regulated environments; small/simple firms may face different constraints.
- As tooling/platforms evolve and absorb seams, the absolute height of the Deployment Wall may fall though its structure likely persists.

Overall, the paper reframes the enterprise-AI problem from “which model?” to “which seams are removed?” and supplies a practical instrument (Seam Index) and framing (Deployment Wall / Deployment Debt) to guide investment, procurement, and research in the Deployment Era.

Assessment

Paper Typedescriptive Evidence Strengthlow — The paper is primarily conceptual and design-oriented: constructs and an instrument are inductively derived from the author’s participant-observation across deployments and from a structured synthesis of published industry studies and surveys. It cites large external surveys (e.g., MIT Project NANDA, McKinsey, BCG, RAND) to triangulate claims, but it presents no original, systematic empirical test or causal identification of the proposed mechanisms or instrument performance. Methods Rigormedium — The development used a defensible mix of practitioner experience and structured synthesis of multiple published field studies, and the paper provides an explicit scoring protocol for the Seam Index. However, the sampling of deployments is non-random and observational, there is no independent validation or statistical testing of the instrument, and possible biases from the author’s consulting engagements or affiliation are not empirically addressed. SampleNo original randomized or observational dataset is presented. The constructs are derived from (a) the author’s participant-observation and professional experience deploying frontier models across multiple enterprises and engagements (non-random, consulting-case evidence), and (b) a structured synthesis of publicly available industry studies and surveys (e.g., MIT Project NANDA: ~150 interviews/350 surveys/300 deployments; RAND: 65 expert interviews; McKinsey: ~1,900 respondents; BCG: ~1,000 executives; S&P Global: ~1,000+ firms; Menlo Ventures enterprise buyer survey), plus market and vendor signals and public disclosures. Themesadoption org_design productivity human_ai_collab GeneralizabilityCalibrated to medium-to-large, regulated or data-sensitive enterprises; may overstate seams for small/simple organizations, Derived from non-random practitioner engagements—susceptible to selection and confirmation bias, Instrument and propositions lack empirical validation across industries, geographies, and firm sizes, Assumes current generation of frontier models and enterprise tooling; applicability may change as platforms natively absorb seams, Potential conflict of interest / vendor perspective given the author’s industry role (HCLTech) that could shape emphasis on platform/architecture choices

Claims (13)

Claim	Direction	Outcome	Confidence & Evidence	Details
Enterprise spending on generative AI tripled in a single year to approximately US$37 billion. Fiscal And Macroeconomic	positive	Enterprise generative-AI spending	Reading fidelity high Study strength medium	tripled to roughly US$37 billion 0.18
Approximately 95% of enterprise generative-AI pilots produce no measurable profit-and-loss impact. Firm Revenue	negative	Measurable profit-and-loss impact from enterprise generative-AI pilots	Reading fidelity high Study strength medium	n=300 approximately 95% 0.18
The divide between organizations that capture value from enterprise AI and those that do not is determined by organizational approach rather than model quality. Organizational Efficiency	mixed	Enterprise AI value capture	Reading fidelity high Study strength medium	n=300 0.18
Most AI projects fail to reach production, and the causes of failure are largely organizational. Adoption Rate	negative	Reaching production deployment	Reading fidelity high Study strength medium	n=65 0.18
The Deployment Wall models enterprise AI value realization as six cumulative stages: model selection, integration, governance, workflow redesign, enterprise adoption, and realized business value. Organizational Efficiency	negative	Realized business value from enterprise AI initiatives	Reading fidelity high Study strength speculative	not reported 0.03
Moderate attrition at each of the six Deployment Wall stages can reduce cumulative survival to the single digits. Adoption Rate	negative	Share of enterprise AI initiatives surviving to realized value	Reading fidelity high Study strength speculative	single-digit cumulative survival rate 0.03
Model selection accounts for a small share of the work determining enterprise AI success, while most effort and failure risk lie in integration, governance, workflow redesign, and adoption. Organizational Efficiency	negative	Distribution of deployment effort and failure risk across activities	Reading fidelity high Study strength speculative	not reported 0.03
Data integration routinely consumes the single largest share of effort in enterprise AI deployment. Organizational Efficiency	negative	Deployment effort required for enterprise data integration	Reading fidelity high Study strength medium	not reported 0.18
A majority of corporate generative-AI connections lack single sign-on, creating an identity and access gap for enterprise AI. Governance And Regulation	negative	Presence of single sign-on and identity controls in corporate generative-AI connections	Reading fidelity high Study strength medium	a majority 0.18
Only a minority of organizations report having robust, enterprise-wide AI governance. Governance And Regulation	negative	Presence of enterprise-wide AI governance capabilities	Reading fidelity high Study strength medium	only a minority 0.18
Only a minority of firms redesign workflows around AI, and adoption rather than model capability ultimately gates value. Task Allocation	negative	Workflow redesign and enterprise adoption of AI	Reading fidelity high Study strength medium	n=1900 only a minority 0.18
Ongoing integration, maintenance, monitoring, and inference costs dominate the lifetime cost of operating enterprise AI systems. Organizational Efficiency	negative	Total cost of ownership and operating cost of enterprise AI systems	Reading fidelity high Study strength medium	not reported 0.18
The paper's Deployment Wall, Seam Index, and Deployment Debt are original conceptual models, while the paper is not an empirical study. Other	null_result	Empirical validation status of the proposed framework and instrument	Reading fidelity high Study strength high	not reported 0.3