Formal enterprise architecture is rare in post‑conflict Afghanistan but correlates with markedly better digital outcomes; a pragmatic 'EA‑Lite' model is proposed to deliver coordination, reduce duplication and support AI/digital investments where institutional capacity is weak.

Main Finding

Despite widespread digital transformation (DT) failures in fragile states, formal enterprise architecture (EA) is rare (16.2% of organisations) yet—where present—is associated with substantially better outcomes (standardisation, innovation enablement, technology risk reduction). Conventional EA frameworks (TOGAF, Zachman, FEAF) are mismatched to post-conflict contexts; the paper proposes a pragmatic “EA‑Lite” model and a revised conceptual framework tailored to fragile states, emphasising lightweight, value-driven, politically savvy, leadership-centric practice.

Key Points

• Adoption paradox: only 16.2% of surveyed organisations have a formal EA function, but those that do report markedly stronger outcomes:
  • Standardisation mean = 4.18
  • Innovation enablement mean = 4.00
  • Technology risk reduction mean = 3.94
• Organisations without EA rate long‑term technology planning lowest (mean = 2.84).
• DT implementation intensity correlates strongly with perceived DT success (r = 0.654, p < .001).
• Sector membership does not predict DT success (t(80) = −0.57, p = .572) nor EA adoption (χ²(1) = 0.045, p = .832) — the deficit is system‑wide rather than sector‑specific.
• Dominant barriers are human/contextual, not technical:
  • Shortage of skilled professionals (mean = 3.66), exacerbated by brain drain
  • Political and economic instability (mean = 3.65)
• The single decisive enabler identified: committed senior leadership.
• Established EA frameworks assume institutional stability, specialist labour markets, and formal governance — assumptions invalid in post‑conflict economies.
• Proposed remedy: “EA‑Lite” — a practitioner‑derived, lightweight, value‑driven, politically aware, leadership‑centred EA approach and a revised conceptual framework for leveraging EA in fragile states.

Data & Methods

• Design: Convergent parallel mixed‑methods case study (triangulation of quantitative and qualitative data).
• Quantitative: Survey of 105 public and private sector professionals in Afghanistan.
• Qualitative: Seven in‑depth interviews with senior leaders.
• Analytical highlights:
  • Correlation: DT implementation intensity ↔ perceived DT success (r = .654, p < .001).
  • Group tests: Sector membership × DT success (t(80) = −0.57, p = .572); sector × EA adoption (χ²(1) = 0.045, p = .832).
  • Descriptive measures reported for outcome and barrier items (means as above).
• Interpretation framed against the institutional realities of a fragile, post‑conflict economy.

Implications for AI Economics

• Investment targeting: Heavy, classical EA investments (designed for stable institutions) are likely to deliver low marginal returns in fragile states. Allocate scarce funds toward lightweight coordination mechanisms (EA‑Lite), leadership incentives, and targeted capacity building.
• Labour market & human capital: Brain drain and skill shortages are central constraints for AI adoption. Economic models of AI deployment in fragile states must incorporate high human‑capital leakage, raising the cost and time to scale AI systems.
• Governance and risk: Political instability and weak governance heighten technology risk and increase the probability of fragmented, duplicated AI deployments. EA‑Lite and leadership‑centred governance can reduce duplication and systemic fragility, improving social returns to AI investments.
• Project design and evaluation: DT implementation intensity strongly predicts perceived success — donors and implementers should prioritise sustained implementation effort and leadership commitment in project design and conditional funding, rather than sectoral targeting alone.
• Scalability and standards: Because EA presence materially improves standardisation and risk reduction, lightweight standard‑setting (interoperability baselines, minimal data standards, reusable service primitives) can yield outsized efficiency gains for AI systems across organisations.
• Policy levers: Policies that (a) retain and retrain local talent (incentives, diaspora engagement), (b) stabilise political/economic conditions for digital initiatives, and (c) empower senior leaders to champion coordination will be more effective than purely technical assistance.
• Research & modelling: AI‑economics analyses should treat fragile‑state contexts as distinct regimes with different parameter values (higher institutional risk premiums, lower effective labour supply, greater coordination failure). Counterfactuals and ROI estimates must include costs of fragmentation and the benefits of lightweight governance interventions.
• Risk management: Rapid AI adoption without EA‑Lite coordination risks lock‑in of duplicated, incompatible systems that worsen long‑term costs—economic appraisals must internalise these systemic externalities.

If you want, I can (a) produce a short checklist for designing AI/DT interventions in fragile states based on EA‑Lite principles, or (b) map EA‑Lite elements to specific AI governance mechanisms (data sharing, model procurement, infrastructure reuse). Which would be most useful?