Enterprise AI, written to be cited.

To establish, from the public record, what enterprise AI has proved and falsified in its first sustained period in production — and to set the frame, the maturity baseline, and the forward research agenda for the five-pillar programme.

• P1-O1Assess headline AI adoption rates against actual production impact (e.g. ~95% of pilots showing no measured P&L impact — NANDA/MIT).
• P1-O2Identify the structural failure modes: buying tools vs building capability; central AI labs vs line-manager ownership.
• P1-O3Document the ‘maturity ladder’ and where most organisations sit (stalled at scaled pilots).
• P1-O4Establish the forward research agenda for the publication and the organisation.

To demonstrate that production AI outcomes are determined by the architecture wrapped around the model, and to give technical leaders an evidence-based framework for moving from demo to reliable production.

• P2-O1Document the retrieval architecture patterns: naive RAG → hybrid → reranking → hierarchical/graph.
• P2-O2Map agentic maturity across a five-tier taxonomy (Tier 1: deterministic-with-LLM-glue → Tier 5: fully autonomous).
• P2-O3Identify which tiers are production-ready for which workloads.
• P2-O4Provide a decision framework keyed to corpus characteristics, query complexity, latency budget, and error tolerance.

To demonstrate that the operating model — not the technology choice — determines enterprise AI outcomes, and to define the archetypes, components, and conditions that separate success from failure.

• P3-O1Define the four operating-model archetypes: Centralised Platform, Centralised Delivery, Federated Platform, Federated Delivery/CoE.
• P3-O2Articulate the six conditions of success: production reach, evaluation in production, systems-of-record integration, governance + delivery, talent retention, executive sponsorship.
• P3-O3Ground the analysis in case studies: JPMorgan Chase, Sanofi, Walmart, the Klarna reversal, the McDonald’s failure.
• P3-O4Identify the five components of an AI operating model: design authority, build capacity, governance regime, run model, funding flow.

To give enterprise decision-makers a fully-loaded view of what production AI actually costs, and a vendor-neutral basis for choosing platforms on the factors that genuinely determine fit — data residency, regulatory accreditation, existing stack, and language — rather than on headline model capability or per-token price.

• P4-O1Decompose the full enterprise-AI cost stack across eight categories and establish that inference is only 20–40% of run-rate cost.
• P4-O2Map the proportional cost shape across the five workload archetypes and the pilot / department / enterprise scale tiers.
• P4-O3Audit the global platform landscape — Western, Chinese, Indian, Korean, Japanese, Middle-Eastern — against capability, cost, data residency, and accreditation.
• P4-O4Establish that platform choice is governed by platform, identity, and regulatory gravity rather than raw capability.

To establish that evaluation and governance are a single operational system, and to give enterprises an inspectable maturity yardstick and a set of operational components and regulatory mappings that turn AI trust from aspiration into verifiable practice.

• P5-O1Establish the seven evaluation dimensions and five methods that constitute a production evaluation practice.
• P5-O2Define the seven operational components of a working governance system and the test that distinguishes governance from compliance theatre.
• P5-O3Map enterprise obligations across the EU AI Act, NIST AI RMF, and ISO/IEC 42001 to concrete engineering work items.
• P5-O4Provide a combined maturity framework that locates a programme on a four-level ladder by inspection.