Sovereignty audit · Whitepaper

AI readiness before you build

Four scores. The evidence behind them. A dated list of what would fail if you started today.

54east · Abu Dhabi · Updated 2026-08-28 · Open access. Cited. No invented case studies.

Winning institutions own the system. They score what has to be true before a build is worth starting — then they own the result.

The expensive failure is the one that ships

The project that never starts is cheap. The expensive failure is the one that gets built, deployed, and then produces nothing: the data was never usable, no one was named to operate it, or a governance sign-off that should have happened before the build happened after, forcing a rebuild.

That is not a 54east statistic, and it is not “quietly unused within six months of production.” Quote the studies as they stand.

MIT NANDA’s The GenAI Divide: State of AI in Business 2025 found 95% of organizations getting zero return from GenAI — “the vast majority remain stuck with no measurable P&L impact.” Success, as the authors define it, is a marked and sustained productivity or P&L effect, not a demo that survived a steering committee.¹²

Gartner, reviewing GenAI implementations through the end of 2025, reports that at least 50% of projects were abandoned after proof of concept — poor data quality, inadequate risk controls, escalating costs, or unclear business value.³

Shelfware after go-live — a production system with no operational owner that is ignored rather than decommissioned — is practitioner judgment. Adjacent to those findings. Not the same number.

A readiness assessment exists to catch this before the build starts. The axes are not original. The deliverable is: four scores with the evidence attached, and a dated list of what would fail if you started today.

The UAE case is speed without the foundations

UAE institutions are under real pressure — competitive, and often regulatory — to show AI progress quickly. The pace is on the record. The foundations are not matching it.

Oxford Insights’ Government AI Readiness Index 2025 ranks the UAE 19th of 195 countries. On Public Sector Adoption the score is 97.27 — among the highest in the table. (The 2025 Index uses a new methodology; do not read the rank as a fall from prior editions.)⁴

In November 2025, Omar Sultan Al Olama, Minister of State for Artificial Intelligence, put utilisation of AI tools across government entities at 97%, and launched a federal AI Readiness Index for those entities.⁵

Enterprises are spending. They are not mature. ServiceNow’s Enterprise AI Maturity Index 2026, run by ThoughtLab with 100 UAE executives, found AI spend up 105% year on year and a maturity score of 48 out of 100. Seventy-seven percent of those executives cite inadequate data accuracy, access and management as a major barrier. Sixteen percent have implemented AI testing, auditing and risk-management processes.⁶⁷

Adoption and spend are unusually fast. Data and governance are not. That gap — not a maturity-model poster — is why you score readiness before you build.

The four dimensions are not ours

Data, infrastructure, workforce, governance. Every major readiness model already uses some version of these.

Cisco’s AI Readiness Index scores six pillars: Strategy, Infrastructure, Data, Governance, Talent, and Culture.⁸ Gartner’s AI maturity toolkit scores strategy, data, governance, engineering, operating model, culture, and AI product/value.⁹ Accenture and Carnegie Mellon SEI’s AI Adoption Maturity Model (June 2026) scores eight dimensions, including workforce and culture, risk and governance, data, engineering, and operations.¹⁰

We use the four foundations those models already share. We do not claim them as 54east IP.

The differentiator is what the score is attached to. A maturity-model poster — an arrow from “ad hoc” to “optimized” — is a self-reported position with no evidence. A skeptical stakeholder cannot check it. A useful assessment produces four scores, each with the specific evidence used to produce it, so the skeptic can see why the score is what it is.

Score each one against evidence

Data. Every organization has data. The question is whether the specific data this initiative depends on is structured, accessible, and clean enough to use. ServiceNow’s UAE finding — 77% citing data accuracy, access and management as a barrier — is the market version of this question.⁶

“We have a data lake” is not evidence. This illustration is hypothetical: “the fields this initiative needs are populated and consistent for 80% of records, with the remaining 20% concentrated in one legacy system we are migrating” is.

Infrastructure. Two things, not one. A deployment environment that meets the residency and security requirements the initiative will need — and the integration reality: can the system connect to what it has to act on, or does that require an integration project that has not been scoped? An AI build that assumes API access to a legacy system with none will surface that gap during build, regardless of how well the model is designed.

Workforce. Not “do we have AI talent.” Most enterprises correctly answer no, and correctly plan to hire or partner. The narrower question: once the vendor leaves, is there a role — filled or being filled — whose job is to operate it. Systems with no operational owner tend to degrade: no one watches for drift, no one maintains prompts or retraining triggers, and the system is ignored rather than decommissioned. That is the shelfware pattern above. Practitioner judgment.

Governance. Named owner for the system’s outputs. Defined approval process for changes. For anything touching regulated data, the specific sector sign-offs required before go-live. An initiative that is technically ready and governance-unscored risks a late-stage block from legal or compliance that should have been resolved in scoping. ServiceNow’s 16% — the share of UAE organizations with AI testing, auditing and risk-management processes — is the citable proxy for how thin this layer still is.⁶⁷

Each dimension scores against pre-defined criteria: not present; partially present with a defined gap; present and verified. A “partially present” data score comes with the remediation required and an estimate of effort to close it. The score converts into a plan. A single “70% ready” number does not.

The cheapest deliverable is the dated gap-list

The most valuable output is often the shortest: a dated list of what would have to be true before a build is worth starting, ranked by how much control the institution has over closing each gap.

A data-quality gap the organization can fix with an internal cleanup is a different blocker than a governance gap that depends on a regulator’s timeline. This list is not a roadmap for the AI system. It is a roadmap for becoming ready to build it. Shorter. More honest. It often reveals why a previous initiative underperformed: not that the AI was wrong, but that the organization built on a gap in one of these four dimensions that was visible in advance and was not scored.

Hypothetical: customer-service AI

Consider an enterprise assessing readiness for an AI-driven customer-service initiative. This example is hypothetical — including the 80% / four-week figures.

Data — partially present. Ticket categorization is inconsistent in recent history and complete only for the two most recent quarters. Remediable with a defined four-week reclassification effort.
Infrastructure — present and verified. The required CRM integration exists and meets residency requirements.
Workforce — not present. No internal role identified to operate the system post-deployment. Hiring not yet underway.
Governance — partially present. An accountable executive sponsor is named. No documented change-approval process.

The honest recommendation is not “not ready, wait” or “ready, proceed.” It is: proceed with the data remediation and workforce hiring as explicit, funded prerequisites running in parallel with initial build scoping; complete governance sign-off before production launch, not before build starts. That is a plan a sponsor can execute. A single aggregate score never produces one.

Not a one-time gate

Treating readiness as a gate passed once is a common mistake. Each of the four dimensions can degrade during a build: a planned hire falls through; the accountable sponsor changes role; a data pipeline that was clean at assessment drifts when source systems change. Institutions that reassess at defined milestones through the build catch degradation while it is still cheap to correct.

What 54east will defend in a briefing

Readiness is not a position on a spectrum. It is four evidenced scores — data, infrastructure, workforce, governance — each attached to a named gap and a plan to close it. The axes are standard. The cheapest document you can buy is the dated list of what would fail if the build started today.

54east runs this as a scoped, short engagement: four scores, the evidence behind them, a dated gap-list. Not a maturity-model poster. If you are not sure whether a planned AI initiative is actually ready to start, start with a briefing.

Notes

Aditya Challapally, Chris Pease, Ramesh Raskar and Pradyumna Chari, The GenAI Divide: State of AI in Business 2025, MIT NANDA. Coverage and quotations: The Register, 18 Aug 2025; Fortune syndication, “MIT report: 95% of generative AI pilots at companies are failing”. The report’s own line: 95% of organizations getting zero return; “the vast majority remain stuck with no measurable P&L impact.” Success for task-specific tools is defined as a marked and sustained productivity and/or P&L impact.

Secondary summaries of NANDA describe a roughly six-month post-pilot window for that P&L test. That clock is not a 54east measurement, and it is not “quietly unused after production.” See Governance AI, “Why AI projects fail”.

Gartner, “Why Half of GenAI Projects Fail”: at least 50% of generative AI projects abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value. Gartner’s earlier July 2024 forecast was 30% abandoned after PoC by end-2025: press release.

Oxford Insights, Government AI Readiness Index 2025 (updated January 2026; 195 countries; new six-pillar methodology). UAE rank 19; Public Sector Adoption 97.27. PDF: Government-AI-Readiness-Report-2025-1.pdf. An earlier December 2025 posting of the Index carried incorrect scores; this is the corrected file.

WAM, 5 Nov 2025, “Artificial Intelligence Readiness Index for Federal Entities launched during UAE Government Annual Meetings”. Minister Al Olama: 97% utilisation of AI tools across government entities.

ServiceNow / ThoughtLab, Enterprise AI Maturity Index 2026, UAE cut (100 executives). Spend +105% year on year; maturity 48/100; 77% cite inadequate data accuracy, access and management; 16% have AI testing, auditing and risk-management processes. ServiceNow UAE findings as carried by Zawya and Arabian Reseller, 19 Aug 2026.

Same ServiceNow 2026 UAE cut: Khaleej Times, 18 Aug 2026.

Cisco AI Readiness Index 2025: Strategy, Infrastructure, Data, Governance, Talent, Culture. Index landing page; 2025 report PDF.

Gartner, AI Maturity Model and AI Roadmap Toolkit: strategy, data, governance, engineering, operating model, culture, AI product/value.

Accenture and Carnegie Mellon Software Engineering Institute, AI Adoption Maturity Model, 8 June 2026. Eight dimensions: organizational strategy, workforce and culture, workflow re-engineering, risk and governance, data, engineering, operations, ecosystem. Accenture newsroom; SEI announcement.

Related whitepapers

Start with one decision.

If this paper describes the problem in front of you, the next step is a briefing with the architects.

Request a briefing