01
Source and Ownership
Where is the data, who owns it, and who can approve changes?
AI fails when no one owns source truth or remediation.
Evidence: System inventory, owner map, steward list, source-of-record decision.
The Readiness System: Data Readiness
Is your data trusted enough to power AI?
Data Readiness is 20 percent of the AIR APAC readiness score — and the component most likely to block AI from working. Organisations discover this only after training, when their copilots hallucinate, their agents retrieve the wrong document, and their dashboards show numbers no one trusts.
Documents that humans can read are not automatically AI-usable. Data is AI-ready when it can become trustworthy evidence inside a decision loop: sourceable, permissioned, contextual, retrievable, and cited.
AIR APAC Methodology, Data Readiness component (20 percent of the readiness score)
Seven scoring criteria
What Data Readiness assesses.
02
Quality and Consistency
Is the data accurate, complete, deduplicated, current, and internally consistent?
Bad inputs produce unreliable retrieval, analytics, recommendations, and automation.
Evidence: Sample profiling, duplicate rate, missing-field rate, update age.
03
Permissions and Privacy
Who is allowed to access what, for which use case, under which policy?
RAG and agents become dangerous when access control is flattened.
Evidence: Classification map, PDPA and privacy treatment, permission matrix.
04
Provenance and Lineage
Can the organisation prove where data came from, when it changed, and how it was transformed?
AI outputs need receipts, auditability, and defensibility.
Evidence: Lineage notes, source URLs, file hashes, method versions, artifact paths.
05
Semantic and Context Fitness
Does the data preserve local language, terms, business meaning, hierarchy, and APAC or regulatory context?
Generic models lose meaning when context is stripped.
Evidence: Terminology map, language coverage, local-market variants, decision-trace examples.
06
AI Retrieval and Agent Readiness
Can the data be chunked, embedded, retrieved, cited, and used by an AI workflow safely?
Documents that humans can read are not automatically AI-usable.
Evidence: Chunking test, retrieval evaluation, citation test, stale-document test.
07
Connector Readiness
Can the data become a repeatable evidence packet inside a decision loop?
Preparedness should feed decisions and remeasurement, not static reports.
Evidence: Sample evidence packet, confidence grade, evidence state, provenance record.
Posture logic
Every assessment ends in a posture, not a scorecard.
Sufficient for a controlled AI use case with known limits.
Viable after named fixes. Do not scale broadly.
Major blocker. Funding AI tooling now would create waste or risk.
Hard caps
Five conditions that block a READY posture.
- No data owner: cannot be READY.
- No permission model for sensitive data: cannot be READY.
- No provenance for critical source data: cannot be READY.
- No update or freshness path for operational data: cannot be READY for live AI use.
- No evaluation set: cannot be READY for production RAG or agent deployment.
Evidence
Where Data Readiness meets the real world.
Asymmetric Visibility
The AI Readiness Pulse documents the visibility gap that Data Readiness exists to close.
Read the PulseTPO and Destination Intelligence
AIR APAC's tourism and trade promotion work includes source-readiness scoring, official-source share measurement, and evidence-receipt generation. This is live Data Readiness work in a public-information domain.
See the TPO collaborationThe APAC Data Desert
The Q1 2026 Index found that 70% of mid-market APAC companies do not produce enough publicly observable signal to be assessed with confidence. Much of that gap traces back to data: fragmented sources, unclear ownership, no provenance, and no retrieval strategy.
See the findingsBefore you fund an AI pilot, test the data.
The AI Use-Case Preparedness Audit assesses whether your data can power AI, before you commit capital.
Explore the Preparedness Audit