Glossary
AI data readiness
AI data readiness is an assessment of whether the data and infrastructure behind a specific AI project can be reached, trusted, lawfully used and delivered at the required rate. It is always judged against one use case, because a policy assistant and an image model need very different data.
Why is readiness judged per use case?
Programs that try to make all enterprise data ready before choosing a project rarely finish, because there is no single standard all data can meet. Readiness gains meaning only against one project's needs. An assistant answering questions on internal policy needs current documents with their permissions intact. A defect-detection model needs tens of thousands of labeled images. The same dataset can pass for one and fail for another.
Four dimensions of the assessment
| Dimension | Question asked | Typical gap |
|---|---|---|
| Access | Can the project reach the data at the scale and speed it needs? | Data locked in a legacy format or an application silo |
| Quality | Is it complete, consistent, deduplicated and labeled for the task? | Duplicates and missing labels |
| Governance | Do privacy, consent, licensing and retention rules allow this use? | Consent gathered for a different purpose |
| Infrastructure | Can storage, network and compute carry the volume and read rate? | Data reachable only by slow bulk export |
Running an assessment
- State the use case and the data it needs, in volume, freshness and format.
- Locate where that data lives, including the copies nobody listed.
- Test whether it can be read at the required speed and scale.
- Check metadata and lineage: source, version and owner for each set.
- Confirm with legal and security that the intended use is permitted.
Step four is where many assessments stall, because missing lineage cannot be rebuilt cheaply. That dependency is the link to AI data management, the continuous practice that makes readiness answerable at all. The infrastructure dimension is decided in AI data infrastructure.
Readiness beside its sibling terms
Readiness is a verdict, not a system. The AI data architecture and the AI data pipeline describe what exists. Readiness asks whether that existing estate supports one more project, and the answer for a retrieval AI knowledge base can differ from the answer for fine-tuning on the same documents.
Frequently asked questions
How often does AI data readiness need to be assessed?
For each new use case, and again when a source changes. A dataset that passed once can fail later if its permissions, freshness or structure change.
Is unstructured data harder to make ready?
Generally yes. Documents, images and recordings carry little built-in structure, so they need extraction, tagging and permission mapping before a project can rely on them.