Data Readiness Checklist for Enterprise AI Transformation
The short answer
Your data is ready for an AI use case when the specific sources it needs are identified, accessible through governed channels, documented with owners and lineage, good enough in quality for the task, and protected by permissions that carry through to the AI system. Readiness is judged per use case — not as a whole-estate cleanup before any AI starts.
Why 'fix all the data first' fails
Enterprise-wide data programs can run for years before delivering value. AI programs that wait for them stall. A better pattern is to pick a use case, assess only the data it needs, and fix those gaps — building reusable foundations one use case at a time.
Six dimensions of data readiness
| Dimension | Ready looks like |
|---|---|
| Availability | Required sources identified and reachable via approved integrations |
| Ownership | Each source has a named data owner who can approve use |
| Quality | Completeness and accuracy measured against what the task needs |
| Lineage | You can trace where data came from and how it was transformed |
| Access and privacy | Permissions and sensitivity labels carry through to the AI system |
| Freshness | Update frequency matches the decision the AI supports |
The checklist
- List the exact data sources the use case needs — and those it must not touch.
- Confirm a data owner has approved the AI use of each source.
- Measure quality on a sample against task-specific criteria.
- Document lineage for any data feeding a regulated decision or report.
- Verify sensitivity classification and that access controls are enforced end to end.
- Agree refresh frequency and monitoring for upstream changes.
- Record gaps and decide: fix now, work around, or defer the use case.
Connect data and execution
Data readiness is one half of what Curate calls the Two-Body Problem. Even ready data stalls without an owner, success metric, and governance for the use case — which is why we assess both together.
Frequently asked questions
How do I know if my data is ready for AI?
Assess per use case: are the needed sources available, owned, of sufficient quality, traceable, properly permissioned, and fresh enough? Gaps in any of these are the work to do first.
Do we need a data lake or warehouse before starting AI?
Not necessarily. Many use cases can start with governed access to a few well-understood sources. Central platforms help scale, but should not block the first use cases.
What data problems most often block AI projects?
Unclear ownership, missing access approvals, inconsistent definitions, and permissions that do not carry into the AI system are common blockers, often more than raw data quality.
Is data readiness different for generative AI?
Generative AI relies heavily on unstructured content such as documents and emails, so document permissions, versioning, and sensitivity labels matter as much as structured data quality.
Official sources
- NIST AI Risk Management Framework (AI RMF 1.0)
- NIST AI 100-1: AI RMF 1.0 (PDF)
- NIST AI 600-1: Generative AI Profile