Enterprise AI now runs on documents, email, tickets, contracts, images and recordings, yet most organizations cannot say what that content holds, who owns it or who can see it. In a 2026 BARC survey of 225 data and AI leaders, 70% said less than half of their unstructured data is discoverable and usable for AI, and Gartner predicts that through 2026 organizations will abandon 60% of AI projects not supported by AI-ready data.
This paper turns that gap into a scored instrument: ten dimensions, a 0-3 rubric for each, a sampling plan and a four-level readiness model. The evidence is analyst research, industry surveys, vendor documentation and US regulation.
It describes the market, not your estate, and the rubric is a method to apply, not a measured result.
The whitepaper sets out a working method, not a survey result. It scores each repository on ten dimensions, samples content in a way a small team can finish, and converts scores into readiness levels that decide which AI use cases go first. Each part produces one document a sponsor can review.
scored 0-3
Inventory, parse-ability, quality, metadata, ownership, access control, sensitive data, lineage, storage cost and tooling are each scored per repository, from 0 for unknown to 3 for controlled, measured and automated. Every score needs evidence such as a report export, a sample result or a policy document, and a score without evidence drops a level. The artifact is a scorecard for each repository with the evidence linked beside every number.
Cheap metadata checks run on the whole population, while parsing and human review run on stratified samples drawn by content type and age band. High-risk areas such as HR folders and claims content are oversampled, golden documents are added by judgment, and a slice is double-reviewed for consistency. The artifact is a recorded sampling frame with parse and quality results by stratum.
Dimension scores sum to a maximum of 30 and map to four levels: Unmapped, Inventoried, Governed and AI-ready. Gates on access control and sensitive data stop a repository from reaching Governed while either is weak, so a high average cannot hide a blocking risk. The artifact is a readiness level per repository and a use-case ranking set by each use case's weakest repository.
Shortlist the use cases first, then the three to eight repositories each one needs. Get legal sign-off and read-only, logged scanner access before any content is touched.
Run full metadata, sharing and sensitive-data scans across each repository, then draw stratified samples, parse them with at least two candidate tools and have reviewers grade accuracy and currency.
Rate all ten dimensions from 0 to 3 with linked evidence. Treat a 0 or 1 on access control or sensitive data as a blocker that caps the readiness level.
Pilot use cases whose content already passes the gates, fund targeted remediation and a rescore date for high-value ones that fail a gate, and defer or stop the rest.
It is a time-boxed audit that scores documents, email, tickets, contracts, images and recordings on ten dimensions, including parse-ability, quality, metadata, ownership, permissions, sensitive data and lineage. Each repository gets a score with evidence behind it.
The result shows which content can safely feed a given AI use case and what has to be fixed before a project starts, rather than after a model has already indexed it.
Start from the AI use cases and the repositories they need, not from the whole estate. Run metadata scans on the full population for inventory, sharing and age, and scan for sensitive data.
Then stratify content by type and age, sample within each stratum, parse samples with candidate tools and have reviewers grade them. Score each dimension from 0 to 3 and apply gates for permissions and sensitive data.
The plan in this whitepaper is built to run in four to six weeks for three to eight repositories. Week one covers scope and approvals, the middle weeks cover metadata scans, sampling and review, and the final week covers scoring and use-case ranking.
Smaller estates with fast approvals can compress it to four weeks. Most elapsed time goes to approvals and scanner access, not analysis.
The whitepaper does not give a price, because cost depends on the number of repositories, approval speed and the tools you already license. It recommends using trial or existing licenses during the assessment and choosing production tools after the scores are in.
For context on the cost of skipping it, IBM's 2025 report found shadow AI breaches added about $670,000 to the average breach cost.
Teams usually combine three tool families. DSPM tools such as Microsoft Purview, BigID or Varonis handle discovery and exposure. IDP and OCR services such as Azure AI Document Intelligence, Amazon Textract and Google Document AI, plus open source parsers such as Unstructured, test machine readability.
Records management and eDiscovery tools handle retention and legal holds. The paper advises running your own sample through at least two parsers.
They set hard limits on what an AI pipeline may touch. HIPAA requires covered entities to keep required documentation for six years, PCI DSS v4.0.1 requires a check at least every three months that expired account data has been deleted, and Federal Rule 37(e) allows court measures when preserved information is lost.
Label regulated data before indexing and keep held content out of any process that alters or deletes it.
You get ten scoring rubrics, a week-by-week plan, a sampling method, a 14-item checklist, a readiness level model, a tool comparison and use-case decision rules. The statistics come from IDC, Gartner, BARC, the Cloud Security Alliance and IBM, and describe organizations in general.
None of it measures your estate, and the rubric and levels are a recommended method rather than benchmark data.
Drop your details and we'll send Unstructured Data Readiness Assessment straight to your inbox - no spam, unsubscribe anytime.
Bring the AI use cases on your shortlist and the repositories they would draw from. We will work through how the ten dimensions and readiness gates apply to that content, and what an assessment would need to cover. SECTION 7 - FAQ - 5 to 8 questions
Book a readiness session