Also called data readiness for AI, preparing unstructured data for AI, or making data decision-ready.

The order of operations is easy to get backwards. Teams buy or build a model and then discover the data it needs is scattered, inconsistent, unlabeled, and locked in formats nothing can read. The capability stalls not because the model is weak but because the data underneath it was never made ready. Readiness is the precondition, not the afterthought.

That makes this the foundation the rest of this hub is built on: you cannot resolve entities, fuse sources, or reason over a knowledge graph if the underlying data was never made visible, understandable, and linked in the first place. Readiness is where entity resolution and fusion become possible.

Key takeaways

• What it is: turning raw, fragmented, mostly unstructured data into data AI and analysts can reason over, measured by the DoD's VAULTIS goals, without losing provenance or classification.
• Why it matters: the DoD has made data a strategic asset, and oversight has found that having usable data is one of the hardest parts of fielding AI, harder than the model itself.
• The unstructured problem: most mission data is unstructured text, imagery, and logs that traditional pipelines were not built to turn into something an AI can use.
• What to require: readiness that preserves provenance and classification, works on unstructured data, and keeps the result government-owned rather than trapped in a vendor's format.
• The payoff: data that is actually ready for resolution, fusion, and reasoning, with a line back to every source so the output can be defended.

Why Most Mission Data Never Reaches a Decision

The problem is not a shortage of data; it is that most of it never becomes usable. Mission data arrives overwhelmingly as unstructured content, reports, message traffic, scanned documents, imagery, and sensor logs, in incompatible formats that were never designed to line up. It sits in systems that cannot talk to each other, and the work of making it usable is slow, manual, and rarely finished.

Oversight has put a sharp point on this. The U.S. Government Accountability Office found that having usable data available is one of the central challenges in fielding AI for defense, offering the concrete example that teaching an AI to detect an adversary's submarines requires gathering and labeling many images first (GAO, Artificial Intelligence: Status of Developing and Acquiring Capabilities for Weapon Systems, GAO-22-104765, 2022). The model is not the bottleneck; the ready, labeled, trustworthy data is.

The Department has named the goal directly. Its data strategy declares the ambition of becoming "a data-centric organization that uses data at speed and scale," insists that leaders "treat data as a weapon system," and sets seven goals, known as VAULTIS, for what ready data must be (DoD, DoD Data Strategy, 2020). The strategy is explicit that data has to be high quality, trustworthy, and linked to be useful for decisions. That is a readiness requirement, expressed at the top of the Department.

For the mission the payoff is the difference between data you have and data you can use: content that is ready for an analyst and an AI to reason over, with provenance intact, instead of a growing archive no decision ever reaches.

What "AI-Ready" Means: The VAULTIS Test

The DoD's own standard for ready data is the VAULTIS acronym. It is a useful checklist because it names what "ready" actually requires, each goal a test the data either passes or fails:

• Visible. The data can be found; consumers can locate what exists.
• Accessible. The data can be retrieved by those authorized to use it.
• Understandable. The data carries enough description of its content, context, and applicability to be interpreted correctly.
• Linked. Related data connects through its innate relationships, rather than sitting in isolation.
• Trustworthy. Consumers can rely on the data for a decision, with its quality and lineage known.
• Interoperable. Producers and consumers share a common representation so the data means the same thing across systems.
• Secure. The data is protected from unauthorized use and manipulation throughout.

The point of VAULTIS is that "AI-ready" is not one property but several, and unstructured mission data usually fails most of them at the start: invisible in a silo, unlabeled, unlinked, and in a format nothing else understands. Readiness is the work of moving data from failing those tests to passing them, without sacrificing the "secure" goal to achieve the others.

Structured vs. Unstructured Data for AI

Bottom line: structured data is already in rows and fields a machine can read; unstructured data, the majority of mission content, is text, imagery, and logs that must first be read, extracted, and organized before any AI can use it. The hard, expensive part of AI readiness is almost always the unstructured half.

Dimension	Structured data	Unstructured data
Form	Rows, fields, defined schema	Free text, documents, imagery, audio, sensor logs
Readiness out of the box	Largely machine-readable already	Must be read, extracted, and organized first
Share of mission data	The minority	The majority
What it takes to use	Normalization and linking	Extraction of entities and meaning, then normalization and linking
Typical failure	Schema mismatch across systems	Never processed at all; sits unread
Role of AI	Consumes it directly	AI both helps prepare it (reading, extraction) and later reasons over it

The takeaway is that a readiness effort that only handles structured data has solved the easy fraction of the problem. The content where the mission's hardest questions actually live, the reporting and imagery, is unstructured, and making it ready is the work that separates a usable capability from a demo.

From Raw Files to Decision-Ready: How Data Is Made AI-Ready

Readiness is a pipeline, not a one-time cleanup. In a mission setting the stages are:

• Ingest in place. Connect to data where it lives across many systems, including unstructured repositories, without forcing a migration into one store.
• Read the unstructured content. Extract the entities, relationships, and meaning buried in text, documents, and imagery, so content becomes something a machine can work with.
• Normalize and structure. Put heterogeneous inputs into a common representation so they can be compared, without flattening away what makes each source distinct.
• Link and resolve. Connect related data and resolve records that describe the same real-world entity, turning isolated files into a linked picture.
• Attach provenance and classification. Carry every element's source and classification through the pipeline, so readiness never comes at the cost of traceability or security.
• Govern and maintain. Keep the data ready as new content arrives, rather than treating readiness as a project that ends.

The load-bearing stage is reading the unstructured content, because that is where most mission data is stuck. Everything downstream, resolution, fusion, GraphRAG, depends on it, and it is also where the "trustworthy" and "secure" VAULTIS goals are most easily lost if the work strips provenance or pools data across classification boundaries.

AI-Ready Data vs. Related Terms

These terms show up together in data-readiness discussions and are easy to blur. They name different things. The table separates them.

Term	What it is	Relationship to AI-ready data
ETL / ELT	Extract, transform, load pipelines that move and reshape structured data	A mechanism within readiness; necessary for structured data but insufficient for unstructured content
Data labeling	Annotating data so a model can learn from it	One readiness task, prominent for training; AI-ready data is the broader state, not only labeled training sets
Data catalog	An inventory that makes data findable and described	Supports the "visible" and "understandable" VAULTIS goals; a catalog alone does not make data linked or trustworthy
Data readiness levels	A maturity scale for how ready data is for AI	A way to measure readiness; VAULTIS describes the goals, readiness levels grade progress toward them

What Breaks Data Readiness in a Mission Context

• The unstructured majority. Most mission content is text and imagery that traditional, structured-data pipelines never turn into something usable.
• Provenance loss. Cleanup that strips the line back to source produces data that is convenient but no longer defensible.
• Classification constraints. Making data visible and linked cannot mean pooling across classification or releasability boundaries that forbid it.
• Format and system sprawl. Data sits in many incompatible systems, and a readiness effort that demands a rip-and-replace migration stalls.
• Staleness. Readiness is not a one-time cleanup; new data arrives constantly, and a picture that is not maintained decays.
• Vendor formats. Data made "ready" only inside a vendor's proprietary model is ready for that vendor, not for the mission.

Why AI-Ready Data Has to Stay Government-Owned

Readiness changes who can use a force's data, which makes ownership a first-order question. If the only "ready" copy of mission data lives in a vendor's proprietary format, the government has made its data usable on terms it does not control. The table contrasts the two models against what a mission actually needs.

Consideration	Typical commercial platform	Government-owned reasoning infrastructure
Control of the ready data	Held in the vendor's proprietary format	Customer keeps the ready, linked data in open form
Relationship to existing systems	Often a migration onto a new platform	Readies data in place; no rip-and-replace
Unstructured coverage	Often tuned for structured data	Reads and structures unstructured text and imagery
Provenance and classification	Varies; often lost in cleanup	Preserved through the pipeline
Portability of the result	Locked to the vendor's model	Portable; feeds resolution, fusion, and reasoning
Governance	The vendor's cadence	The customer governs and maintains readiness

Government-owned does not mean the government does all the work itself or owns a vendor's underlying intellectual property. It means the ready data, and the logic that made it ready, stay under the customer's control and in a form it can reuse, rather than being usable only inside a proprietary platform. Whether a specific deployment is government-owned (GOTS) or commercial (COTS) depends on the system the customer installs and purchases; when the asset being made ready is the force's own data, the government-owned model is the one to evaluate.

What to Require to Make Mission Data AI-Ready

The comparison explains why ownership and unstructured coverage matter; the checklist is what to require in an evaluation.

1. Handles unstructured data. Reads and structures text, documents, and imagery, not only clean tabular data.
2. Readies data in place. Connects to data where it lives without forcing a single repository or a rip-and-replace migration.
3. Preserves provenance. Every element keeps its line back to source through the pipeline, so readiness never costs traceability.
4. Respects classification. Makes data visible and linked without pooling across classification or releasability boundaries.
5. Measurable against VAULTIS. Moves data demonstrably toward visible, accessible, understandable, linked, trustworthy, interoperable, and secure.
6. Government-owned and portable. The ready data stays in open form the customer controls and can reuse, not locked to a vendor.
7. Maintained, not one-time. Keeps data ready as new content arrives, rather than decaying after a single cleanup.

Evaluating a capability? The seven requirements above are the backbone of a data-readiness evaluation you can score vendors against. Bring them to a scoping call and we will walk each one against your environment: request a technical walkthrough.

How Torch.AI Makes Mission Data AI-Ready

Torch.AI builds reasoning infrastructure the customer can own and govern, offered as a government-owned (GOTS) deployment when a mission requires it. It readies data where it lives rather than demanding a migration. ORCUS ingests and normalizes fragmented, multi-source data, including the unstructured content that defeats traditional pipelines, and NEXUS reads that unstructured text and imagery to extract the entities, relationships, and meaning inside it. The result is data that is visible, understandable, linked, and trustworthy enough for an analyst and for AI to reason over, with provenance and classification carried through so nothing is made ready at the cost of being defensible. You can see how this is packaged as a capability on the Torch.AI software page.

Because the data is readied in place and kept in open, owned form rather than a proprietary model, it stays portable: the same ready data feeds entity resolution, fusion, and GraphRAG instead of being trapped where it was prepared. This is the systems of record versus systems of reason distinction at the center of Torch.AI's approach: readiness is the first step of a system of reason that sits on top of the customer's systems of record, leaving the authoritative sources intact.

Torch.AI's approach is built for the conditions this page describes: unstructured, multi-source data; provenance and classification preserved through the pipeline; readiness measured against the VAULTIS goals; and a result the customer owns and can reuse across the mission.

For evaluators scoping a capability, see how Torch.AI readies unstructured mission data and turns it into resolved, fused, reasoning-ready data on the software page, or request a technical walkthrough and we will run it against a sample of your own unstructured data, with provenance traced end to end.

Sources

• U.S. Department of Defense, DoD Data Strategy (2020), media.defense.gov - makes data a strategic asset and sets the seven VAULTIS goals (visible, accessible, understandable, linked, trustworthy, interoperable, secure) for ready data.
• U.S. Government Accountability Office, Artificial Intelligence: Status of Developing and Acquiring Capabilities for Weapon Systems, GAO-22-104765 (2022), gao.gov - finds that having usable, labeled data is a central challenge in fielding defense AI, harder than the model itself.

Frequently Asked Questions

What does it mean for data to be AI-ready? It means the data is usable by AI and analysts: findable, retrievable, understandable, linked, trustworthy, interoperable, and secure, which is the DoD's VAULTIS standard, with provenance and classification intact. Readiness is the precondition for any AI built on top of the data to be trustworthy.

What is the difference between structured and unstructured data for AI? Structured data is already in rows and fields a machine can read; unstructured data, the majority of mission content, is text, imagery, and logs that must first be read and organized before AI can use it. The hard part of readiness is almost always the unstructured half. See the structured vs. unstructured table.

Why is unstructured data so hard to make AI-ready? Because it has to be read and interpreted before it can be used: entities and meaning extracted from prose and imagery, then normalized and linked, all while preserving provenance and classification. Traditional structured-data pipelines (ETL) were not built for that.

Is making data AI-ready the same as data labeling? No. Data labeling is one readiness task, important for training models, but AI-ready data is the broader state in which data is visible, linked, trustworthy, and usable, not only annotated training sets.

Does making data AI-ready require moving it into one database? It should not. A mission-grade approach readies data in place across existing systems, preserving provenance and classification, rather than forcing a rip-and-replace migration into a single store.

What does government-owned AI-ready data mean? It is a model (GOTS-deployable) in which the ready, linked data and the logic that made it ready stay under the customer's control and in open form it can reuse, rather than being usable only inside a vendor's proprietary platform. The same capability can also be delivered commercially (COTS); which applies depends on the system the customer installs and purchases.