The AI Readiness Problem Most Legal Teams Are Missing

AI is generating real interest across entity management teams. Yet workflows that appear ready for automation are still difficult to accelerate because the data beneath them is fragmented, inconsistent or incomplete.

There is no lack of use cases: KYC, entity data search, audit support, structure charts, filings and recurring reporting are strong workflows for AI because they are repetitive, information-heavy and hard to scale manually. The real constraint is control over data those workflows depend on.

Entity data often sits across different systems and owners. Before the business can rely on an answer, someone has to confirm if the data is current and reconcile competing versions. AI can accelerate parts of that process, but it cannot make fragmented data trustworthy on its own.

That is why the idea often summarized as “IA before AI,” or information architecture before AI, is becoming more relevant to governance operations. Before governance teams can rely on AI, the information architecture behind the work has to be accurate, connected, permission-aware and usable when the business needs to act.

AI readiness starts with control over entity data

When governance teams ask what AI can do for entity management, Athennian CEO Adrian Camara often starts with another question: “Where’s your data?” It is a simple question, but it changes the conversation.

It forces teams to ask whether the data they are giving AI is reliable, accessible and connected because without it, AI has little to build on. That is the difference between having governance information and having operational control over it.

Fragmented records turn AI outputs into review work

Governance teams have worked around scattered data for years. Experienced operators know which source to trust, which advisor to call and which document to check before a decision can move. AI does not have that institutional knowledge unless it is reflected in the data. If every AI output has to be verified, the work has not become more reliable. It has simply moved downstream.

That is why Caitlin Melchior, a partner in Paul, Weiss’s investment funds group, frames the issue as one of diagnosis before automation. Her instinct is to pull the problem apart before deciding where AI belongs.

Some steps may be suited to AI, such as extracting information, summarizing a document or preparing a response. Others may be slow because ownership is unclear, the authoritative system is not defined or the workflow has not been formalized.

Defined entity relationships determine whether AI has context

Adrian describes entity structure as “a context layer for AI.” Governance work does not run on isolated data points. It depends on the relationships between entities, owners, directors, officers, jurisdictions, authority, approvals and supporting records.

AI may help prepare a KYC response, generate a chart, summarize a document or route a request. But the entity model tells it which entity is involved, how that entity fits into the wider structure, what authority applies and which evidence supports the answer. This is why information architecture is more than a data-cleanup exercise. It connects the information the business needs to act.

Access and authority cannot be separated from AI

Governance data sits inside a permission structure. Finance, tax, treasury, compliance and legal teams may all rely on the same entity data, but they do not need the same level of access. Sensitive information needs to remain restricted and certain actions need review. AI has to operate within those boundaries. 

Adrian captured the point simply: “the permission matrix would like to have a conversation with the AI.” Permissions are not something to add after the AI use case has been proven. They are part of the information architecture the use case depends on. 

Start AI adoption where speed will not weaken control

“IA before AI” can sound like a call to fix every record before AI implementation begins. That is not the point. Teams do not need perfect data before they start. They need to know which workflows have a strong enough foundation to move faster and which ones would become riskier if AI accelerated them tomorrow.

Start with the work where speed, control and confidence matter most. Then run a focused workflow review:

1. Identify the workflow. Is it KYC, diligence, structure charts, authority checks, entity changes, filings or recurring reporting?
2. Map the source data. Where do the entity record, ownership data, supporting documents, approvals and authority details live?
3. Confirm the governing record. Which system or source should the business rely on when records conflict?
4. Check the evidence trail. Can the team trace the answer back to the document, approval or record that supports it?
5. Review permissions. Who can view the information, who can update it, who can approve it and who is allowed to act on it?
6. Decide whether AI should accelerate it. If the workflow has current data, clear ownership, connected evidence and defined permissions, AI can reduce effort. If not, AI will surface the gaps.

That is the role of information architecture before AI: an operating sequence that separates useful acceleration from expensive noise.