Law Firms Don’t Have an AI Problem, They Have a Data Problem

By Kevin Walker, CEO, Centari.

In recent months, several of the world’s leading law firms have announced investments in legal technology at a scale only few of us would have predicted two years ago. The media coverage has labeled this ‘AI adoption.’ But that misreads what’s happening.

These firms already have access to AI, and so does everyone else. They are not committing nine figures to buying more licenses to ubiquitous legal AI workflow tools. What they are paying for is a proprietary asset that the firm owns outright, and that simultaneously mitigates AI’s existential threat to the business of the firm, while preparing it to extract more value from AI than their competitors.

They are paying for data; decades’ worth of knowledge, turned into a structured and compounding record of the expertise and insights that command over $2,000 per billable hour in the market for top-tier legal services.

Firms compete on expertise, and expertise is built one deal at a time.

With legal AI products, including the ones firms license at scale, clients can now self-serve almost any piece of work product imaginable. If the practice of law were little more than drafting and workflows, law firm revenues would crater as most of this work would move in-house. Instead, revenues continue to expand, and some firms are seeing their profits spike.

When a client has a bet-the-farm transaction, they pay for expertise. They engage the attorneys who have seen their situation before and know how to navigate the maze of relationships, regulations, and risks to achieve the desired outcome.

That expertise accumulates across deals, not across documents, and the distinction matters a great deal.

A purchase agreement read on its own tells you what the parties agreed. It does not tell you that the seller gave up the survival period only after the buyer conceded the indemnity escrow in a prior redline. A credit agreement on its own will tell you the rates agreed to in that financing, but it cannot tell you that the maintenance covenant has loosened across the last forty financings in this sector. And a defined term renegotiated in the third amendment to a limited partnership agreement might change a covenant calculation in the original agreement.

Attorneys earn that often-used but hard-won title of ‘strategic advisor’ by seeing the pattern. And the pattern is exactly what’s hardest to see. It lives in fragments across drafts, side letters, exhibits, closing checklists, notes, and the memory of whoever was staffed on the deal, whether or not they still work at the firm.

The industry likes to call law a knowledge business. Unfortunately, most firms, and the technology they are adopting, still treat it as a document business.

Your tech investment decides whether you get a knowledge asset or an approximation.

The wave of ‘horizontal’, give-us-every-use-case products is real, and it’s correct for firms to adopt general-purpose tools for research, drafting, data extraction, and summarization. They belong in every firm that’s serious about innovation, and a firm that has not put AI in front of its lawyers is behind. None of what is said here is an argument against those tools. It is an argument about layers.

Today’s legal AI handles documents, where the lawyer’s workflow sits. The asset sits underneath those products: the structured, queryable record of the firm’s expertise that everything on top draws from. Reading a document and building that record are different problems, and the second is far harder than it looks.

Consider what it takes to get this part right. A general-purpose system reads text as text: a sequence of words frozen in a moment. Ask it to summarize a single agreement and it will do that well. Ask it to build a comprehensive profile of a closing set where the agreement, amendments, schedules, and side letters all reference and modify each other, and the real problem comes into view. Humans can spend days untangling how those pieces fit.

Reaching for the general-purpose tool here is the natural instinct, and the specific vendor matters less than the problem. Building the data layer a firm can rely on is fundamentally harder than reading a document, and any system a firm trusts to do it has to clear the same bar:

Does it carry a defined term across the amendment that changed it? Does it hold up not on one clean document but across the whole portfolio, extracting the same provision the same way every time rather than three different ways? Does it understand how the pieces of a transaction fit together, or is it just reading text?

Those questions separate a knowledge asset from an expensive approximation, and they are the right questions to ask of any vendor claiming to sell the answer.

A data asset you cannot trust is not a data asset. If the table says a borrower is in compliance when an amendment puts them in breach, every answer built on that record is wrong, and very well may be wrong at the exact moment the stakes are highest.

The firms making the largest commitments already understand this. They are investing in the specialized capability to own a competitive data layer underneath the tools everyone else is renting.

—

—

AI adoption is the right instinct, but the data is the unfinished part.

The firms committing real resources to capture their own knowledge are reading the moment correctly. Knowledge is the asset, and the instinct to capture it at scale is sound. But this instinct only pays off if the data can be trusted, and that trust depends entirely on whether the system used to create it understands the deal or merely reads the text.

The most valuable thing a firm owns is its accumulated and ever-evolving knowledge of the market and of the strategies to navigate the complexity of modern business. For the first time, that knowledge can live somewhere other than fragmented records and the people who happened to be in the room.

A firm that has command of its data can tell a client what is genuinely market on a given term right now, not what was market last year. It can sit across from a repeat counterparty already knowing how they moved on the last ten deals, and price a covenant against how it has loosened across the sector.

Those are the strategic insights clients pay top rate for, and the firm that can produce them on demand wins the mandate over the one still reconstructing them from memory or reaching to get there with genera- purpose legal AI.

The firms that get this part right will run the best available AI models on top of a record they can actually trust. The models will keep improving, and they will keep getting cheaper and more available to every competitor a firm has, and eventually to that firm’s own clients.

The commoditization of legal AI is not something any firm can defend, but the data underneath is the part only the firm can own. Every deal it touches makes that data sharper, and makes the firm that owns it harder to replace.

To learn more about how Centari’s deal intelligence platform can help you, please see here.

—

[ This is a sponsored thought leadership article by Centari for Artificial Lawyer. ]