What a Legal Ontology Actually Means, and Why it Matters

We've spent the last several months building legal ontologies alongside lawyers, working through real matters, real data, and the decisions that shape how a firm operates.

A legal ontology will make your precedent searches faster, better, and cheaper, but the value should not stop there. A real ontology’s value is to help a firm make better decisions.

A firm's alpha is its judgment. An ontology is what allows that judgment to be captured, applied, and compounded to establish data sovereignty and generate more alpha for your firm.

Meanwhile, "ontology" has entered the lexicon of almost every legal AI release or RFP. Ontologies significantly predate the development of AI, but the term is now being molded to fit the world of models, agents, and tokens. 

Some define an ontology as a way to "distill" knowledge, condensing your entire complicated world into a graph or summary that fits within a context window. These new definitions already miss the big picture of what makes an ontology valuable, especially to a law firm.

For a firm, the real value is a transactional ontology built matter by matter, client by client. One that organizes the messy reality of your world and reflects how the firm actually operates, the decisions its lawyers make, and the judgment behind them.

Ontology is a compounding asset

Ontologies cannot simply be synthesized by a model or regurgitated from a set of data or documents. They are built on the edge, with lawyers sitting next to engineers, across countless rounds of feedback and testing.

They convert messy data, concepts, and workflows into structure and organization.

No two ontologies are alike, although they may bear many resemblances. A public ontology, or even one built for the same practice area at another firm, will not capture the particular way your firm operates.

The more of a firm's work an ontology connects, the more useful it becomes.

Ontologies within a firm for one use case or another connect to each other and stack up into a firm-wide view. You start small with a single pilot, and then it layers and scales, so clients look like other clients, with similar documents, relationships, and patterns.

An ontology, when properly connected, touches everything and becomes "part of the brain" of the firm. The point of all this is better decision-making at scale, not better lookup.

The ontology exists so that a person, or increasingly an agent, can act on the organization's data inside the organization's rules, and so that every action can add more value.

Once that model exists, building on top of it is fast. Non-technical people can define rules and automations in minutes because the hard questions have already been answered: What are the things? How do they relate? Who decides what about them?

That is how an ontology rewards you. Define your world up front, then get compounding value you can own and build upon however you want.

What a legal ontology is, and what it is not

Let's translate some of these concepts to a law firm.

A firm's alpha is not its access to case law or its ability to find the right precedent quickly. Every firm can find things.

A firm's alpha is the thousands of decisions its lawyers have made and the judgment behind them: which position to open with, which fallback to accept, when to walk, who signs off on the exception.

That judgment is mostly tacit, mostly in partners' heads, and mostly lost the moment the deal closes. The firm's ontology is the map of that messy reality.

It's the deal data and financials, the counterparties, the LPs, what the leverage looked like, why the indemnity cap landed where it did. 

And critically, it includes the decisions themselves, so that the next lawyer facing the same question starts from what the firm already decided rather than from a blank page and a search box.

You should be able to open up the ontology in your browser and recognize what you're looking at, because it's the language you use every day.

This is also where some important distinctions emerge between different approaches being described as legal ontology.

An ontology of law

An "ontology of law" built from public sources is a reference work. It can be useful, but it's also the same for every firm that buys it.

It cannot contain the thing that makes one firm's advice worth more than another's, because that thing is not public. It helps you find the law but knows nothing about what your firm decided to do with it.

It's a legal database with better structure, which is fine. But it's not your ontology.

A context graph or knowledge layer

A context graph or knowledge layer, some machine-built graph of document, matter, party, and version relationships for an LLM to use, gets closer to your firm's data.

That's useful progress. But there's a difference between knowing that two documents are connected and understanding why the relationship matters.

You can end up with a graph that nobody at the firm designed, and nobody at the firm can read. Ask a partner to open it and explain why two drafts are linked. Can they tell you what changed, who made the decision, and why?

If not, you've captured that the documents relate. You haven't captured what the firm decided or the judgment behind it. It's context, detached from judgment.

Both approaches have their uses. But if the goal is to capture and compound a firm's alpha, I think we should hold legal ontology to a higher standard.

What actually makes a real ontology valuable?

Working through pilots and building ontologies sitting next to lawyers, here's how I've come to judge an ontology.

It belongs to the firm

Your firm's alpha shouldn't become someone else's product.

Vendors and labs naturally want to improve their own products so they can sell them to everyone else. But helping your firm beat its competition by building its own unique and proprietary asset is not necessarily aligned with that incentive structure.

Your firm's ontology should be a portable, readable asset the firm owns, independent of any model or application vendor, including us.

If the ontology cannot leave with the firm,you should ask if it's really an asset or a cleverly disguised lock-in mechanism.

It is efficient to run

Without structure, every request has to rediscover or remap the firm's knowledge from scratch.

The agent re-reads the precedent library, re-infers the relationships, re-derives the positions, and burns tokens in the process. And after all that work, the output may be barely different from what any competitor with the same model and the same documents would produce.

An ontology front-loads that work once, as structure, so outputs can be more precise and deterministic because they're grounded in the firm's data rather than the model having to reconstruct everything.

Our initial benchmarks using a real customer's ontology are at least 5x more efficient at using tokens than a context graph approach.

That matters as firms start thinking seriously about the cost of running AI at scale. But it also means the system spends less time rediscovering what the firm already knows.

It compounds

An ontology isn't a summary of what the firm knows today. It needs to capture what happens next.

As lawyers and agents work, new decisions and outcomes can be written back into the ontology, adding to the firm's accumulated knowledge.

Agent memories or skills can rot over time and become outdated. Ultimately, they're preferences and instructions.

An ontology evolves with the firm's reality. It's a world model, not an attempt to codify your complicated world as a skill, workflow, or README.

The more the firm uses it, the more valuable that underlying structure can become.

It evolves

Firms change.

A new practice group, a new deal structure, a new regulatory regime, or a new source of data that isn't a document at all, like a deal tracker, billing system, or conflicts database, will sometimes require changing the structure itself, not just appending nodes.

An ontology needs to accommodate that.

It is built by people

Real ontologies cannot simply be synthesized.

They are extracted by forward-deployed engineers and legal engineers working on the ground alongside the firm's lawyers, naming the object types and relationships, resolving the bizarre data types like billing spreadsheets and handwritten PDFs, and deciding what exactly a "position" is.

Machines help enormously with the volume, but they cannot decide what matters. That takes people who understand the work, and a lot of back-and-forth to get the structure right.

Why this matters now

Models are converging. Your competitors already have access to the same models and vendors as you do. What they won't have is your firm's alpha.

They won't have the thousands of decisions your lawyers have made, the reasoning behind those decisions, and the judgment developed through years of working with clients, unless you make that knowledge available to them.

That is what a firm's ontology is built to capture.

A structured, governed, portable record of the firm's judgment that compounds and grows with every matter, negotiation, and decision. One that the firm owns and can build upon however it wants.

That’s what ontology boils down to: Retrieval is a commodity. Your judgment is not.