Law firms have bought AI. Few have changed how they work because of it. On 30 September, Valliance’s Tarek Nseir and Chris Lormor showed a room of law firm leaders what it takes to close that gap, including a live ontology demo on Palantir. The full recording is below.
The market is moving fast, and unevenly
The past six months have been busy for legal AI. Legora tripled its valuation to $5.55 billion. Harvey raised $550 million at a $15.6 billion valuation. OpenAI entered the market with Astra for Law. At the same time, Sullivan & Cromwell apologised to a federal judge for AI-fabricated citations, and the billable hour is under real pressure.
Kirkland & Ellis is spending $500 million building its own AI platform. A firm of that size could license every tool on the market. Kirkland is also investing in something no competitor can buy: its own knowledge, in a platform it owns.
'If everybody's using these tools in exactly the same way, what's the point of difference for the firm?' – Tarek Nseir, Valliance
The adoption gap sits between the lawyer and the firm
Adoption inside firms is slower. 74% of UK lawyers use AI for legal work at least weekly, according to LexisNexis. Only 29% of organisations have AI embedded in strategy and operations, and just 18% of professionals know whether their organisation tracks the return on its AI tools (Thomson Reuters Institute).
Individual lawyers have moved faster than their firms.
Our own interviews with lawyers bear this out. We showed it as a day in the life. A lawyer's day today is split between client work and judgement on one side, and a long tail of chasing replies, re-keying markups, pasting precedents into AI tools and writing time narratives on the other. Each AI tool helps with one task, so lawyers still carry the admin between them. That's time that could go on clients and judgement.
Your only advantage is your people. We call this your alpha.
Today, a firm's real edge lives in its people, in the house positions, the judgement calls and the history of what was agreed and why. Every competitor can rent the same tools. When a partner leaves, much of that knowledge leaves with them.
An ontology keeps that knowledge in the firm. It's one living map of the practice, linking matters, clients, documents and deals to the positions and decisions that today exist only in people's heads and inboxes. Workflows reach lawyers where they already work, in Word, Outlook and one workflow app. Models become interchangeable, so firms can keep the tools they like. The alpha stays with the firm, and it compounds with every matter.
What we showed
Chris walked through a working ontology on a fund formation scenario, using Claude in Word on top of a Palantir ontology.
'All the heavy lifting you're seeing is being done by the ontology, and not the agent itself.' – Chris Lormor, Valliance
Term sheet to first draft. A lawyer attaches a term sheet in Word, and the ontology drafts the LPA from the firm's house positions. Every clause is annotated with its reasoning and sources.
Departure and approval. When a lawyer moves a clause away from the house position, the change is routed to a partner to approve, and their reasoning is recorded.
Publish to obligations. Once approved, the LPA and side letters become live obligations the firm can track.
Diligence crossover. When the fund later makes an acquisition, the ontology flags issues from the fund's own commitments and from outside events, as well as from the data room. In the demo, it picked up investors whose co-investment rights apply above a set threshold.
The house position learns. When lawyers depart from a house position often enough, it is flagged for review. Once it is updated, the next draft reflects it.
At each step, the ontology knew something a standalone tool couldn't, because it holds what the firm has already concluded.
'The ontology isn't just a static model of our institutional knowledge. It has a feedback loop. It learns and evolves with the firm.' – Chris Lormor, Valliance
The demo took about two weeks to build, for a fictional firm. A firm's own ontology would be built around its practices and positions, which is what makes it the firm's alpha.
Why this matters for a global firm
Owning this layer gives a firm an edge competitors can't rent. It also protects the quality of the work as more AI output reaches clients, because every output is anchored to how the firm thinks and every departure from a house position is recorded and approved.
It matters for talent too. As AI takes on the groundwork junior lawyers used to learn from, the ontology gives them handrails to develop under the firm's oversight. And the better the models get, the more the firm's own record of positions, judgement and history is worth.
How you get there
Design for the whole firm, and start in one practice. We build in two complementary motions, and the steps alternate between them.
Value First builds the system. Choose an anchor practice where value, partner appetite and build potential line up. Build the ontology first. Put the first use case in lawyers' hands. Extend to adjacent practices. Own your ontology.
People First brings the firm with it. Build fluency in the partnership. Map how work really gets done. Set guardrails that give lawyers a safe, fast route to approval. Scale your power users. Drive adoption and measure it. Build a culture of shared expertise.
It can build in stages, practice by practice and partner by partner.
'A common mistake is to build an ontology that looks like the systems you're integrating with. An ontology is there for humans and agents to understand.' – Chris Lormor, Valliance
Talk to us
If the session raised questions specific to your firm, we'd be glad to talk them through one to one. Get in touch
Find out more about how we’re helping legal teams put AI to work.
FAQs
Do we need Kirkland's $500 million budget?
No. Kirkland's figure covers several years, platform licences, services and the time of its own lawyers. A global firm can make the same change for many multiples less, starting in one practice and extending from there.
How do we choose an AI partner for this?
Look for a partner that builds working software on your data from week one, holds no vendor allegiances, and ties its fees to the outcomes it creates. At Valliance, small senior teams do the work themselves, and most of our fees are paid when we create value.
How do you bring lawyers with you?
Through our People First motion, which runs alongside the build. It starts with fluency in the partnership, maps how work really gets done, sets guardrails that give lawyers a safe, fast route to approval, and scales what power users already do.
How is client confidentiality and privilege protected?
Access is governed object by object, so lawyers only see the matters they are entitled to see. The ontology runs inside the firm's own secure environment, and every departure from a house position is recorded with who approved it and why.
Is Palantir Foundry worth it for a law firm, and what does implementation involve?
Foundry gives the ontology security and governance built in, and what the firm builds stays in its own instance. Implementation starts with one workflow in one practice. We sit with lawyers, map the decisions they make, and model the ontology on how the business works rather than on the source systems. The demo took about two weeks to build.
We already use Harvey or Legora. Do we need to replace them?
No. The ontology sits underneath the tools a firm already uses. In the demo, Claude in Word did the drafting, but the context came from the ontology, so Harvey, Legora or another agent could sit in the same place. Most legal AI tools learn from public documents, and some practices, such as fund formation, have very few public examples to learn from.
What is an ontology, and how does it make a law firm's data AI-ready?
An ontology is a model of how the firm's work gets done. It links matters, clients, lawyers, documents and deals to the things that make the firm distinct, such as house positions, model clauses, approvals and departures. AI tools can then draw on what the firm has already concluded, instead of a generic public corpus.
What return should we expect from an ontology programme?
Each use case is measured against the baseline set at the start, such as hours per task and turnaround. The ontology shows up in the compounding. Each new use case should be faster and cheaper to deliver than the last, because it builds on what is already there.
