AI strategy for a law firm.

Five firms today, five firms with the same AI. The bars are illustrative, not measured. On the right, every firm that takes the AI stands on the same base, and the base is taller than the old differences between them. What sits on top is no longer what the firm had before. It is what the firm built around the AI.

That is why the order changes. Firm D leads today and does not tomorrow. Firm C, mid-table today, built more around the AI and leads. AI does not just narrow the gap. It can reorder the market. The client rises too, and can do more of the work alone. And Firm B, which did not adopt AI, is behind everyone.

McKinsey (opens in a new tab) describes the same shift: general-purpose legal AI turning into a commodity, with the advantage shifting to what a firm owns: its proprietary data and the way it works.

The client's legal workload in two bars, drawn as shapes, not measurements. Today, roughly half of a corporate legal budget already goes to outside counsel, according to the Association of Corporate Counsel's 2023 benchmark (opens in a new tab) (53% internal, 47% external). Once the client's own legal team has a good AI, the routine share moves in-house and the firm is left with the work that needs judgment, risk-taking and someone to be accountable.

That last share is not small, and it does not shrink. It is where the price holds.

The answer to a client that does more itself is not to do the same work cheaper. It is to offer what the client could not buy before: better quality, and services built on the firm's own expertise. Thomson Reuters' 2026 survey (opens in a new tab) puts the gap in numbers: 77% of clients want AI-enabled quality improvements, 5% get them from most of their firms. What those services look like is section 06.

The pressure is already visible. According to the Financial Times (opens in a new tab), Goldman Sachs asks its law firms to quantify their AI savings and uses the number in fee negotiations, and Citi asks firms bidding for its work to disclose the same. In Thomson Reuters' 2026 survey (opens in a new tab), 71% of in-house legal professionals expect their outside firms to change their commercial model as AI use grows.

Clients who do more themselves also expect the AI efficiency to show in the price of what they still buy. That is the decision in section 07.

Three layers. At the bottom, the AI anyone can buy: the models (Claude, GPT, Gemini) and the legal platforms built on them (Harvey, Legora). Every firm and every client can buy the same, and the model vendors keep adding what the platforms sell today. Nothing in this layer separates one firm from another.

In the middle, what only this firm has: its knowledge, its workflows and playbooks, and its own way of checking whether the AI got it right. On top, the legal services the client actually buys.

The principle: the bought layer can be swapped. The middle layer should stay with the firm wherever possible. Harvey and Legora also offer to hold the firm's workflows and knowledge. Whether that layer should live inside a vendor is the question in section 05.

A simple test. If the firm dropped its AI vendor tomorrow, what would be left? If the answer is "our people know how to use AI", the firm owns nothing yet. If it is playbooks, workflows, evaluations and a knowledge base that work with the next model too, the firm has built an asset.

Two sources. The documents: contracts, precedents, past matters. Indexing them is the easy part. The other source lives in the lawyers' heads: experience, judgment, how a negotiation is run. Getting that into a form AI can use is the hard part, and the one that pays.

"Curated, kept current, with access rights" is not a technical footnote. It is what makes the knowledge trustworthy enough to act on, and it is part of the asset. Knowledge nobody maintains or permissions is a liability with a search box.

The technology is not the point. The point: if twenty years of a partner's expertise exist only in that partner, AI cannot multiply it. Made accessible, the same expertise reaches the junior, the other practice area, and a service sold to the client.

This is also the answer to "we already have Legora". The question is not which tool. It is whether the AI the firm uses reaches the firm's own expertise, or only the general capability every competitor has. A&O Shearman describes distilling 20 to 30 years of expertise (opens in a new tab) out of its lawyers' heads into its own tools, down to client playbooks embedded in the product (opens in a new tab).

Without it, AI makes individual lawyers faster. With it, AI scales the firm's expertise.

Not a product comparison. The picture shows what the firm owns in each option, layer by layer. Above the dashed line is the firm's. Below it is bought.

On a specialist platform the firm owns only its services. Its knowledge goes into the vendor's workflows and agents. Fast to adopt, little to build, and dependent on what the vendor allows and what can be moved out later. With an own AI layer the firm owns everything but the models. Most control, most building.

The hybrid is the interesting one. The firm owns its knowledge and key workflows and buys the rest where it is best. A&O Shearman's ContractMatrix (opens in a new tab) works this way: their own product and legal layer, developed by their own innovation group with their lawyers (opens in a new tab), with Harvey and Azure OpenAI inside. Owning the advantage did not mean building all the technology.

The decision is where the line goes. Not which product, not yet. At which layer does the firm say: from here up, this is ours?

Four stages, left to right. Generic AI is where everyone is, clients included. Firm AI is where the model starts working with the firm's precedents and playbooks. AI-enabled workflows is where the AI stops answering questions and does steps of the work the firm's way. AI-enabled services is where new revenue appears.

The strip underneath says why to move right: commodity, differentiation, competitive advantage, new revenue. The strip on top says what changes: AI assists on the left, AI executes on the right. That shift is what "agents" means: from answering a question to carrying out a defined process.

Two firms are already at stage four. A&O Shearman turned its lawyers' playbooks into ContractMatrix (opens in a new tab), a contract review product its clients use. Kirkland & Ellis built its fund formation workflow, institutional knowledge and senior judgment into one AI-enabled service (opens in a new tab).

Stage four has a second branch. Once the firm owns its knowledge, workflows and AI framework, the same capability can be used to build a service with a client, on the client's data and process. Linklaters' Applied Intelligence (opens in a new tab) pairs lawyers with data scientists to co-design bespoke AI-enabled solutions for clients' problems. The firm stops selling only its own product and starts building the client's. A service at this stage can still have lawyers in it. What changes is that it runs continuously and scales.

Efficiency scales lawyers. Productization scales the firm's expertise. AI stops being an IT cost paid back in saved hours and becomes a way to build new services and revenue.

This is no longer hypothetical. According to the Financial Times (opens in a new tab), Goldman Sachs asks its law firms to quantify their AI savings and uses the number in fee negotiations, Citi asks firms bidding for its work to disclose theirs, and Morgan Stanley is putting most of its outside legal work out to competitive bids with fixed fees on the table. At Legalweek 2026 (opens in a new tab), DLA Piper's head of AI innovation put it as: do not wait for the client to set the terms, lead the conversation about AI ROI.

The shared model already exists in practice. Clifford Chance (opens in a new tab) publishes cases where technology-assisted due diligence cut the client's cost by 40 to 50 percent, and says so in the numbers it shows clients.

The AI dividend and the pricing model are two different decisions. The first is who keeps the saving. The second is what the firm charges for once hours stop measuring value. Houthoff (opens in a new tab) sets fixed fees for AI-assisted work from the measured time saving and the complexity of the task, or raises the hourly rate to reflect more value per hour. Addleshaw Goddard (opens in a new tab) names the same three structures: fixed fees, subscriptions and outcome-linked pricing. A subscription fits the recurring services of section 06. None of them removes scope risk. A fixed fee moves it to the firm, so it fits predictable, repeatable work first.

AI efficiency does not set the price. The pricing model decides who captures the AI dividend, and the largest clients have started to negotiate it (opens in a new tab).

The page so far says what to build and why. A partner's next question is what it takes. The answer from the firms already doing it is a small, deliberate organisation, not a tool rollout.

Linklaters (opens in a new tab) recruited 20 AI lawyers, a mix of its own practising lawyers and external hires, trained them in the firm's tools, prompt and workflow creation and change management, and placed them in practice groups to build with the client-facing lawyers. Its Applied Intelligence (opens in a new tab) team pairs lawyers with data scientists for client builds. Goodwin (opens in a new tab) launched an internal AI platform and a Builders programme where attorneys and technologists develop solutions on real workflows, with a target of 90% of employees using AI in everyday work by the end of 2026.

The same organisation has to answer a question the old one answered by accident: how the firm grows its next seniors. Lawyers learned by doing the research, the first drafts and the review. AI now does much of that. Simmons & Simmons (opens in a new tab) puts it directly: learning itself has to evolve, and the firm has to be deliberate about what replaces that experience. Vorys (opens in a new tab) built AI personas of 19 partners so associates can get feedback in a partner's way of thinking. Whatever the method, it is a budget line and a design decision, not a side effect.

Change management and AI training belong here too, as the enabling work. They are not a strategy of their own.

This does not end in a ready-made strategy. It ends with six decisions, one per section, that the partners now have enough evidence to take a position on. Each card links back to the case for it.

The timeframe matters more than it looks. Six to twelve months of thinking before anything starts is too slow. Linklaters (opens in a new tab) went from recruiting 20 AI lawyers to launching a team that builds AI solutions for clients in six months. Freshfields (opens in a new tab) had Gemini inside its own legal solutions within a year. Kirkland & Ellis announced a $500M AI programme (opens in a new tab) and a live client-facing platform (opens in a new tab) two weeks apart. The scale of the money is the other message: this is not a ten-thousand-euro experiment. The long-term strategy can be two or three years. The first decisions and pilots should take ninety days.

Left open on purpose: which vendor, and the security and governance details around the work. Those are settled in the implementation, whatever the choices are. The capabilities to build it are not left open. That is decision six.