Is your AI strategy a castle built on sand?

Why the firms that win the AI transition do so on structure, not software.

Picture the legal market around two years on from the arrival of ChatGPT. Leadership at many firms decided it was time to take the technology seriously. An AI steering committee was formed, with a respected partner as its sponsor. A pilot was commissioned to test the leading legal AI platforms: Harvey and Legora.

The early feedback was positive. There was genuine interest, a flurry of internal momentum, a sense that the firm was moving in the right direction.

A year later, the firm either rolled out a single platform or adopted a portfolio approach, supplementing its legal platform with frontier models such as Claude or Gemini. And yet usage quietly tailed off. Aside from a small group of enthusiasts, most people drifted back to working exactly as they did before.

I have sat in those rooms, at the optimistic launch and at the quieter post-mortem a year later. It is one of the most common outcomes in legal technology today, and it has very little to do with the technology.

Strategy, it turns out, is the easy part. Almost any firm can draft a credible AI strategy or commission one.

What firms find far harder is implementing it and, harder still, achieving genuine adoption. The two are not the same. Implementation puts the tools in place; adoption changes how people work. And adoption is an organisational and behavioural problem masquerading as a technological one. Any firm that treats it as a software rollout will keep producing the same result: impressive pilot statistics, disappointing business impact.

Why it fails

When AI adoption stalls, the causes are remarkably consistent.

1. Ownership

   The first is ownership, though not in the way it is usually framed. The question is not whether someone has been put in charge of AI. It is whether whoever owns it has a real mandate, the seniority and the remit to lead change across the firm. In many firms, the conversation defaults to IT, which has an essential role to play but is naturally oriented towards procuring and running systems rather than reshaping how legal work is done. In others, responsibility sits between Knowledge and IT, or within a dedicated legal technology and innovation function. Plenty of firms now have such a function, and on paper, that looks like the problem solved. But a function is only as effective as the remit it is given. Too often, these teams are scoped as a support service, evaluating tools and fielding requests, when the moment calls for a transformation remit: the standing to shape how matters are run and how knowledge moves through the firm. That work requires an owner who understands both the business and the practice of law, and who has been given real license to act. The test is not "do we have a function?" It is "have we empowered it to transform, or to maintain the status quo?"

2. Data

   The second is data. AI is only ever as good as the knowledge it can reach. A firm's lasting competitive advantage does not lie in the models it licenses, which its competitors can license too. It lies in the firm's own material: its precedents, its closed matters, its house views, the accumulated judgment of its people. If that material is locked inside disconnected systems, or scattered across an architecture nobody has rationalised, the tools are reduced to performing party tricks on documents a lawyer has pasted in by hand. The unglamorous work of structuring the firm's data into something an AI can genuinely use is precisely the work most firms skip, and precisely the work that determines whether the investment pays off.

   
   

   The clearest evidence that this is where the real contest lies comes from the top of the market. Kirkland & Ellis, the world's highest-grossing firm, has committed up to $500 million to building its own AI capabilities and has partnered with Palantir to develop a proprietary solution for private equity fund formation. The logic its chair offered is revealing: widely available tools are "raising the floor for everyone," but the firm does not get hired for the floor. The solution is being built from information drawn from 250 of its lawyers about how they actually work, with the explicit aim of capturing the institution's collective judgment and making it available at scale across more than a thousand lawyers. Most firms neither need nor could justify spending at that level. But the principle scales down, and it is the right one: a firm's durable advantage is its own knowledge, and capturing it is an act of deliberate construction, not a by-product of buying software.

3. Incentives

   The third is incentives. Most firms still measure and reward what they always have: time recorded and hours billed, and then ask their people to adopt tools whose whole purpose is to compress the time a task takes. The firm is asking for one behavior while paying for the opposite. Lawyers are not naive. They respond to how they are measured, and until that changes, adoption stays shallow.

None of these is a failure of technology. Each is a failure to build the right structure around it.

What the firms that succeed do differently

The firms making real progress see the task differently. They treat the transition to AI-enabled operations as a business transformation, led from the top, rather than an IT project delegated down.

In practice, that begins with empowering a single, central capability to lead the firm's response, whether by establishing a new function or, more often, by giving an existing one a genuine mandate for transformation and the standing to use it. In the transformation projects I have led, I have found it useful to describe this as a control tower for AI: a function with the multidisciplinary skills, the authority and the vantage point to coordinate the firm's response across every practice area and business service, rather than leaving each team to experiment in a silo. It is best understood not purely as an efficiency measure but as a mechanism for both defense and enablement. Its real value is speed: the ability to move the whole firm in a chosen direction, deliberately and quickly, as the ground shifts beneath it. A firm without one will always be reacting, a department at a time, while its competitors move on.

Why speed now matters

Speed has become urgent because traditional firms are no longer competing only with one another or with alternative legal service providers.

A new category of competitor has emerged: the AI-native firm, built from the ground up with AI at its core rather than bolted on at the edges, and they are well-funded to press the advantage. Crosby, backed by Sequoia, pairs lawyers with proprietary AI to turn around commercial contract review at a speed the conventional model cannot match.

Garfield, in the UK, uses AI to serve small-claims and debt-recovery work that traditional firms have never found profitable, expanding the market rather than merely contesting it. Further out, agentic experiments such as Lavern, an openly published system of sixty-seven specialist legal agents, ask the question more sharply still: what defines a law firm once much of the work can be done by an agentic team, and how much of the firm's value collapses back to judgment alone? Nor is it certain that the platforms themselves will stay in their lane; the frontier labs and legal AI providers now serving firms may, in time, have ambitions of their own.

Whatever one makes of these new entrants, they move quickly by design. A traditional firm's only structural defence against a competitor built for speed is its own ability to respond at speed, and that is exactly what a mandated, central capability provides, and what most firms have not built, or have built but scoped too narrowly to matter. The threat is real. The more useful point is that the moat is still buildable. It is a question of organizational structure and will, not of buying better software.

A question of leadership, not technology

All of which makes this a leadership matter, and not one to be delegated and forgotten. Success in this era turns on the choices only leadership can make: what the firm values, what it measures, what it rewards, and, hardest of all, what it is prepared to change to get there.

The sequence is not complicated, though it is demanding. The first step is to accept that standing still or copying the firm next door is not a strategy, and to put the person charged with leading the transition on the executive or management committee, with AI as a standing item rather than an occasional one. The second is to build a genuine consensus around a vision of what the firm's operating model should look like in five to ten years, clear enough to justify the effort and disruption required to get there. The third is to translate that vision into the practical business of how work is structured, priced, resourced and led. This reaches well beyond technology, into change management, governance, incentives and, hardest of all, culture.

The firms that escape what we might call pilot purgatory are not those with the best tools, or even the sharpest strategy. They are the ones who built the internal structures to turn strategy into execution, and treated the whole exercise as the business transformation it always was. The tools are not the foundation. The structure is. The question facing every firm is no longer which tools or strategy to adopt, but whether it is building on rock or on sand. The ground beneath the profession is shifting, and what is a quiet advantage today will be decisive tomorrow.

Tom Connor partners with Lexington Consultants, working with its clients on digital transformation, AI strategy and legal technology projects. To find out what this might look like for your firm, please contact us at contact@lexingtonconsultants.com