Your firm bought an AI legaltech tool. That isn't transformation.

In a previous piece we wrote about the cultural barriers to generative AI adoption in law firms: the pyramid of perfection, the fixed mindset it rewards, and how badly that culture fits a technology that demands experimentation. This piece is about the structural half of the same problem, and about a comparison that we think the profession has not yet made.

The money is now real. Harvey raised at a $15.5bn valuation in September 2026. Legora reached $5.55bn in March and passed $100m in annual recurring revenue eighteen months after its platform became generally available. Thomson Reuters and LexisNexis both relaunched their flagship products around agentic orchestration within three days of each other in August. Legal technology spending grew 9.7% in the most recent year measured by the Thomson Reuters Institute, the fastest rate it has ever recorded.

And roughly 90% of the money in the legal industry still flows through the billable hour, unchanged. Thomson Reuters put it better than we could: firms “have built incredible efficiency machines while using outdated billing systems”.

That sentence should be familiar to anyone who lived through the last great enterprise transformation — Agile.

Somewhere around 2010, every large organisation decided to become agile. They bought Jira. They sent people on two-day courses and gave them certificates. They renamed project managers as product owners, installed stand-ups, and put a burndown chart on a wall.

A great many of them did not become agile. Teams got faster at the bit in the middle, and the organisation around those teams carried on exactly as before: annual budget cycles, business cases written a year ahead, quarterly release boards, a governance process designed for a world of fixed scope. The result was a quicker middle inside an unchanged system.

Forrester's Dave West named the pattern in 2011 and called it Water-Scrum-Fall. It is the single most useful idea in this article, because it describes with precision what is happening inside law firms right now.

We say this as people who were in those rooms. Futurice was founded in 2000 and built agile from the inside out — it was how we worked before it was something clients asked us for. Through the 2010s we spent a decade inside other organisations’ agile transformations, and the pattern was consistent enough to be predictable: the teams changed quickly, and the system around them did not. Delivery got faster. Funding cycles, governance gates and the definition of a successful year stayed exactly where they were. What worked, when it worked, was never the framework. The organisations that broke through were the ones willing to change how work was funded and who was allowed to decide. That is precisely the move legal has not yet made.

The firms buying Harvey and Legora have not made a mistake. They have made a good decision — and then, in most cases, stopped, as though the decision were the work. Buying the tool is where the transformation starts. It is not evidence that one is happening. The tools are genuinely good, the adoption is genuinely real, and the time savings are genuinely being achieved. At Legalweek in 2026, DLA Piper's Barclay Blair said due diligence on certain financial contracts had already been compressed from fifteen or twenty hours to two. That is not a pilot result. That is production.

The question is what happens to the other thirteen hours — and the answer, in most firms, is that nobody has decided. The firm does not get to take its time over that. The general counsel on the other side of the matter already knows the diligence takes two hours, because their own team is running the same tools on the same documents. This is the part with no agile equivalent: the client is holding a stopwatch. Whatever a firm concludes about its pricing model, the conclusion has a deadline, and the firm is not the one setting it.

The faster middle

A two-hour due diligence is not worth very much if the matter took three weeks to open, the fee was estimated on an hourly assumption before the work started, the billing system cannot express any other kind of fee, and the pitch that won the work was written from last year's template. The legal work got faster. The firm did not. This is Water-Scrum-Fall with different job titles. The efficiency is real and local. It does not compound, because nothing around it changed to let it compound.

The funding model nobody changed

The most consistent finding in the research on why agile transformations stalled is not about tooling or training. It is about money. Annual budget cycles allocating fixed sums against fixed scopes are structurally incompatible with a method built on continuous re-prioritisation. That is why the Beyond Budgeting movement exists, and why the organisations that genuinely changed usually had to change how they funded work before anything else would hold. Law firms have a harder version of this problem. Partnership profits are distributed annually. Multi-year capability investment has to be argued for out of this year's drawings, by the people whose drawings they are. The firms currently reporting lower profit per equity partner because of technology investment are not doing something wrong; they are doing something structurally difficult, in public, with a constituency that votes.

Will those firms be the winners? Not automatically — suppressed profit is a signal, not a strategy, and a firm can spend heavily on licences while changing nothing that matters. The distinction is what the money is buying. Investment that buys tools shows up as cost and stays as cost. Investment that buys a changed business model — what the firm sells, how a fee is constructed, how a matter is staffed and supervised, how a career is built — is the kind that compounds — and it is the kind that is hardest to argue for in a room where the people voting are the people paying. Our view is that the partners who win that argument now will be running the firms everyone else is trying to copy in five years’ time, and that it will be described, at the time, as luck.

Certification theatre

The sharpest criticism of the scaled agile frameworks was that they let an organisation purchase the appearance of transformation. Roles were renamed. Ceremonies were installed. Certificates were issued. Decision rights were untouched.

The legal equivalent is already taking shape: AI champion networks, prompt-writing training, adoption dashboards, a percentage-of-lawyers-onboarded metric reported to the board. None of it is bad, and that is precisely the difficulty — it is doing a job. It gives the board a warm and entirely sincere sense of progress, and that comfort is what allows the fundamental questions to stay unasked for another year. Ninety per cent onboarded goes on a slide. How the firm makes money, how a matter is staffed, who is accountable for an agent’s output and what a trainee does in their first year do not go on the slide, because none of them has changed.

Which gives us a test, and it is an uncomfortable one. If the same people are making the same decisions in the same forums, the firm has not transformed — it has upgraded. Decision rights are what actually moved in every transformation that held, in any industry, and they are the one thing no platform can shift on a firm’s behalf. A firm can run an AI champion network, hit ninety per cent onboarding and publish a maturity score, and if pricing is still set the way it was set in 2019, by the people who set it in 2019, nothing structural has happened.

Measuring the wrong thing

Agile measured velocity and story points because they were countable, and then spent a decade discovering that they measured activity rather than value. Legal is measuring hours saved, for exactly the same reason and with exactly the same risk.

The client-side evidence already shows the gap. In the ACC and Everlaw survey of in-house counsel, 64% expected generative AI to reduce their reliance on outside counsel and around half expected lower costs — while nearly 60% reported seeing no savings at all. Thomson Reuters has issued the warning in plain terms: premium billing tied to AI adoption is sustainable only when clients clearly see the added value. Put less diplomatically, the premium survives exactly as long as the client cannot see what the work now costs to produce.

It is worth being specific about where this ends, because the profession is not being specific about it. A firm that buys the tools and changes nothing else arrives, in about three years, somewhere recognisable: a cost base that has not moved, a client base that knows what the work now takes, and a fee conversation in which the only remaining lever is discount. Efficiency that is not converted into a different commercial model does not stay with the firm. It leaks to the client through the negotiation, slowly and without credit, because by the time it is visible it is expected.

An analogy is only worth using if you know where it fails. This one fails in four places, and three of them should make a law firm leader more concerned rather than less.

Legal has a regulator. Agile never did.

Nobody was ever struck off for bad Scrum. No court sanctioned a firm for its sprint cadence. No insurer priced professional indemnity cover against story-point maturity. In legal, all three of those exist. In Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank (2025) EWHC 1383 (Admin), the Divisional Court dealt with a barrister who cited five non-existent cases and a solicitor who relied on AI-generated research containing forty-five problematic citations, eighteen of them entirely fictitious. The court's position was that lawyers using AI must check accuracy against authoritative sources before professional use. A further High Court decision in 2026, Cork & Anor v Smith, found that a firm's own supervision had failed — a senior associate and a partner had approved letters to the court containing fabricated statutory wording without checking them.

The Solicitors Regulation Authority has issued a warning notice on the misuse of AI. Professional indemnity insurers, according to practitioners writing in the trade press in March 2026, now expect to see evidence of how a firm is adapting before they will price cover.

Every one of those failures was a failure of supervision, not of technology. In the 2026 case, the AI tool itself warned the user to verify the material and was ignored.

This is the most important difference between the two eras, and it points the opposite way to the usual conclusion. The agile comparison is not a reason to expect law firms to move slowly. It is the reason to expect that they will not be allowed to.

A partnership is not a hierarchy

Corporate agile transformations had, in the end, a chief executive who could mandate. A law firm has hundreds of owner-operators, each with a vote and a client base. Laura Empson's research on professional partnerships describes the governance condition precisely: ambiguous authority and hidden hierarchy. Larry Richard's personality research on lawyers supplies the behavioural half — as a population, lawyers score around the ninetieth percentile for scepticism and the eighty-ninth for autonomy. These are not flaws. They are the professional virtues that make someone a good lawyer, and they are also the reason that a firmwide mandate does not work the way a corporate mandate does.

The practical consequence is that in a partnership, change is won by demonstration rather than instruction. Not a firmwide programme announced at a partners' conference, but one practice group, one matter type, one visible result that other partners want for themselves. That is slower to start and considerably faster to spread.

This is also the only way we have seen the decision-rights problem actually solved. A pyramid does not hand down authority because a strategy document asked it to. It releases authority to the practice group that has already demonstrated it can be trusted with it. Demonstration is not an alternative to changing who decides. It is the mechanism by which who decides gets changed.

The economics run the other way — and there is no precedent

Agile promised more throughput from a fixed-cost engineering function. The value accrued to the organisation adopting it, and nothing about it threatened anyone's revenue.

Agentic AI compresses the unit that a law firm sells. We looked for a historical case in which an organisation's own revenue model was dismantled by a technology it chose to buy, and we could not find one. Newspaper classifieds, high street travel agents, the audit-to-advisory shift — these are all stories of somebody else taking the revenue. This is different. The firm is purchasing the thing that shrinks its own billable base, and doing so because its clients and its competitors leave it no realistic alternative.

We will put that as a challenge rather than a claim: if there is a precedent — an industry that voluntarily bought the technology that dismantled its own pricing unit, and came out of it intact — we would like to know what it is. We looked, and we could not find one. Which means nobody can tell a law firm how this was handled last time. There was no last time.

The demand is arriving from both directions at once

The final difference is in the profession's favour, and it is underrated.

Agile began as a practitioner movement in 2001 and spent a decade pushing upward before most executives engaged with it seriously. Legal AI is arriving from the top and the bottom simultaneously: bought as a firmwide programme by leadership, and pulled by lawyers who actively want it. The standard set at Legal Geek in 2025 was memorable — a tool has succeeded when the lawyers would riot if you took it away.

Executive sponsorship and genuine user demand at the same time is a condition that agile transformations rarely enjoyed. It removes the most common excuse for failure. If a firm has the budget, the mandate and the appetite, and still ends up with nothing more than efficiency, the missing ingredient was never the technology.

There is a second structural problem, and it is almost entirely absent from the industry conversation.

Legal technology is aimed at legal work product — research, drafting, review, discovery. But a large law firm is not only a collection of lawyers doing law. On the clearest available data, which is the Lawyer's UK 200 business services benchmarking for 2016/17, the ratio of business services staff to fee-earners across the UK market was around 0.70 to 1. That is roughly forty per cent of headcount. Some firms ran close to parity.

The figure is nearly a decade old, and the fact that nothing more current exists is itself revealing: the profession is restructuring a population it is not measuring.

Look at where the reductions are actually landing. In November 2025 Clifford Chance announced the loss of up to fifty London business services roles, citing greater use of AI among its reasons. In May 2026 Clyde & Co opened consultation on moving around two hundred UK roles in legal support, purchase-to-pay and IT to Glasgow and to a shared services centre in Manila. These are not lawyers. They are the functions that have to work differently for anything in the legal delivery model to change.

And the vendor landscape reflects the neglect almost perfectly. Harvey and Legora together are worth something in the order of five to six times the entire public market value of Intapp, the most advanced and most acquisitive vendor serving the business of law. Further down the scale the gap becomes absurd: one business-of-law vendor raised a $23m Series B in late 2025, two orders of magnitude below a single funding round on the legal work product side. The industry's own assessment, from the business-of-law track at a 2026 conference, was that firms are “buying licences and throwing them over the fence”.

Client onboarding. Conflicts and anti-money-laundering checks. Pitches and panel submissions. Pricing and matter scoping. Billing and e-billing compliance. Knowledge management. Career frameworks and training design. None of this is doing law. All of it determines whether doing law faster is worth anything.

This is why we think the business services question is not a second, smaller opportunity to be addressed after the legal one. It is the constraint on the legal one. A firm that transforms legal delivery inside unchanged surrounding processes has bought itself a faster middle — which is exactly what the agile era delivered, and exactly why so little of it compounded.

Not more pilots. Five structural moves, none of which require a firm to become a technology company — and none of which are only about price. Each of them goes at an assumption underneath the business and operating model: what the firm sells, how it is made, who makes it, and how anyone knows it is any good.

1. Make the work explicit before pointing agents at it

Agents automate tasks inside workflows that were never written down. Both leading platforms ingest a firm's playbooks and precedents rather than creating them, which means the encoded judgement of what good looks like has to exist in the firm's own hands first.

The bold move is unglamorous and highly effective: take one matter type, decompose it into its actual constituent tasks, and decide task by task what stays with a human, what a human leads with agent assistance, what an agent leads with human verification, and what can be automated outright. Most firms have never had that conversation about any matter type. It usually takes a room, a wall and a day, and it reveals more than a year of pilots.

It is worth being blunt about what this means commercially. Both leading platforms ingest your playbooks and precedents; neither writes them. What a firm is buying is leverage on judgement it has not yet written down — and if it is not written down, the leverage has nothing to act on. This does not come up in the sales process, and it should not: it is not a vendor’s job to tell a firm that the missing input is the firm itself. But it is the reason two firms buying the same licence get results a year apart.

2. Change the commercial wrapper, not just the delivery

The managing partner of Linklaters told an audience in November 2025 that the structure for monetising production will have to change. He is right, and almost nobody has done it, partly because the finance system is often the binding constraint: a billing platform built for time cannot represent a fee that is not time.

The bold move has a number attached. Pick one service line and stop billing it by the hour — not a pilot, not a hybrid, the whole line, for twelve months. Design the entire wrapper around it: the scope, the price, the guarantee, the way it is sold, the way it is staffed, the way it is billed, on the assumption that agents do most of the production. Accept that the first six months will cost the firm money and treat that as tuition rather than failure. Every organisation we have watched genuinely change a commercial model did it by removing the fallback; the ones that left the old option open defaulted back to it inside a quarter, every time. Then find out what breaks. It will not be the AI. It will be the finance system, the matter-opening process, the resourcing model and the conversation about what a partner is now for — which is precisely the list a firm needs, and precisely the list it cannot obtain by reasoning about it in advance.

3. Build the accountability thread before a court builds it for you

The reported judgments are not really about hallucination. They are about who checked, who approved, and whether anyone could later reconstruct why a piece of work was trusted. Every firm will need an auditable thread running from a decision back to the evidence and the person who owned it. The bold move is to design that thread as workflow rather than policy. A policy document telling lawyers to verify AI output is what every firm already has. A verification step that is built into how the work actually moves, with a named owner and a record, is what the courts, the regulator and the insurers are converging on requiring.

4. Treat the other forty per cent as the main event

The temptation is to answer the business services question with labour arbitrage, because that is what the market is currently selling: move it offshore, centralise it, shrink it. We would be cautious. Offshoring a function that agentic capability is about to change fundamentally risks building a service centre sized for a volume of work that never arrives.

The bold move is to redesign those functions for what is about to become possible, rather than relocating them as they are. Client onboarding, pitching and pricing are journey and process design problems. They happen to be the problems that most respond to exactly the kind of redesign that the legal side is getting all the attention for.

5. Redesign the first five years of a legal career, on purpose

Here the data is genuinely surprising. Associate leverage reached a seventeen-year record in the final quarter of 2025, above even the 2008 peak, while firms continued to hire. The pyramid is not collapsing. Trainee numbers in the City have been broadly stable for five years.

So the question is not how many juniors a firm needs. It is what they now do, and how they learn judgement when the first draft no longer comes from them. That is a curriculum and career design problem, and the firm that solves it will have a recruiting advantage that compounds for a decade. Most firms are waiting to see what everyone else does.

Everything above asks a firm to change things it has never changed, and there is an obvious objection to all of it. A law firm is not a laboratory. It has clients mid-matter, partners mid-career, an insurer, a regulator and a reputation that took a century to build. The risk of reopening the business model is not theoretical; it is that you get it wrong in public with a client watching.

So the answer is not courage in the abstract. It is being structured about being uncertain. Choose a contained part of the firm — one practice group, one matter type, one client relationship deep enough to survive a conversation about doing things differently. Change several things at once inside that boundary, because changing one thing at a time is exactly what produced the faster middle. Put a time limit and a defined cost on it. Agree in advance what would count as a failure, and agree, in advance, that failure is an acceptable outcome. Then run the next one with what you learned. This is ordinary experimental discipline, and it is almost entirely absent from how firms are approaching AI, where the unit of change is currently the licence renewal.

The harder half is unlearning. Almost everything a senior lawyer knows about how a good firm runs was learned inside a system where the hour was the unit, the trainee was the first draft, and seniority was the quality control. Those beliefs are not wrong. They are load-bearing, and they were built for conditions that are now moving. A firm cannot design its way past assumptions it has never said out loud.

The most productive day we have ever spent with a leadership team is the one where the organisation writes down what it currently believes to be permanent — and then goes through the list, item by item, asking whether it still is.

That is the real dividing line, and it is not between the firms with the best technology and the firms without. It is between the firms with a method for being wrong safely and the firms with no method at all. The first group can afford to question a thirty-year-old assumption, because they have a way of testing it that does not put the firm at risk. The second group cannot, so they buy another tool instead, and call it progress.

The legal profession is at the point in the cycle where the technology is proven, the adoption is real, and the transformation has not happened. That is precisely where the agile era sat around 2014, and the organisations that broke through then were not the ones with the best tooling. They were the ones that changed how work was funded, governed, priced and led.

The honest summary:

• The tool is not the change. Buying Harvey or Legora produces efficiency. Efficiency is worth having and is not transformation.
• Watch for the faster middle. If the legal work is accelerating while intake, pricing, billing and pitching are unchanged, the gains will not compound.
• This time there is a regulator. Courts, the SRA and professional indemnity insurers are all converging on the same question, and they will not wait for the profession to answer it voluntarily.
• Mandate will not work; demonstration will. In a partnership, one visible result that partners want for themselves travels further than any firmwide programme.
• Forty per cent of the firm is not doing law. It is the least examined and most constraining part of the problem, and almost nobody is selling a solution to it.
• It is not only about price. Pricing is the most visible assumption, not the only one. Staffing, supervision, career design and what a partner is for are all built on the same foundation.
• Structure beats bravery. The firms that break from the pack will be the ones with a method for running contained experiments and unlearning safely — not the ones that announced the boldest programme.

One prediction, so that this can be marked. We do not think the billable hour dies this decade. We think it survives — and that its survival is the trap. It will keep working well enough that most firms never build the alternative alongside it, right up to the point where a client asks for a price the firm has no way of constructing. The billable hour will not be killed by AI. It will be quietly outcompeted by firms that spent three unglamorous years learning to sell a different way.

Our own view, stated plainly enough to be argued with: the firms that look transformed in five years will not be the ones that picked the best platform. They will be the ones that went back to the assumptions underneath the business — what they sell, how it is priced, how it is produced, who produces it, what a partner is for and how a career is built — and were willing to find out, in a contained and deliberate way, which of those assumptions no longer hold. They will have done it while the change was still voluntary rather than negotiated. We would rather be wrong about that in public than vague about it.

None of this requires a firm to predict the technology, which is fortunate, because nobody can. It requires a firm to be honest about what it has actually changed, as opposed to what it has bought.

The firms that treat this as an organisational design problem rather than a procurement decision will spend the next three years building something their competitors cannot copy by signing the same contract. That has always been the difference between the firms that transform and the firms that merely modernise.

Tom Castle is VP Strategy & Transformation at Futurice. This piece follows “Bridging the Culture Gap: Generative AI Adoption in the Legal Industry”, published November 2025.