Made to measure

There’s a well-worn story about FE Smith, the Earl of Birkenhead and one of the most notorious barristers of the early 20th century. Asked to advise on a libel action one evening, he was handed a four-foot pile of papers to review before the next day. He proceeded to send out for a bottle of champagne and a dozen oysters, and worked through the night. The following morning, he produced a short opinion: “There is no answer to this action for libel, and the damages must be enormous.” 

Like many legal anecdotes, it improves with age, repetition and a forgiving attitude to detail. But it also poses a serious question: what, exactly, was the client paying for? Not the hours alone, nor time-and-a-half after midnight, the length of opinion or the faint glamour of professional stamina sustained by shellfish. The client was paying for Smith’s judgement: the ability to see the point quickly, identify the answer and state it with economy. It reflects the proposition that the measure of legal fees is really about value rather than time – and artificial intelligence (AI) is now set to bring this question into much sharper relief.

This is why AI matters for costs. Not because machines are fashionable, and not because solicitors must all become amateur legal engineers, but because much legal practice (and hence a firm’s profitability) still rests on time as the basic unit of billing fees and costs recovery. AI will weaken and possibly destroy this connection at its most vulnerable point: work that is language-heavy, repetitive and document-rich – and once comfortably chargeable by the hour.

What does AI mean for lawyers?

AI will not destroy the world. AI will not take your job. AI will not drink your coffee. It’s neither sorcery nor a swindle. In practice, it’s a family of software applications that can recognise patterns, process language, classify information, generate text and make predictions from data. It’s a probability machine that puts words together in their most sensible pattern to create a solution to a query. This makes it highly relevant to law, which is, among other things, a profession conducted largely in words.

Large language models such as ChatGPT, Claude and their rivals do not think like lawyers. They predict and generate language from vast training data. They can summarise, draft, compare, extract themes, build chronologies and suggest lines of enquiry. They can also invent authorities, miss nuance and produce polished nonsense with unnerving confidence.

The lawyer’s task isn’t to worship the machine or sneer at it from a safe, Edwardian distance. Rather, it’s to use AI where appropriate, supervise it properly, verify what matters and remain responsible for the final product. AI may prove an excellent tool, but it’s not, even by some miracle of branding, a qualified fee earner. 

2022: a tech odyssey

AI did not begin in 2022, but for many lawyers this is when it became real. Earlier forms of machine learning, analytics and e-disclosure had been around for years. The change came when conversational tools made the technology feel accessible. The machine no longer had to be programmed at arm’s length in code; it could be asked to do things in English.

This matters because legal work is built from language: instructions, pleadings, advice, witness statements, disclosure, submissions, orders, bills and points of dispute. Technology that handles language quickly goes directly to the centre of daily practice. 

The effects of using AI will appear first where documents are plentiful and issues recur, such as with disclosure-heavy disputes, high-volume claims, routine drafting, standard correspondence and data-rich litigation like credit hire. In more bespoke litigation, including catastrophic injury work, human judgement remains central, but the gathering, sorting and testing of material becomes faster and cheaper.

Compression of legal work

The most obvious effect of using AI is the compression of time taken to perform mundane tasks. It reduces the time needed for tasks that once generated a healthy crop of recorded units, such as from the armies of fee earners deployed on disclosure.

Research is another example. Once, this meant libraries, indices and several hours of disciplined rummaging. Databases reduced that burden; AI reduces it again. A lawyer can now ask a question, refine it, test the distinction, request counterarguments and produce a provisional structure within minutes. Verification remains essential, however, and anyone who files AI research without checking it deserves whatever procedural storm follows. But the economics of arriving at a first position, or the production of a first draft, have plainly changed.

Document review is even more susceptible. Machines can sort, classify and interrogate large data sets far faster than human teams, identifying themes, repetitions, gaps and anomalies. The lawyer still decides what matters, but the route to this decision may be much shorter than it was before.

Drafting is also vulnerable to compression, though lawyers may resist admitting it. Some documents are original works of thought. Many are precedent-based or in template format, and structurally familiar. AI can often produce a serviceable first draft of a pleading, letter, witness statement, instruction or note. It may be too generic, miss the best point or strike the wrong tone, but it’s still a first draft. The professional task becomes editing, testing, improving and taking responsibility for it.

Once a sound draft can be produced in a fraction of the previous time, our original question returns: for what is the client paying? It cannot honestly be just the minutes spent prompting, and it becomes harder to defend the notion that the work still ‘really’ took eight hours. The answer lies in value rather than duration, and shifting the charging model to reflect the value conferred, rather than time spent. 

Billable hours were never holy writ

The billable hour wasn’t handed down on tablets of stone in the mists of antiquity. In truth, it’s a relatively modern management device, useful for understanding overheads, salary costs and profit. Over the last 70 years or so, time recording became billing, billing became costs recovery and, before long, parts of the profession spoke as though value and time were interchangeable.

They are not. Time may be evidence of complexity, urgency or difficulty, but it isn’t value itself. A poor lawyer can spend 10 hours misunderstanding a problem; a good lawyer may solve it in one. No sensible client believes the first performance is worth more merely because it was longer. AI makes this truth impossible to evade. It accelerates competent lawyers, but it also exposes how fragile the assumption that time equals value has always been.

Zero marginal cost legal production

Once AI is built into legal workflows, the marginal cost of producing certain outputs falls sharply. Another summary, chronology, comparison table, first draft or issue list can often be generated at incredibly low, incremental cost. This doesn’t make legal expertise free. Systems must be bought and governed; confidentiality must be protected; lawyers must supervise the process and assume responsibility for the result. But the cost of producing the written artefact may drop significantly.

The value therefore lies elsewhere: in judgement, strategy, experience, risk allocation, knowing which point matters, recognising when the machine has wandered off into fantasy and deciding when to settle and when to fight. This is what clients should pay for, and what firms must learn to describe and price with more honesty than ‘0.3 to review AI output’.

Value-based charging

The obvious consequence is a greater move toward value-based charging: fixed fees, staged fees, capped arrangements, retainers, subscriptions or other structures that reflect the worth of the service rather than the duration of the task. This does not mean every dispute can be fixed fee from the outset. Litigation is an untidy business. Opponents misbehave, courts alter timetables, clients change instructions and documents breed with a zeal that would impress a rabbit.

A bad fixed fee is often just hourly billing in a fake moustache. But the direction of travel is plain. As AI reduces the time needed for many tasks, firms must rethink how they charge, or clients and competitors will do the thinking for them. Practically, firms should review retainers and client care documents; decide when and how AI may be used; adopt clear rules on confidentiality, verification, supervision and audit trails; and train lawyers not merely to prompt systems, but to challenge them. They should also decide how AI-assisted work will be explained later to clients, insurers, paying parties and costs judges. ‘The computer did a lot of it’ will not do.

Inter partes recovery: CPR 44.4

Looking at the recovery of costs in litigated matters, even now, inter partes assessment is not, in principle, chained exclusively to time. Part 44.4 of the Civil Procedure Rules (CPR 44.4) directs attention to whether costs were reasonably and proportionately incurred and reasonable and proportionate in amount, having regard to matters including conduct, value, importance, complexity, skill, effort, specialised knowledge, responsibility, time spent, circumstances and any approved or agreed budget. Time is there, but it’s only one factor among several. This means the existing framework can already accommodate value as a gauge for assessing costs. 

The obstacle to doing so is the conservative habits of the profession and the judiciary. Detailed assessment still tends to revolve around hourly rates and anxious disputes about units, attendances and whether an attendance note proves that work was done, and how long it took. 

AI-assisted work will strain this practice. A receiving party seeking to recover a fixed or value-based fee will need to explain the work done, the skill involved, the responsibility assumed, the importance of the issue and why the amount is proportionate. A bald entry such as ‘AI-assisted review: £5,000’ may well elicit a frosty judicial response. Equally, a paying party shouldn’t be heard to say that work is worth almost nothing simply because the machine made parts of it quicker. The real question is what professional value the lawyer added and whether the charge is reasonable in context.

Risks of using AI

Although AI presents possibilities for firms to reduce overheads and increase profit, there are obvious risks. Efficiency may outrun pricing, so firms become faster but oddly less profitable. Or the reverse happens: firms rebadge time-based billing as ‘value’ without being able to justify the figure by reference to risk, expertise or outcome. There is also the old-fashioned risk of negligence. AI can hallucinate authorities, miss procedural traps and build persuasive prose on false premises. The lawyer who signs the document owns the error. 

Confidentiality is another risk: firms must know where client data is going, how it is stored, whether it is used for training, and what contractual protections exist. And there is the ever-present danger of complacency: using AI to produce average work more quickly and then invoicing it as excellence. This isn’t innovation; it’s merely faster mediocrity with better marketing.

Fixed recoverable costs

Fixed recoverable costs (FRC), which have expanded in scope to cover all money claims of up to £100,000, have often been criticised by the profession as blunt and insensitive to the untidiness of real litigation, and sometimes with good reason. But in an AI-assisted world, they may look less crude and run with the grain of how much of the work will be done. If production cost is no longer sensibly measured by hours, a tariff linked to case type, stage and value may be more coherent than a reconstruction of time spent.

This is especially true in high-volume work, such as portal claims, or credit hire, debt recovery, low-value injury litigation and similar cases where there is greater scope to deploy standardised processes, and where AI-native practice models will emerge first. A firm designed around automation, data, fixed stages and close supervision may enjoy a structural advantage over one trying to bolt AI onto an hourly-rate model built for another age.

What should firms do now?

Panic is seldom a business plan, and buying every shiny legal AI product sold by a man in a fleece who says ‘solution’ before coffee is not a strategy, either. Instead, as part of the arrival of a new industrial revolution, firms should do the following: 

• Identify where AI can safely reduce time in existing workflows.
• Decide how the resulting value should be priced.
• Amend their retainers, client care documents and internal policies that will largely be drafted to reflect hourly rates charging.
• Act now to train lawyers to verify outputs, protect confidentiality and preserve audit trails. 
• Prepare to justify AI-assisted work by reference to value, not merely time. The better question is not ‘how many hours did AI save?’ but ‘where, exactly, is our value, how do we price it, and how do we prove it?’

Conclusion: time must serve value

FE Smith’s world of oysters, champagne and mountainous piles of paper has sadly long gone, but the problem his anecdote represents stills remains. How does one price exceptional legal judgement? The billable hour was never a full answer. It was at best a proxy: sometimes useful, often convenient and frequently misleading.

AI won’t remove the need for legal advice, advocacy or responsibility. If anything, it may make judgement more valuable, because fluent words will become easier and cheaper to produce. What it will do is destroy the assumption that time is the best measure of value. 

For solicitors, the task is to build a principled bridge between new methods of production and old methods of charging and recovery. CPR 44.4 already supplies much of the architecture.

Time is the servant of value. AI has simply reminded the profession of this age-old truth.