top of page

The Missing Savings: AI, Efficiency, and the Legal Bill That Won’t Fall

Writer: Novara Advisory Partners
Novara Advisory Partners
Aug 13
10 min read

The tools have never been more capable, or the work faster to produce. So why aren’t legal bills falling, and why, for some clients, are they beginning to rise?


Authored by: Mason Gregory, Marcus Wolter and Lukas Kratochvil


Ask a general counsel what artificial intelligence has done to their legal spend, and you will rarely hear the answer the technology seemed to promise. The tools are remarkable. Leaving aside questions of accuracy, a first draft that once took a junior associate a day to produce can now be generated in minutes. Diligence that once occupied a floor of trainees can be run by a handful of people and a model. The work has never been faster to produce. Yet the invoices from their most trusted firms are, at best, unchanged, and, for a growing number of clients, quietly climbing.


Somewhere between the promise and the bill, the savings have gone missing. According to a July 2026 report from Deloitte, only 4% of in-house general counsel have reported experiencing any AI benefit from their external firms. The same survey found 78% want lower costs above all else, highlighting the widening gap between what clients expect AI to deliver and what they actually receive.


It is tempting to read this as a technology story. The news cycle is filled with early declarations of AI failing to live up to the hype across many industries. In this case, however, it is not. The capability is real, and improving by the month. Thomson Reuters estimates that AI could free up roughly 190 hours per lawyer per year, representing around $20 billion of freed-up work in the US alone that could be put to better use. The reason clients are not seeing it on their bills has almost nothing to do with the software and almost everything to do with the structure it is being dropped into. Understanding why may be the single most valuable thing a sophisticated buyer of legal services can do today.


None of this has gone unnoticed. As the established firms have struggled to turn AI into anything their clients can feel, a new kind of competitor has emerged to fill the gap: AI-native firms, built around the technology rather than bolting it on, and structured to pass those savings through to clients. They are still small and still new, but they are increasing.


What the bill was always for


To see why AI has not lowered the cost of legal work, it helps to be honest about what has driven that cost for the past half-century. Strip away the paneling and the legacy law firm is, at its core, a leverage machine. A partner sits atop a pyramid of associates, and the firm’s economics depend on billing those associates’ time at a multiple of what they cost. The more juniors beneath each partner, and the more hours they record, the better the model performs.


That structure carries a cost the profession has rarely named aloud, and the client has always borne it. Clients pay for first-year associates to learn on the job. They pay for work to move through successive layers of review: drafted by someone junior, revised by someone more experienced, and ultimately checked and signed off by someone senior. They pay for the handoffs, the catch-up calls, and the review layers that exist for a single reason: the people doing most of the work are not yet the people who can be trusted with the judgment.


For decades, this was simply the price of legal work, because there was no alternative. Leverage was the only way to produce advice at scale, and when every serious firm was built the same way, the client had nowhere else to take the problem. What AI changes is precisely that. For the first time, the leverage no longer has to come from an army of billable juniors. This means the pyramid stops being the inevitable shape of a law firm and becomes an organizational choice, one whose cost the client is entitled to question.


Bolted on, not built in


Faced with that shift, most established firms have done the understandable thing: bought the technology and bolted it onto the firm they already had. AI becomes another tool on the shelf, with Harvey and its brethren playing much the same role that Westlaw and LexisNexis did in the early internet age: more capable tools, but deployed to essentially perform the same tasks inside the same structure, just faster. However, a tool bolted onto the pyramid leaves the pyramid, and the price, exactly where they were. The firm captures a little efficiency at the edges: a quicker first draft here, a faster review there. But the client, watching from outside, sees very little of it reach the invoice.


In the best case, then, bolting AI to the legacy model nibbles at the edges of what the technology can do while the underlying cost structure holds. In a growing number of cases, it does something worse. The cost of tokens keeps compounding. The leading legal-AI platforms are moving away from flat per-seat licenses towards consumption pricing, and as autonomous agents take on more end-to-end work, usage and cost are climbing steeply. Firms are already being prepared for token bills that run into eight figures. When a new and variable cost lands on the firm, the instinct is to treat it as an expense and pass it through to the client as another line on the matter. Handled that way, AI does not lower the client’s bill. It adds to it.


Clients have seen this film before. A decade ago, it became common practice to simply strike legal-research charges from their invoices, having decided, correctly, that they would not, and should not, pay a premium for a database subscription dressed up as billed time. The same reckoning is coming for AI, and the question driving it is disarmingly simple: where are the savings? A charge that reflects genuine, disclosed value will survive that question. A charge that reflects the same work, produced more cheaply but billed at the old rate with a software meter attached, will not.


There is a further hazard for firms that lean on third-party platforms to get there. Those platforms are inexpensive today because they are buying market share; as firms grow dependent on them, the same vendors are well placed to take a rising share of the margin. Many vendors are already selling directly to the firms’ own clients. A firm that has outsourced its core capability to a supplier that also courts its customers has done little to strengthen its long-term competitive position.


Even where a firm genuinely commits to AI, there is a quieter obstacle that has nothing to do with pricing. You can buy a tool. You cannot buy consistent, sophisticated use of it across a firm of hundreds or thousands of lawyers who never signed up for a new way of working. Roll a platform out across that population and adoption follows a bell curve: a handful of genuine converts, a broad middle that dabbles, and a long tail that logs in once so it can say it tried. The firm’s effective “AI capability” becomes an average of that adoption curve, which is why, from the client’s side, the results so often look like marginal improvement rather than a genuinely different way of practicing. Fluency cannot be mandated into existence, nor trained into a workforce numbering in the thousands of attorneys on any timeline that matters to a live deal.


Less value, higher price


The always-beleaguered junior associate, meanwhile, arguably provides less of a value-add than ever. Through no fault of their own, young lawyers joining the profession are now from a cohort that has leaned on AI at every stage: university, law school, and on into practice. Many have become little more than intermediaries between the client and a model that the client could often have prompted directly (and without incurring fees of close to $1,000 an hour). Worse still, because AI now automates so much of the work once reserved for juniors, the understanding and experience that work used to build arrive more slowly, and far more unevenly, than they did for the partners training them. The firms seem to agree: despite record revenue, entry-level hiring across the Am Law 200 has stayed flat since 2022, according to SurePoint's law-school hiring data.


Yet the price of that work has moved in the opposite direction to its value. In 2015, a $1,000 hourly rate belonged to a senior partner at a top global firm; today it is roughly what a junior associate is billed out at. And the climb hasn't stopped: according to Wells Fargo, standard rates rose 9.6% in 2025 alone (following a 9.1% increase the year prior). The firms have prospered accordingly: average profit per partner across the Am Law 100 has more than doubled since 2015, from $1.61m to $3.59m, and the most profitable firms have gone from about $6.6m per partner to an eye-popping $12m. London’s magic circle has ridden the same wave, with profit per partner up from roughly £1.4m in 2015 to £2.48m at Linklaters today, and reportedly nearer £4m at the top end. Whatever AI is doing to the cost of producing legal work, remarkably little of it is reaching the people who pay the invoices.


There is a reason this pattern is so stubborn. The legacy firms are not simply slow to adopt the new model; they genuinely cannot escape the old one. A law firm is a partnership: its profits are paid out to the equity partners each year, not retained to create and fund innovation and improvements. Genuine AI capability takes serious, sustained investment, and every dollar of it comes straight out of money that would otherwise be distributed as partner take-home pay. Worse, if the investment succeeds, the prize is a tool that does in minutes what the firm used to bill for by the hour, cannibalizing the very revenue the partnership lives on. Asking a leveraged partnership to forgo distributions to fund a tool that shrinks its own billings is asking it to vote for a pay cut today in order to disrupt itself tomorrow. Most, understandably, will pass on such a proposal. That is why the reflex is to bolt on someone else's tool rather than build one, and why the firms most likely to change how legal work is priced are the ones with no partner profits, and no billable hour, to protect.


The real question


The firms furthest ahead have drawn a sharper conclusion. AI is not a technology cost. It is a staffing cost. An agent that runs diligence from end to end is not software; it is capacity. It belongs in the leverage model beside the associates, attributed to matters, budgeted like headcount, and measured against the work it replaces. Rate cards and matter budgets built around human timekeepers now have to describe a blended team of people and machines. Clients expect it, too: 85% think AI will reshape law-firm billing, and Deloitte's respondents expect the share of work billed by the hour to fall from 72% to 44% within three years.


But notice what even that sophisticated view can quietly preserve. Drop an agent into the pyramid, in the slot where a junior used to sit, and you still have the pyramid: the review layers, the handoffs, the supervision, and the markup the whole structure was built to justify. You have made an increasingly outdated operating model more efficient without changing the model itself. The honest question underneath all of this was never simply who pays for the AI. It is what legal work actually costs to produce now that so much of it can be produced differently. That question cannot be answered from inside the structure AI is making obsolete. It can only be answered by starting from the new cost of production and building outward.


Where the savings reside


AI-enabled savings in legal services do not come from the tools a firm licenses. Rather, they come from how the work itself is produced. A faster junior is still a junior on the client’s clock. Genuine savings require a fundamentally different approach to the practice of law, which in turn requires a different kind of firm: one designed around the economics of AI rather than retrofitted to accommodate them.


Novara was established on precisely this premise. Rather than asking how AI can improve the economics of a traditional leveraged partnership, Novara focuses on what the cost of producing high-quality legal advice should be when AI is part of the production process from the outset. This leads to a model built around today’s economics, rather than yesterday’s leverage structure that legacy firms cannot escape from.


At Novara, there is no traditional pyramid of associates beneath each partner. Matters are led throughout by experienced senior lawyers. AI provides much of the leverage that previously required large teams of junior associates. Now, junior lawyers are only utilized where they add genuine value, rather than because the economics of the partnership require the recovery of training costs through billable hours. Large legacy firms are structurally prevented from adopting this approach because it would mean dismantling the leverage their profits are built on.


What sets Novara apart is not simply its use of AI, but the combination of two capabilities that rarely exist together. The first is the judgment that only years of practicing at the highest level can develop; the second is a sophisticated command of the AI tools that are reshaping legal practice. Every senior lawyer leading a matter acquired the former at some of the world’s most elite law firms long before AI emerged as a meaningful part of legal practice. The firm’s training is built around both disciplines, but in the appropriate order: judgment first, because judgment remains the one capability technology cannot replace.


For clients, the distinction is significant. Efficiency becomes part of the firm’s operating model, and passing those savings on to the client is not a commercial concession to be negotiated; it is the very foundation of the business model. Clients retain what they value most from leading international firms: senior, experienced legal judgment coupled with impeccable client service. What falls away is the overhead that adds cost without adding commensurate value: large teams of juniors learning on large matters, multiple layers of review, repeated handoffs, and the coordination that such structures inevitably require. The quality of the legal advice remains, and the efficiency appears where the client had always expected to see it: on the invoice.


None of this requires predicting the end of the billable hour or the collapse of the established firm. Both have been forecast for years, yet neither appears imminent. The more significant change is already underway. Clients are increasingly asking where AI-enabled savings are actually being realized, and discovering that the answer depends far less on which technology a firm has purchased than on how the firm is organized. They are increasingly also acting on those conclusions. Thomson Reuters reports that nearly a third of clients are reconsidering their law firm relationships in search of demonstrable gains in efficiency. Technology-driven efficiency is real. Whether its benefits reach the client’s invoice is a question of structure, not software.


That leaves clients with a straightforward question to ask of every firm, on every matter, and preferably before the engagement begins: what does this work cost to produce today, and where do the savings go? A firm genuinely built around the new economics should be able to answer that question clearly, and demonstrate its answer through its fees. Others will try to explain why their invoices have changed so little despite the technology behind them changing so much. For clients deciding whom to entrust with their next transaction, the distinction is becoming increasingly difficult to ignore.


Sources:

Thomson Reuters, 2026 Future of Professionals Report, June 2026. https://www.thomsonreuters.com/en/institute/future-of-professionals-2026/report

SurePoint, Law School Hiring Report 2026 (reported by Artificial Lawyer, July 2026). https://www.artificiallawyer.com/2026/07/20/new-associate-hiring-is-flat-is-ai-the-cause/

Wells Fargo Legal Specialty Group, Year-End 2025 Survey (reported by ABA Journal, 29 January 2026). https://www.abajournal.com/news/article/revenue-billing-rates-and-demand-all-up-at-law-firms-in-2025-survey-shows

Comments

bottom of page