Who Gets the Savings? AI, Law Firm Economics, and the Conversation Clients Are Already Having

Clients are starting to ask three questions of their outside counsel: Are you using AI? Can you show me what it saved? And how much of that savings is mine? Firms that can answer all three are the ones with a working AI governance program underneath them, because governance is what turns "we use AI" into a documented, measurable, defensible account of where the value in legal work now sits. A firm that opens this conversation with clients on its own terms, before a revised set of outside counsel guidelines opens it, keeps the ability to shape the answer.

That is the short version. The longer version starts with what each side of the table is seeing.

Both Sides of the Table Already Have Numbers

On the client side, the expectation has arrived ahead of the paperwork. In the Thomson Reuters Institute's 2026 Future of Professionals legal report, 71 percent of in-house professionals said they expect law firms to change how they charge as AI use grows, and only 15 percent expect firm fees to hold steady or rise. Thirty-two percent said they are already reconsidering, or will within twelve months, relationships with firms that cannot demonstrate clear AI-enabled value, and a third of that group put more than $1 million in annual work on the table.¹ BCG's 2026 Legal AI Survey found that 57 percent of firms with more than 200 attorneys report clients asking about AI's effect on fees, and 81 percent of legal professionals expect AI to materially change the law firm business model within three to five years.² Litera's spring 2026 sentiment survey put it more bluntly: 85 percent of firms already feel, or expect, direct client pressure on their AI strategy, and a former Skadden chairman summarized the dynamic as clients expecting "to share in the AI productivity dividend" and having "the leverage to ask for it."³

The leverage is now being used in the open. In September, the Financial Times reported that Goldman Sachs, Morgan Stanley, and Citigroup are pressing their outside law firms to cut fees as AI shortens routine work. Citigroup rebuilt its bidding process to ask firms how much AI is saving them, and its global head of legal said the bank's "expectation is for costs to come down significantly per transaction." Morgan Stanley's general counsel said most of the bank's outside work would move to competitive bidding and alternative fees by year end, and called the current compensation model "extraordinarily unstable." Goldman is examining how AI efficiencies should show up on its bills. The backdrop is an average associate rate at the largest firms of $798 an hour, up 33 percent since 2023.¹⁰ When the most sophisticated legal buyers in the country make the question part of the RFP, the rest of the market follows.

On the firm side, pricing has barely moved. The same Thomson Reuters report found that only 28 percent of law firms have changed their pricing structure in response to AI.¹ Clio's 2026 solo and small firm report found that 86 percent of solos and 78 percent of small firms have not changed pricing because of AI at all, and fewer than a third of them report any revenue gain from it, against nearly 60 percent of enterprise firms.⁴ Litera found that 32 percent of firms cannot confidently demonstrate AI value to their most important client.³

The written guidelines are the last thing to catch up. Fulkerson Advisors reviewed 1,054 outside counsel guidelines published on the open web since the release of ChatGPT. Only 20 of them, about 2 percent, mention AI. Of those 20, twelve require the firm to disclose AI use, usually in time entries or invoices; nine change how AI work is billed; three expect the savings to show up as fewer billed hours; and two refuse to pay for time spent using or learning AI tools.⁵ The sample skews toward public bodies and universities, which write their guidelines in public, so the corporate figure is almost certainly higher. But the shape is instructive: the expectation is widespread, while the contractual language that enforces it remains rare and is being drafted now.

The window between a client expecting something and a client writing it into the guidelines is where a firm sets the terms instead of receiving them.

Why the Current Model Cannot Absorb This Quietly

The Thomson Reuters Institute's 2026 State of the US Legal Market report found that roughly 90 percent of legal dollars still flow through hourly billing.⁶ When a tool finishes in twenty minutes what used to take four hours, an hourly firm has exactly two choices. Bill the twenty minutes and lose the revenue, or bill something closer to four hours and create a problem under Rule 1.5.

The second choice is closing fast. ABA Model Rule 1.5 requires that fees be reasonable, and the reasonableness factors start with "the time and labor required."⁷ The Texas Professional Ethics Committee's Opinion 705 made the application to AI concrete in February 2025: under hourly billing, efficiencies gained through AI have to benefit the client financially, a lawyer cannot bill for time that was not worked, and the reasonable costs of AI tools can be passed through only with the client's advance agreement.⁸ Other jurisdictions are reaching the same conclusions by the same route, and the ABA's own nonbinding guidance says the same thing in the same words.⁹

BCG calls the result a pricing paradox: "Firms that layer AI onto current workflows risk accelerating margin compression instead of preventing it."² The firm pays for a tool that compresses the very hours it bills. Meanwhile corporate legal departments are deploying their own AI against the other side of the same transaction, running invoices through automated guideline enforcement that flags unapproved rates, duplicate entries, and vague narratives at a scale no human reviewer could match. A firm that does not adapt its pricing is getting squeezed from both directions at once.

The hourly model is unlikely to disappear next quarter, but a firm can no longer treat the question as hypothetical, and the answer for a given firm and a given client has to rest on something more solid than a rate card.

Governance Is the Foundation, Because Governance Is the Evidence

Here is the argument we make to every firm we work with. The value of AI-assisted legal work is real, and it is defensible, but only if the firm can show where the value sits. That showing depends on governance work that has to be done before any pricing decision is made.

Consider what a client is asking when it asks a firm to demonstrate its efficiency gain. It is asking which tools were used, on which tasks, by whom, with what data, reviewed by whom, against what standard, and how long each step took. A firm that cannot answer those questions has no gain to show and no basis to defend the fee if it is challenged. Answering them gives the firm something better than a discount to offer: a documented account of what the lawyer contributed.

Every one of those answers comes from a governance artifact that a well-run program produces anyway.

An inventory of what the firm uses and where

The first discipline in any serious AI governance framework is an inventory: a register of every AI system in use, the use cases it serves, who operates it, what data flows through it, and who owns the outcome. A firm cannot govern what it has not mapped, and it cannot bill transparently for it either. The inventory is what lets a firm say to a client, with a straight face, "here is every tool that touched your matter and here is where your data went." It is also what surfaces the unsanctioned tools: Thomson Reuters found that 34 percent of law firm professionals use AI their firm has not authorized.¹ That is a Rule 1.6 exposure, and it is also a pricing gap, because the firm cannot account for work it cannot see.

Principles a client can hold the firm to

A governance program that says "we use AI responsibly" has said nothing a client can hold the firm to. The discipline that works runs values down into principles, principles into commitments, and commitments into measurable specifications. "AI output is verified by a licensed attorney before it reaches a client" is a commitment. "Every AI-assisted work product carries a logged verification step with a named reviewer" is a specification. A client reading the second sentence knows what it is paying for, and a client reading the first has been handed a slogan.

Named accountability and supervision

Model Rules 5.1 and 5.3 already require lawyers to supervise the people and the nonlawyer assistance that contribute to a matter, and Rule 1.1's competence duty, read with Comment 8, requires understanding the technology used to deliver the work.⁷ A governance program operationalizes those duties: a named system owner for each tool, a defined review standard for each category of output, an escalation path when the tool gets something wrong. Those are the same facts a firm needs when a client asks why a document that an AI drafted still carries a professional's fee. The answer is that the judgment, the verification, and the accountability were the lawyer's, and the firm can show the record.

Transparency as an ongoing practice

Rule 1.4 requires lawyers to keep clients reasonably informed about the means by which their objectives are pursued.⁷ Done well, that duty produces a documented account, maintained as the matter proceeds, of where AI was used and how it was checked, rather than a single line in an engagement letter. Firms that build this into their workflow are producing exactly what the newest outside counsel guidelines are beginning to demand, before being asked for it.

One more point about measurement. Thomson Reuters' advice to firms is to "measure behavior change, not adoption rates," and the report notes that roughly a fifth of professionals given professional-grade tools never use them.¹ Tracking seat licenses tells a firm what it spent; tracking which workflows run through AI, what they cost, and what lawyer time they displaced tells the firm what it has to offer a client and what it needs to keep. That second kind of measurement is what makes a pricing conversation possible.

What the Conversation Should Look Like, and Why It Should Happen Now

The reason to open this conversation early is leverage. A term the firm proposes, backed by an inventory and a verification record, is a negotiation, while a term the client imposes in the next revision of its guidelines, with a 30-day acceptance window, is a condition. The firms that treated AI disclosure as something to handle when asked are the ones now reading guidelines that refuse to pay for AI time at all.

The conversation itself has a manageable shape. The firm brings five things.

What the firm uses, and where client data goes. The inventory, in plain terms: which tools, on which categories of work, under what retention and training terms, and which matters are walled off from AI entirely. Clients whose own guidelines now ask where their data travels will respond to a firm that can answer before the question is posed.

How the fee reflects the tool. For hourly work, the firm bills actual time and says so. For work where AI has changed the economics, the firm offers a structure that reflects it: a flat fee for the defined deliverable, a blended rate, a subscription for recurring work, or a shared-savings arrangement. The Clio report found that 71 percent of clients already prefer fixed or flat fees.⁴ AI gives the firm, for the first time, the unit economics to offer them without guessing.

Who pays for the tooling. Opinion 705 and the ABA's guidance converge here: general-purpose tools are overhead, matter-specific usage can be an expense, and either way the client agrees in advance.⁸ Settling this up front removes the single most common source of invoice disputes about AI.

The verification standard. What gets reviewed, by whom, at what level of seniority, and what the client can expect to see in the file. This is where the firm makes the case that the work has value: the output may have been drafted faster, but the judgment that stands behind it, and the liability that attaches to it, remain the lawyer's.

What the firm needs in return. The complete version of this conversation includes the firm's economics. Professional-grade tools cost money, training costs money, governance costs money, and a client that wants all of the efficiency gain and none of the cost has not proposed a viable arrangement. The goal is a structure that is economically sound for the firm, cheaper or faster or better for the client, and safe for both: confidentiality preserved, output verified, fees reasonable on their face.

The output of that conversation should be written down. A short AI addendum to the engagement letter, or a firm-proposed rider to the client's guidelines, covering tools, data handling, disclosure, verification, and fee treatment. A page or two is enough, as long as it exists before someone else drafts it.

Where This Leaves a Firm

The BCG survey found that firms actively redesigning their workflows around AI report significant value at scale at nearly twice the rate of firms still running pilots.² If that gap holds, it shows up in client retention before it shows up in profits per partner. The firms in good shape are the ones that have stopped treating AI governance as a compliance exercise and started treating it as the operating record that makes their pricing defensible and their client conversations easy.

At InGlobo AI, that is the work: building the inventory, writing the operational commitments, designing the verification workflow, and helping firms walk into the client conversation with a record instead of a reassurance. The collision between the old economic model and the new tools is coming either way. What a firm controls is whether it meets that collision with its own terms in hand.