AI and the Billable Hour: Can You Bill for Time AI Saved, and How to Price Instead

Lawyer bills $300 an hour for estate planning. Standard documents used to take about three hours to draft; with AI they take one. May Lawyer bill three hours? The North Carolina State Bar’s answer, in its 2024 Formal Ethics Opinion 1, is the plainest sentence written on AI and the billable hour: “No, Lawyer may not bill a client for three hours of work when only one hour of work was actually experienced.”

That settles the ethics question, not the business one: what to do with the two hours. Most firms have done nothing: Clio’s 2026 data show 86% of solos and 78% of small firms made no pricing change after adopting AI, and only 32% and 31% have grown revenue since. Wall Street’s general counsel are now asking why.

The ethics answer: bill the hour it took

ABA Formal Opinion 512 (29 July 2024) is the reference text. Hourly billers “must bill for their actual time”. A lawyer who “expends 15 minutes to input the relevant information into the program… may charge for that time as well as for the time necessary to review the resulting draft”, but “in most circumstances, the lawyer cannot charge a client for learning how to work a GAI tool”. And: “A fee charged for which little or no work was performed is an unreasonable fee.” And a warning that reaches flat fees: if a tool “enables a lawyer to complete tasks much more quickly than without the tool, it may be unreasonable under Rule 1.5 for the lawyer to charge the same flat fee”.

The states are more specific, not less; the state-by-state map covers consent and disclosure too.

The North Carolina hypothetical, and the sentence after the “No”

The sentence after the “No” is the business advice: “Lawyer may enjoy the benefit of those new efficiencies by completing more work for more clients.” The opinion then opens the door most firms have not walked through: flat fees are permitted “provided the flat fee charged is not clearly excessive and the client consents”, and AI expenses may be passed through if “accurate, not clearly excessive, and the client consents to the charge, preferably in writing”. The floor is not “give the client the hours back”; it is “do not pretend the hours happened”. A consented flat fee for the same package, priced on today’s cost, is compliant everywhere in the table above.

The paradox: billable hours rose 4.2% anyway

If AI were eating the hour, the numbers would show it. Half-year figures reported by Best Law Firms in September 2026 put industry billable hours up 4.2% in the first half of 2026 against a norm of 1.5-2%, with AI investment at about 0.25% of revenue. Thomson Reuters’ 2026 State of the US Legal Market found worked rates up 7.3% in 2025 and about 90% of legal dollars still flowing through hourly arrangements. The verdict: “This creates an almost absurd tension that sees firms deploying technology that can accomplish in minutes what once took hours, then trying to bill for it by the hour.”

BigHand’s 2026 survey of 800-plus finance leaders shows how the tension is managed: 96% raised rates, nearly two-thirds report fewer billable hours, 99% plan to raise targets, and only 34% updated pricing to reflect AI. Predictions of the hour’s death are old; Crosby’s co-founder notes that “people have been predicting the death of the billable hour for, like, 70 years… it’s just really durable”.

Client pressure: Wall Street banks and Meta’s Haven

What changed in 2026 is who is complaining. On 1 September the FT reported that Goldman Sachs, Morgan Stanley and Citigroup are pressing outside counsel to cut fees because of AI. Citi’s global head of legal, Adam Meshel: “If the number of hours they’re working on a matter has come down because of AI… our expectation is for costs to come down significantly per transaction.” Morgan Stanley’s general counsel Eric Grossman said top lawyers’ “compensation model is now extraordinarily unstable” and is moving most outside work to competitive bids with fixed-fee alternatives; associate rates are up 33% since 2023, to $798 an hour.

Meta’s head of legal operations, Mike Haven, told the CLOC Global Institute in May 2026 that hourly billing will be “the exception, not the rule” within five years and that he already asks firms for fixed fees: “Now, I am telling top law firms that you need to do this for you.” Keep the denominators straight, though: Axiom found 92% of in-house teams expect or are negotiating AI-related rate cuts, while 8am found only 6% of legal professionals see clients explicitly pushing for them. The pressure is loud at the top, latent elsewhere, and latent pressure shows up as attrition. The in-house guide explains the other side’s arithmetic: an hour saved in a legal department is capacity, not lost revenue.

Four pricing models for AI-assisted work

Clio found 71% of clients prefer flat fees for entire cases; among wide AI adopters, 45% have adjusted pricing.

Token pass-through: MyShingle and the price of Fable 5

The pass-through question became concrete in July 2026, when Anthropic’s expected shift of Claude Fable 5 to metered usage put a number on it, about $10 per million input tokens and $50 per million output tokens, roughly twice Opus 4.8 ($5 and $25). Carolyn Elefant’s MyShingle Ethics Opinion 2026-1 synthesised ABA 512 and the state opinions into a rule: subscriptions are overhead and may never be billed; metered charges may be billed only when attributable to a specific matter, at actual cost, without markup, disclosed in advance in writing.

Her policy advice goes further: absorb ordinary AI cost as overhead; pass through only extraordinary single-matter consumption. “A lawyer whose ordinary AI costs rise should raise rates or adopt flat fees rather than itemize tokens.” Her precedent is Westlaw: per-search charges were client expenses until flat “commitment pricing” turned them into overhead, and in 2021 a large firm was sued over $100,000-plus of per-search charges billed against a flat contract. Tokens will follow the same arc.

Clients now write the same rules into billing guidelines, with the invoice notation “AI-assisted; attorney reviewed”. What the tools cost, tier by tier, is in legal AI pricing; the client-side clauses are in the OCG guide.

Rewrite these time-entry narratives <entries>...</entries> so each states the task performed, the document or issue, and the purpose, in the client's required format <ocg_format>...</ocg_format>, without changing the time recorded or adding tasks. Flag any entry that looks like block billing or that the guidelines would reject, and any entry that should carry "AI-assisted; attorney reviewed" under the client's convention. Do not merge or split entries, and do not adjust any number: if an entry reads 0.2, it stays 0.2.

CHRGE versus Rapoport and Tiano

The academic argument decides whether you have a revenue problem or a pricing problem. Georgetown’s Jonah Perlin, in “How the Billable Hour Can Survive Generative AI” (2025), writes revenue as the CHRGE equation, Compensation = Hours × Rate − Granted adjustments − Expenses, and argues that “fewer hours means less revenue” is the lump-of-labour fallacy: demand is not fixed, and Jevons Paradox suggests cheaper work gets bought more. He quotes a partner: “So what if you charge $5,000 an hour if it’s the same amount that they expect to pay.”

Rapoport and Tiano’s “Fighting the Hypothetical” (2025) takes the other side; its subtitle is the thesis, “Why Law Firms Should Rethink the Billable Hour in the Generative AI Era”. And Thomson Reuters’ corrective for anyone planning to charge more because AI made the work better: “more marketing hype than a legitimate strategy.” My reading: both are right, and nobody needs to settle the debate to reprice one matter type this quarter.

Sebastian’s rule: split the savings

Clio’s 2025 Legal Trends Report quotes a partner at a small California real-estate firm, Sebastian, on what he actually does: generate the work with AI, estimate how long it would have taken manually, and bill a figure between the two. “I split the savings between the client and myself so that everybody benefits.” An Indiana partner, Ezra, states the floor in the same report: “If you can do the same function in a quarter of the time with AI, you can’t ethically pass that on to your client.”

Clio’s advice is one sentence: “Price one predictable matter type, such as a residential closing or a simple will, as a flat fee against what it actually costs you to deliver today, with AI doing its share.” Then “give the saved hour somewhere to go”: intake, unbilled-time review, referral work. The business development pillar covers that half; efficiency without new work is a pay cut.

Help me price [a residential closing / a simple will package / a standard NDA review] as a flat fee. Here is our time data for the last [20] such matters <data>...</data>, with hours by task and our rates. Compute the mean, median and 80th percentile of hours and cost; identify the three drivers of the outliers; propose a scope definition with explicit exclusions; propose a flat fee at [target margin] with an add-on schedule for the exclusions; and draft a two-paragraph client-facing scope description in plain English. Show the arithmetic in a table. Do not assume any time saving from AI that is not in the data.

Check the arithmetic; models miscount. And price on today’s cost, not last year’s: the ABA’s caveat is that a flat fee AI makes trivially fast can itself become unreasonable.

What to write in the engagement letter

ABA 512 says the engagement agreement “is a logical place” to disclose AI use and record client instructions, and warns that “merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient”. The letter needs four things: which tools touch client information; the billing rule; any metered pass-through, at cost and with consent; and the fee model, with scope if flat. Elefant’s sample clause is the model: charges “directly attributable to your matter”, billed “at the Firm’s actual cost, without markup”, and no charge where the attributable cost cannot be determined.

Draft the fees section of an engagement letter for [matter type] under [jurisdiction] rules that: (1) states our fee model (flat fee of [amount] for the scope in <scope>...</scope>, or hourly at actual time recorded); (2) explains that we use generative AI on enterprise terms that do not train on client information, that a lawyer reviews all output, and that we bill only time actually spent, never time spent learning tools; (3) treats subscriptions as overhead and bills metered AI charges directly attributable to the matter at actual cost, without markup, itemised, with no charge where the cost cannot be determined; (4) gives the client the right to instruct us not to use AI. Plain English, under 300 words; mark any sentence that depends on a state rule with [CHECK: state].

Where to go next: the ABA Formal Opinion 512 guide covers the fee passages in full, AI for RFP responses covers fee modelling before you bid, and the business development hub holds the rest of this cluster, with the prompts in the prompt library. Faster and not repricing is “an 80% discount you never asked for”, which is why the homework in AI Lab for Lawyers includes pricing one matter type flat, on your own time data.