Selling Hours When Hours Shrink

Make the junior staff of a consulting or law team 40 percent faster at their work, keep everything else the same, and keep billing by the hour. Fees fall 17 percent. Profit falls by a third, even if the firm cuts its staff to match the smaller workload perfectly. If it keeps everyone, profit falls by nearly three-quarters.

That's the arithmetic of selling time in a year when time gets cheaper, and it surprises people who have never looked at how these firms make money. Faster people should be good news. Under hourly billing they are a discount the firm gives away.

Here is the argument in two sentences. A firm that bills by the hour passes every efficiency gain to its client at billing rates, which is much more than the firm actually saves, so its profit falls faster than its revenue. And the hours AI removes first are the junior hours that have always trained the next generation of partners, so the firm has to change what it charges for and decide, on purpose, who pays for the apprenticeship.

How a firm that sells time makes money

Law firms, consultancies, accounting practices and IT-services companies mostly sell hours. Revenue is hours worked times an hourly rate, summed over everyone who bills. Profit is a markup: each person's rate sits well above what the firm pays for that hour, and the partners own the spread on everyone below them.

That's why these firms are shaped like pyramids. A few partners at the top, more managers, and a wide base of juniors doing the drafting, the research and the first pass of document review. The firm measures this shape with the leverage ratio, the number of non-partners per partner. More base means more marked-up hours for each partner.

Take an illustrative engagement team, with numbers chosen to be plausible rather than taken from any firm. Three partners bill $600 an hour, five managers $350, and twelve juniors $160. In a year they bill 30,900 hours for $7.69 million in fees, and after paying everyone, partners included, and the overhead, the team keeps $1.79 million in profit.

Two facts about this team drive everything that follows.

Juniors do two-thirds of the hours. And a junior hour bills at $160 while costing the firm about $88 including overhead. The base of the pyramid is where most of the hours sit and where a large share of the markup is earned.

It's also where partners get made. A junior who spends years drafting, reviewing and being corrected learns what a good document looks like. Nobody calls that a training budget. The client pays for it, one billed hour at a time.

What happens when the base gets faster

Now give the same team AI that does the first pass of drafting and review. The tasks AI handles best are checkable, written down and repetitive: first-pass review, summaries, routine drafting, research. Those are precisely the tasks that fill the junior years (A Job Is a Bundle of Tasks).

Assume junior hours for the same deliverable fall 40 percent. Managers spend as long as before, because checking AI output replaces the drafting they no longer do. Partners spend the same time. (The 40 percent is an assumption. In a 2023 field experiment, Boston Consulting Group consultants using GPT-4 on tasks within the model's reach finished more than 25 percent faster. A bigger or smaller cut changes the size of what follows, not its direction.)

Scroll sideways to see the whole figure

The engagement now needs 22,740 hours instead of 30,900. The base shrinks from twelve juniors to the equivalent of about seven, and the top doesn't move.

Three ways to price the same work

The firm has three choices for what to charge.

Keep billing by the hour. Fees fall 17 percent, to $6.38 million. If the firm cuts junior staff to match, profit drops by a third, to $1.20 million. If it keeps all twelve juniors busy on less work, profit collapses to under half a million.

Quote a fixed fee at the old price. The client pays $7.69 million as before, the firm resizes the team, and profit rises 40 percent, to $2.51 million. The client sees none of the gain.

Quote a fixed fee that shares the saving. The firm cuts its price by half of what hourly billing would have given away, about 8.5 percent. Profit ends slightly above where it started.

Scroll sideways to see the whole figure

The interesting number is hidden between the first two. Under hourly billing, the client's bill falls by $1.31 million. The firm's costs fall by only $0.72 million, the cost of the junior positions it no longer needs. The client receives 1.8 times what the firm saved.

Hourly billing hands the client the saving at billing rates, not at cost.

The mechanism is the markup. Every hour that disappears takes its margin with it. The firm saves what the hour cost it, $88. The client saves what the hour was billed at, $160. Multiply that gap by thousands of junior hours and profit falls about twice as fast as fees.

That also tells you the fair price, which is a useful number for both sides of the table. In this example, a fixed fee about 9 percent below the old price keeps the firm's profit exactly whole. Charge more and the firm keeps part of the gain; charge less and the discount comes out of the partners' pockets. The firm and its client are negotiating over the space between zero and 17 percent off, and almost nobody has done the arithmetic to know where the middle is.

Why it hasn't happened yet

A fair reader will object that this has been predicted before. Spreadsheets didn't end accounting hours. A US federal court first approved computer-assisted document review in 2012, and discovery work is still billed. Offshoring cut the cost of a junior hour dramatically, and the big IT-services firms grew into some of the largest employers in the world.

The recent numbers support the objection. US law firms grew demand, measured in billable hours, about 2 percent in 2025, raised rates 7.3 percent, the fastest in years, and increased profit 13 percent. They added lawyers while spending nearly 10 percent more on technology. Accenture's revenue for fiscal 2026 rose 5 percent and it added people. About 90 percent of US legal spending still moves through hourly arrangements. None of this looks like a squeeze.

Two things explain it, and neither lasts forever.

Some hours were never being paid for. Associates, the junior lawyers, had the lowest realization of any group in 2025: on average clients paid about 86 percent of the value of the hours they worked, and the rest was written off. AI can absorb that unpaid slice without touching collected revenue. The pressure starts when savings exceed what clients weren't paying for anyway.

Clients and firms are in a standoff. Corporate legal departments say they want firms to propose billing that reflects AI efficiencies. Firms say procurement just divides any quoted total by estimated hours and compares the result with last year's rates, so a value-based price gets turned back into hours. Each side is waiting for the other to move.

The precedents teach something specific, too. Spreadsheets, e-discovery and offshoring all made a unit of work cheaper while the market wanted ever more units: more models, more documents, more projects. Economists call that elastic demand: when the price drops, people buy enough more to keep total spending up. Hours survived because the work expanded.

This time has three differences. AI cuts the number of hours a deliverable needs, where offshoring cut the price of each hour and kept the pyramid intact. Clients hold the same tool, so they can price the first pass themselves; corporate legal departments have been ahead of their outside firms in using generative AI since 2022. And some work doesn't expand. To restore its fees under hourly billing, our team's client would have to buy about 20 percent more of the deliverable. That's plausible for due diligence across more targets. It's implausible for an annual audit.

Hours survive where the work expands. The question for each practice is whether more of it can be sold.

The apprenticeship bill

The second problem is quieter and, in my view, more serious.

Keeping all twelve juniors, so that they still get years of practice, costs the example team about $720,000 a year, 40 percent of its original profit. Under hourly billing, that money no longer exists: the hours that funded the training are gone and nothing replaces them. Under a fixed fee at the old price, the firm could keep the whole pyramid and match its old profit, which means the cost of training equals exactly the gain it kept. Under a shared fee, the partners choose between profit and pipeline.

A smaller base also changes who becomes a manager. With some plausible assumptions about how long people stay, the firm used to promote about 28 percent of each group of departing juniors. With the smaller base it has to promote about 46 percent, chosen from fewer people who each had 40 percent fewer hours of practice.

There is a collective trap here too. A firm that trims its intake and later hires experienced people from rivals benefits from training somebody else paid for. If every firm does it, the supply of trained people shrinks for all of them. Payroll data already shows employment of 22-to-25-year-olds in AI-exposed occupations running about 19 percent below where it would otherwise be, mostly through fewer hires, though the authors call the result descriptive, not causal (Hiring When Producing Is Cheap).

My view: training was always a cost. Hourly billing simply hid it inside the client's invoice. Once those hours disappear, the cost becomes visible, and a firm that doesn't budget for it explicitly will discover in ten years that it has no one ready to be a partner.

What stays scarce, and who enters

Three things resist compression because they are not hours. Accountability: someone whose name and insurance stand behind the work. Judgment: choosing among options when the facts are incomplete. Access: the relationship and the permission to be in the room, whether that room is a regulator's, a board's or a client's systems.

The clearest illustration comes from England. In 2025 the Solicitors Regulation Authority approved Garfield.Law as the first purely AI-based firm allowed to provide regulated legal services. The system takes each step only with the client's approval, and named solicitors remain accountable for its output. The regulator let the hours go to a machine and kept a person on the hook.

So the market splits. Work whose value is mostly hours, such as first-pass review, content variations or research decks, will see prices follow costs down. Work where the hours shrink but a signature stays, like a regulatory filing or an audit opinion, should be priced for the assurance, with fewer hours behind it. Work sold on who answers, like crisis response or bet-the-company advice, is the safest place to be.

New entrants are what will force the issue, because they set the reference price. In the example, an AI-native firm with the smaller team and the incumbent's old margin could quote about 12 percent below the old price. The incumbent sharing half the saving quotes about 8.5 percent below. The entrant's 4 percent advantage has to buy the trust, track record and regulatory standing it lacks, and sometimes it will.

Underneath sits a problem economists call counter-positioning. Partners are paid out of the markup on junior hours. A model that doesn't depend on that markup is hard for a partnership to copy, because copying it shrinks the pyramid that pays the partners. Partnerships can change how they pay themselves, but it takes years, and the entrants' cost structure is available now.

Firms are already moving. EPAM, an IT-services company, billed about a fifth of its 2025 revenue on fixed prices, up from about an eighth two years earlier. Accenture and Cognizant have run restructuring programs mostly paid in severance and framed partly around AI. Accenture and Anthropic launched a joint business that embeds engineers with clients to make AI work in their real processes, which sells deployment instead of drafting hours (What a Deployment Engineer Should Own, and Leave Behind). And Thrive Holdings, backed by OpenAI, has been buying accounting and IT-services firms to rewire them around AI.

What to do

If you run a practice, do the calculation with your own numbers. What share of your hours sits at the base, what's the markup on them, and at what fixed fee does a shared saving keep your profit whole? In this example the answer is about 9 percent below the old price. Yours will differ. Knowing it turns a defensive negotiation into an offer.

Move to fixed or shared fees before your client asks, and price the estimate carefully. A fixed fee moves the risk of a bad estimate from the client to you, and every competitor is estimating the same hours with the same tools.

Sort your revenue by what you actually sell. Hours, assurance, judgment or access. A firm whose profit sits in commodity hours has a different problem from one whose profit sits in signatures and relationships, and it should be investing differently.

Put the apprenticeship in the budget. Decide how many juniors you train, what they practice on now that the drafting is automated, and which line of the profit and loss pays for it.

If you buy these services, ask for the number. The saving hourly billing would pass you is roughly twice the firm's cost saving. A fixed price that shares around half of it leaves the firm whole and cuts your bill by close to a tenth. Naming that number beats waiting for the other side to blink (Who Receives the Gain?).

The firms that do well won't be the ones that adopt AI fastest. Everyone will adopt it. They'll be the ones that stop selling the hours AI removes, before a client or a competitor does it for them.

Check yourself

Why does profit fall about twice as fast as fees under hourly billing?

Each hour that disappears takes its margin with it. The firm saves what the hour cost it, while the client saves what the hour was billed at.

Why is the loss of junior hours a problem beyond profit?

Those hours were how juniors learned what a good document looks like, and the client paid for it one billed hour at a time. Once the hours are gone, the training cost is visible and someone has to budget for it.

Why did earlier technologies not shrink hours the same way?

They made a unit of work cheaper while demand for units grew. This time AI cuts the hours a deliverable needs, clients hold the same tool, and some work, like an annual audit, doesn't expand.