When AI cuts the hours, what do you charge for?

A firm that uses AI to turn ten hours of work into two, and still bills by the hour, has just cut its own invoice by eighty per cent for the same result. The better it gets at AI, the less it earns. That is the quiet problem sitting underneath every productivity case study, and most firms have not decided what to do about it. Deciding means answering a question the hourly model let firms avoid for a century: what are you actually selling once AI does the doing?

Start with the mechanics, because they are blunt. The hourly model pays you for time. AI's whole effect is to take time out. Put the two together and you get an efficiency discount: every hour the tool saves is an hour you can no longer put on the bill. A piece of comparable-company research that took an associate a day now takes an afternoon. A first draft of a report that took six hours comes back in ninety minutes, good enough to edit. The work is the same quality or better, the client is just as well served, and the fee falls. A firm that has not touched AI looks more expensive for no good reason, and for a while it earns more for being slower. That cannot hold.

So it is worth asking what the hour was ever really standing in for. No client has wanted hours. They wanted a deal that closes, accounts that are right, advice they can act on without checking it themselves, and someone who carries the risk if it is wrong. The hour was a proxy, a rough and convenient way to price all of that, on the reasonable assumption that better work took more time. AI breaks the assumption. Now the best work can take the least time, and the proxy starts to mislead in both directions: it underpays the firm that has built something genuinely fast, and it lets a slow firm overcharge for plodding. When the measure stops tracking the value, you stop measuring value and start measuring effort.

This is not a thought experiment any more. A Thomson Reuters survey of more than 1,500 professionals across legal, tax and accounting found that organisation-wide AI use almost doubled to 40% in 2026, while only 18% of firms track the return on the tools at all, and a rising share now expect AI to land directly on billing and revenue. The tools are in the building; the commercial thinking is lagging behind the usage. On the client side, the expectation is starting to form too. In an open letter published through the Harvard Law School Forum, Salesforce's chief legal officer argued that clients now expect the savings from AI passed to them rather than kept as firm margin. He is a large buyer of legal services talking his own book, so weigh it as one interested party rather than settled fact. But it is the direction of travel. A client who knows the report was largely AI-assisted will, sooner or later, ask why the bill looks the same as last year.

If the hour no longer measures the value, what does? Three things are worth pricing for directly, and they are genuinely different from each other.

The first is the outcome. Some work has a clear deliverable a client would happily pay a fixed price for: a filing, a set of accounts, a due-diligence pack, a contract reviewed and marked up. Price the thing, not the time it took to make it. Fixed fees are not new, but AI changes the maths underneath them. The work that used to make a fixed fee risky, the long tail of hours, is exactly what AI compresses, so the fixed price gets safer to offer and the efficiency gain stays with you rather than leaking out through a falling timesheet.

The second is assurance, which is another word for risk. A lot of what a serious firm sells is the confidence that the answer is right and that someone qualified stands behind it. AI makes a first draft cheap and makes the checking matter more, because a confident wrong answer is now easy to produce. The firms that came unstuck in the past year were the ones that let unchecked AI output reach a client, which is a different and avoidable failure. Pricing the assurance, the review, the sign-off, the named partner who carries it, reflects where the work has actually moved.

The third is access to judgement. Some clients do not want a deliverable at all; they want to be able to ask a good question and get a good answer quickly. That is a retainer or a subscription to your thinking, and it holds up well when the routine production is automated, because what is left is the part AI cannot do: knowing which question matters, reading a situation, telling a client the thing they did not want to hear. This is the same shift Ethan Mollick describes when he argues that management is becoming the core AI skill: as the doing gets delegated to the machine, the value moves to knowing what to ask for and checking what comes back. The doing gets cheap. The deciding does not, and it is the part worth keeping off the machine.

None of this is as tidy as a timesheet, and it is worth being honest about that. Hourly billing survived for so long because it is easy: you do not have to know what your work is worth, you just count. Pricing the outcome or the judgement forces a harder conversation, internally and with the client, about value you may never have had to name. It also depends on the AI-assisted work actually being good, which means the human check is not optional, it is the thing you are now charging for. A firm that drops the hour and also drops the rigour has not moved up the value chain, it has just found a faster way to be wrong. Few firms even track whether their AI use pays its way yet, which is the measurement gap sitting right next to the pricing one.

Working out where AI has changed the value of your own work, and what to charge for instead of time, is the kind of question an AI Lessons for Leaders session is built around: your actual workflows and your pricing, not a generic talk.

The decision to make is narrow and specific, which is the good news. You do not need to rip up how you bill across the whole firm this quarter. Pick the one or two services where AI has most changed how the work gets done, the ones where the hours have quietly fallen. For those, decide on purpose what you are charging for: the outcome, the assurance, or the access to your judgement. Then price it that way before the efficiency discount does it for you, or before a client opens the conversation on their terms. The firms that come out ahead will be the ones that worked out, before they had to, what a client was paying for all along, and priced for that instead of for time.

If you want a second pair of eyes on where this lands in your firm, book a short discovery call.

Frequently asked questions

Does this mean we should stop billing by the hour?

No. It means the hour stops being a safe default for work AI has sped up. For routine, well-defined work where AI now does much of the production, a fixed or outcome price usually serves you and the client better. For genuinely open-ended advisory work, time may still be the fairest measure. The point is to choose per service, not to apply one model everywhere out of habit.

Will clients expect us to pass on the AI savings?

Some already do, and more will. A client who knows AI did much of the drafting will eventually ask why the bill has not moved. The firms that handle this well decide in advance how the gain is shared, framing it as faster turnaround, more access, or a fixed price they can plan around, rather than waiting to be pushed into a discount with nothing offered in return.

Is value-based pricing realistic for a small firm?

It is often easier for a small firm than a large one, because there is no entrenched billing machine to unpick and the leaders setting prices are the same people doing the work. The hard part is rarely the mechanism. It is being willing to name what your work is worth. Starting with one service where the value is clear is a low-risk way to test it.