Who gets the hour AI saves

In-house legal teams have already formed a view on what AI should mean for the bill. Law firms have largely not made the corresponding decision, and the distance between the two can be measured inside a single survey.

The legal section of Thomson Reuters' Future of Professionals Report 2026, fielded in March and April 2026, rests on 736 survey responses from law firm professionals across 46 countries, plus a further 203 responses from corporate legal departments. Of the in-house respondents, 71 percent expect their outside firms to change their commercial models. 28 percent of law firms say they have made any change to their pricing structure in response to AI.

The evidence is American-heavy. Thomson Reuters draws 421 of its 736 law firm responses from the United States, and the Clio material further down is American. Neither source reports a Nordic subset. The figures describe the client expectation a Nordic firm meets, particularly from international clients, rather than Nordic firm structure itself.

The client's expectation on the bill is already set

The two figures above were put to two different groups within the same survey, and they measure different things: one an expectation, the other an action. What makes the distance interesting is that the expectation is priced. Among the in-house respondents, 15 percent expect fees to stay level or rise as AI use increases.

Confidence in delivery is low at the same time. 70 percent of in-house legal professionals consider it very important or essential that their outside firms can demonstrate AI-enabled productivity and efficiency. 6 percent believe most of the firms they work with are doing so. A client who expects a new commercial model, does not budget for higher fees and does not see the demand met has already framed its negotiating position.

The arithmetic behind it is simple enough for a whiteboard. Under hourly billing, every hour AI saves is revenue handed back. Under a fixed price, the same hour is margin that stays. The technology is identical in both cases. It is the fee model that decides who gets the gain.

Fixed pricing is not a new instrument

The demand is not primarily for a discount either. 33 percent of in-house respondents expect new pricing models offering greater cost certainty and transparency, the largest single group, and in practice a description of fixed pricing.

The objection to changing how work is priced is usually that the market is not ready for it. That claim can be tested against material measuring what firms actually offer, rather than what they say they intend to do.

Clio's review of mid-sized US firms, published in March 2025 and built on surveys fielded in April and June 2024 together with aggregated data from tens of thousands of American legal professionals, puts flat fees at 64 percent against hourly rates at 54 percent. Firms offer several models in parallel, so the figures should be read as what is on offer and not as shares of a whole.

The two sources measure different things and can therefore correct each other. Thomson Reuters asks professionals what they expect and what they have done. Clio counts what firms say they sell, in surveys of American professionals and of its own customer base. That flat fees are already the most common offering in the American mid-market suggests the obstacle is rarely the instrument. What is missing is the decision about which matters move there and at what price.

The arithmetic assumes a measurement few firms make

A pricing decision needs a number: how many hours actually disappear from a given matter type. That is where the evidence runs out. In Thomson Reuters' 2026 report on AI in professional services, covering more than 1,500 respondents across legal, tax, audit and accounting, corporate risk and government, 18 percent say their organisation measures the return on AI tools, and 40 percent do not know whether it is measured. The responses were collected in October and November 2025, and participants were screened for familiarity with AI. The share that measures is therefore likely lower across the professions as a whole.

Nor is a firm's own time recording clean evidence. 34 percent of all 1,816 respondents to Future of Professionals 2026, counted across every profession in the survey, say they use AI tools their organisation has not sanctioned and cannot see. If that use saves time, it is saved outside the system, and historic hours then overstate how long the work takes today.

A firm that cannot see where the hours went cannot price the fact that they went.

The measurement needed is bounded and made on the firm's own closed matters: the same matter type before and after, the same staffing, and a comparison of time recorded per step. That question belongs with how an AI programme is followed up in general, which is the subject of Do you know whether your AI programme worked.

The evidence is enough to decide on, not enough to name a cause

There is one relationship that pulls the other way, and it concerns the evidence. The level of fixed pricing in Clio's material cannot be attributed to AI. The material carries no comparison with an earlier year, so it shows a level and not a movement. The surveys behind those figures were fielded in April and June 2024, before the period Thomson Reuters measures, so the 64 percent says something about market maturity and nothing about what generative AI has caused.

Thomson Reuters' 71 percent has a corresponding weakness. It is a buyer's expectation and not a signed agreement, and the survey does not measure how many of them press the demand at the next procurement. The 28 percent has the opposite weakness: the question is explicitly about changes made in response to AI, which excludes firms that changed their pricing for other reasons. The share of firms with a fee mix beyond the billable hour is therefore not limited to those 28 percent.

The evidence shows the instrument exists. It does not show who drove it.

Taken together the objections point the same way as the conclusion, but with a different cause. If the price shift was already under way and AI accelerates it, the decision is not a reaction to a technology shock but a question of who sets the terms first. The firm that waits for the client's demand negotiates a discount. The firm that does the arithmetic first negotiates a model.

Three matter types, and the question that separates them

The criterion is bounded and gives different firms different answers. Take the three matter types that accounted for the most billed hours last year and put two questions to each.

The first: is the client buying an outcome? A dispute of uncertain scope is bought by the hour in practice, so no. An incorporation, a standardised review or a recurring contract sweep is bought as an outcome, and the client already has a view on what it should cost.

The second: could the matter have been delivered with twenty percent fewer hours without the client noticing any difference in quality? If the answer is yes, that difference is already yours, whatever you do with it.

Matter types where the answer is yes to both are the ones where the pricing decision has to be made before the tooling decision. There a saved hour becomes margin if the price is set in advance and a discount if it is not. Matter types where the answer is no to the first will bear hourly billing for a good while yet, and there AI is a question of capacity rather than of price.

The point is that the criterion divides the matter mix rather than passing judgment on the firm. The decision concerns individual matter types, which is why it can be made without anyone having to take a position on the future of the billable hour in general.