A professional firm can make routine work cheaper without losing the revenue needed to train its people. If a client pays an agreed fee for a completed service and the firm delivers it more efficiently, the saving can leave more resources available for supervision, development or specialist capacity. Any argument about AI destroying the financing of expertise has to begin by acknowledging that straightforward possibility.
Something else can nevertheless change. The routine work that becomes automated may have supplied part of the experience through which junior professionals learned. The firm could therefore become more profitable while needing to redesign its apprenticeship. Losing a learning opportunity and losing the money to provide one are different problems, even when they arise from the same technological change.
The legal market offers a useful setting in which to examine the distinction. Thomson Reuters' 2026 research reported that 71 per cent of in-house legal professionals expected outside firms to change their commercial models as AI usage increased, while 28 per cent of law firms reported pricing changes. Those survey responses establish expectations and reported action, not a measured loss of the financial contribution supporting training.[1]
Consider a hypothetical lawyer reviewing a series of contracts under supervision. The exercise may both produce something useful for the client and teach the lawyer how particular risks appear in practice. The value of the learning depends on the work and supervision, not on the number of hours alone. Still, production and development can occur together, allowing a firm to recover some of the cost of employing inexperienced people while their competence grows.
Automation could separate those functions. If software completes much of the initial comparison, a junior may spend less time encountering differences directly and more time assessing the result. That change might improve learning if it exposes the person to useful examples and feedback. It might weaken learning if the person is expected to accept conclusions without understanding them. Neither effect can be inferred simply from the disappearance of manual review hours.
The same Thomson Reuters report found expectations of fewer junior roles and continued demand for structured support from experienced colleagues. It reported that 71 per cent believed early-career professionals would need that support to develop judgement. These are respondents' views about a changing profession, not evidence that every firm will face a shortage of expertise.[1]
The financial mechanism needs its own comparison. Suppose, illustratively, a category of work earns £100,000 and incurs £60,000 of avoidable delivery costs over the relevant period. It contributes £40,000 towards shared costs, development and profit. If AI reduces those delivery costs to £30,000 while the fee stays unchanged, contribution rises to £70,000. The firm has not lost a subsidy by automating the work.
A different outcome follows if the fee falls to £50,000 while the revised avoidable cost is £30,000. Contribution then falls to £20,000. Whether that creates a financing shortfall depends on the capabilities the firm must maintain, the shared costs it can adjust and the contribution obtained from other work. The fee reduction alone is not the lost contribution, and costs already deducted in calculating contribution must not be deducted again.
These examples are not estimates of professional-service margins. They show why task automation and financial loss are separate steps. A shortfall could arise if clients bring work in-house, competitors reduce fees substantially or the firm loses a stream of work while retaining costs that cannot adjust quickly. It could be avoided if the firm retains fees, expands demand, redeploys staff or reduces the cost of developing expertise itself.
It is also important not to call every contribution to shared overhead a cross-subsidy. Firms commonly recover common costs across several services. A genuine claim about one activity supporting another needs evidence about the relevant costs and the counterfactual: what capability would become difficult to sustain without that activity's contribution? An arbitrary allocation of office or management expense would not establish the answer.
The stronger concern is therefore conditional. Some firms may have relied on a profitable stream of routine work both to employ junior cohorts and to support expertise needed elsewhere. If the financial contribution declines and the capabilities remain necessary, they will need another way to pay for them. The technology does not guarantee that combination, and evidence from law cannot establish that it holds throughout accounting, engineering and consulting.
Clients are not obliged to preserve the supplier's inherited organisation. Thomson Reuters' 2026 UK legal-market report describes buyers seeking commercial understanding and improvements in delivery alongside technical expertise. That supports the need to demonstrate value, but not the proposition that buyers have agreed to finance a particular staffing pyramid.[2]
A firm might respond with fixed fees, subscriptions, specialist services or a different mix of work. None is a complete answer on its own. An agreed fee can allow the firm to retain efficiency gains, while competition can subsequently affect the level of that fee. Outcome-based charging requires clarity about the outcome and what the professional controls. The billing unit does not replace an understanding of the costs and obligations beneath it.
Training can be redesigned independently. Supervised cases, simulations, deliberate exposure to exceptions and structured feedback could replace some experience previously acquired through repetition. AI might help provide examples or identify misunderstandings. These are possible approaches, not evidence that professional judgement will develop automatically or that the replacement must be more expensive than the old process.
The useful objective is competence, not preservation of billable busywork. A firm may discover that less repetitive production leaves juniors with more time for valuable instruction. Another may need to create learning opportunities that no longer arise during delivery. The financial question is whether the chosen development process is supported sustainably; the educational question is whether it produces the capabilities the firm needs.
Time horizons can complicate both. Reducing a junior intake can lower expenditure now while affecting the future supply of experienced people. Investment in training can benefit another employer if people move. These familiar organisational choices may become more visible when routine production is less labour-intensive, but they are not evidence that partnership structures or automation necessarily make long-term investment irrational.
Nor can a firm assume that it will recover any shortfall by increasing specialist fees. Customers' alternatives, demand and the value of the service constrain what can be charged. Maintaining a capability may require lower distributions to owners, shared development resources or a different operating model. In some cases the capability may become cheaper to maintain; in others the firm may decide not to provide it.
The most informative evidence would connect matter-level economics with development outcomes over time. How much contribution remains after automation? Which learning experiences disappear, which improve, and how does independent competence develop? Separating these questions would show whether a firm faces a financial problem, an apprenticeship problem, both or neither.
AI could consequently change how expertise is produced and financed without making expertise less sustainable. The opportunity is to use cheaper production to support better professional development. The risk is to subtract work without noticing which useful functions accompanied it, or to assume that a fall in hours tells the whole financial story. A firm needs to understand what it is replacing, what resources remain and which capabilities it intends to maintain.
Editorial revision
Originally published as 'AI Could Destroy the Cross-Subsidies Hidden Inside Professional Services'. The title and argument have been revised to distinguish changes in learning opportunities from losses of training finance, and to make clear that shared overhead alone does not establish a cross-subsidy.