The Expertise Premium Is Moving: How AI Changes Professional Services

AI is making professional-services production cheaper before judgment, mobilization and accountability become cheap. The expertise premium is not disappearing. It is moving. That shift now sits inside the broader AI-Native Professional Services category thesis: professional services should be redesigned around abundant machine-assisted cognition rather than merely adding AI to the old delivery model.

Professional services were built around several forms of scarcity at once. Credible research took time. Analysis required trained people. Synthesis depended on experience. Deliverables absorbed hours of specialist effort. Clients paid for expertise, but they also paid for the expensive human machinery required to turn expertise into something usable.

AI changes that bundle unevenly. Research, synthesis, analysis, modelling, drafting and presentation work can now be produced with dramatically less human effort in many contexts. That does not make expertise worthless. It makes the old value logic easier to inspect. The artifact is becoming cheaper to produce before the decision becomes cheaper to make.

Some of the expertise premium was always a production-scarcity premium

For most of the history of professional services, producing a credible answer required a great deal of human cognition. Someone had to gather information, clean it, compare sources, build models, identify patterns, formulate hypotheses, prepare recommendations, create presentations, rewrite documents and make the result coherent enough for a client to act on.

This was not waste. It was the production technology available at the time. The consulting pyramid, the legal associate model, the audit team and the research analyst all developed around that reality. Senior judgment was scarce, but so was the capacity required to create the evidence and artifacts on which senior judgment depended.

That distinction matters because it prevents an easy moral story. Professional-services firms did not invent labour-intensive delivery merely to protect fees. In many cases, large teams were the only practical way to search, calculate, compare, document and quality-check enough material to support a consequential recommendation. The old model reflected an old production constraint.

AI attacks that production constraint first. The 2026 Thomson Reuters AI in Professional Services report says organization-wide generative-AI use has risen sharply while only a minority of organizations report systematically tracking return on investment. Adoption is moving faster than the commercial and operating models around it.

The same tension appears in my own work. In AI Is Making Me Much Faster at Consulting Work, I described how AI can compress the production layer while leaving the harder judgment intact. Once that happens, hours become a weaker explanation for price. They still matter as a cost and capacity measure, but they become less convincing as a proxy for value.

The professional-services value stack is being repriced unevenly

The easiest mistake is to conclude that AI makes professional expertise cheap. It does not. It makes different parts of expertise differently scarce. A competent first version of an analysis may become dramatically easier to produce while the decision about what to analyze, what to ignore and what the organization can actually do remains difficult.

Production is the layer most obviously exposed. AI can already accelerate search, synthesis, first-pass analysis, documentation and drafting. Human expertise still matters for source quality, assumptions, edge cases and quality control, but the scarcity attached to producing a plausible artifact is weakening.

Interpretation is different. It asks which problem is actually worth solving, which signal matters, which benchmark is misleading and which recommendation survives contact with the client’s history. AI can contribute, but context is not merely more information. It includes knowing which facts are politically sensitive, which compromises shaped the current system and which apparently rational option will fail because the organization is not ready to act on it.

Mobilization is more different again. A recommendation that cannot be translated into action is an interesting document. Professional-services firms often create value by coordinating people who do not naturally cooperate, giving a difficult decision external legitimacy, creating a conversation an internal team could not convene or maintaining momentum across organizational boundaries.

Then comes accountability. Eventually somebody has to decide whether the diagnosis is good enough to bet on. Somebody has to challenge the evidence, validate the recommendation, accept that uncertainty remains and stand behind the decision when consequences become real. AI can inform that process. It does not make the consequence disappear.

AI may commoditize expertise before it commoditizes responsibility.

This is closely related to the Human Premium Enterprise. When ordinary cognitive output becomes cheaper, organizations have to become more precise about the human capabilities that remain consequential. Professional services are simply one of the places where that repricing becomes visible first.

The strongest case for professional-services firms is not production

There is a lazy version of the AI-and-consulting argument: clients can use ChatGPT to make a deck, therefore consultants are finished. That confuses an artifact with a professional service. A presentation is visible; the relationships, pattern recognition, contextual access and organizational work behind a consequential intervention often are not.

A strong professional-services firm can possess genuinely scarce assets: cross-company pattern recognition, deep domain expertise, access to specialists, proprietary or privileged information, trusted relationships, external independence, stakeholder credibility, transformation capacity and experience with situations a client may encounter only once. It may also have the institutional confidence to say that the client’s preferred answer is wrong.

Those capabilities are not automatically reproduced by a general-purpose model. A recent Financial Times debate around the future of consulting captures both sides of the pressure: AI is enabling some work to move in-house and creating cost pressure, while practitioners argue that complex advisory work still depends heavily on private information, contextual judgment and organizational execution. Both can be true. The threat to established consulting economics is real, without implying that consulting itself is dead.

In fact, AI may make the distinction between weak and strong professional services sharper. If a provider’s differentiation was primarily the ability to mobilize enough people to research, synthesize and package information, that differentiation weakens. If its advantage is unusual context, specialist depth, trust, independent challenge or the ability to get an organization to move, the value may become easier to explain.

The apprenticeship paradox: cheap work can still be valuable learning

AI can remove a great deal of junior production work. That sounds efficient because much of that work is repetitive and expensive for the client. But apparently low-value production was also how people became good.

Research. First drafts. Analysis. Client notes. Building a model badly. Presenting an argument that does not survive the first senior question. Watching an experienced colleague notice something you missed. The client may not have been paying for the learning intentionally, but the system was producing two outputs at once: a deliverable and a more experienced professional.

My own move from product leadership into consulting taught me this in a way I did not expect. I had already led three teams and organized asynchronous work across continents. I arrived with a strong instinct to take initiative and lead. On my first project, a peer pushed back hard against that instinct. She was much more versed than I was in consulting methodologies and in recognizing patterns across the kind of work we were doing.

The friction was uncomfortable and useful. Over time, she learned to respect the industry-side experience I brought. I learned to respect the expertise she had that I did not. The relationship shifted from competition over who should lead to a sparring relationship in which the project benefited from both perspectives.

You already have what you have. Growth comes from what the other person can add.

The lasting lesson was humility. AI can expose us to more information, patterns and critique, but it cannot remove the need for the human humility to recognize that somebody else may understand something we do not. That is not a sentimental defense of human interaction. It is part of how capable professionals discover the boundaries of their own judgment.

Preserving manual junior work purely for training would therefore be the wrong answer. We should not make clients pay for inefficient production so firms can maintain an old apprenticeship system. The harder task is to redesign apprenticeship so exposure, friction, feedback, responsibility and correction survive after repetitive production has been automated.

McKinsey’s 2026 work on building expertise in the age of AI makes this problem explicit. It argues that firms can no longer assume informal apprenticeship will happen automatically when AI increasingly handles research, documentation, data cleanup, basic coding and preliminary analysis. It recommends deliberate learning in the flow of work, stronger coaching, codification of expert decision rules and “attempt-then-check” patterns in which people form a view before using AI to test it.

Professional-services firms do not just sell expertise. They reproduce it.

This is the part of the AI debate that deserves more attention. Professional-services firms have historically manufactured new expertise while delivering work. A junior professional encounters a situation. A more experienced professional challenges the interpretation. The team sees how a decision performs in the real world. Patterns accumulate. Exceptions become memorable. The organization develops language, frameworks and judgment that can later be applied elsewhere.

The Expertise Reproduction Loop

1. Experience: professionals encounter real situations with real constraints and consequences.
2. Judgment: humans interpret what mattered, what surprised them and why one answer worked better than another.
3. Codification: useful patterns, decision rules, exceptions and evidence are made explicit without pretending every situation is identical.
4. AI + apprentice: precedents and reasoning become easier to retrieve while less-experienced professionals can attempt work sooner and compare their thinking against accumulated knowledge.
5. Challenge: people defend interpretations, disagree, receive correction and expose apparently plausible reasoning that does not survive expert scrutiny.
6. New experience: the decision creates fresh evidence, renewing the knowledge base rather than merely replaying the past.

If AI removes a large portion of the production layer, this loop has to become deliberate. Experience still has to create judgment. Judgment has to be codified selectively. Apprentices still need to attempt, compare, defend and be corrected. AI can retrieve precedents, surface patterns and preserve institutional knowledge, but the organization still has to generate new experience and decide what deserves to become reusable knowledge.

The future expertise pipeline therefore has two customers: the next generation of professionals and the AI systems they will work with. An AI system that only recycles yesterday’s codified knowledge eventually becomes less useful in a changing environment. Organizations need mechanisms for continuously converting new human judgment into validated institutional knowledge. If you want to extrapolate this into absurdity, think about how useful the AI would be in the dystopian future of the movie Idiocracy.

That does not mean dumping every senior opinion into a knowledge base. Good knowledge management is selective. Expert rules need provenance. Exceptions matter. Contradictions need to be visible. Content needs owners. AI outputs need validation. Tacit knowledge often has to be surfaced through examples, comparison and explanation rather than polished documents alone.

This is one reason redesigning work with AI should begin with the work itself rather than with a list of use cases. The question is not merely where AI can save hours. It is what system will continue producing better judgment after those hours disappear.

A firm that automates production but stops reproducing expertise is liquidating its intellectual capital.

Pricing should follow the movement of scarcity

Only after making that distinction does pricing become interesting. The temptation is to replace hourly billing with a supposedly more modern model and declare the problem solved. I do not think there is one universal answer. Different commercial models allocate uncertainty, access, effort, outcome risk and accountability differently.

Hourly billing can remain rational when scope is genuinely uncertain or provider effort correlates strongly with risk. Fixed fees can work when scope and deliverables are clear. Retainers make sense where access, continuity and accumulated context create value. Subscriptions can fit repeatable services or continuously updated knowledge products. Outcome-oriented and risk-sharing models become attractive when outcomes are measurable and the provider has enough control over the variables that create them.

Pure performance pricing is not automatically superior. Many professional outcomes are jointly produced by client and provider. Attribution is difficult. A consultant can give excellent advice that a client chooses not to implement. A law firm cannot control every judicial outcome. A transformation partner cannot own executive decisions it was never authorized to make.

The principle I would use is simpler: the commercial model should increasingly reflect where scarce value and risk actually sit rather than mechanically mirroring production effort. That may produce a different answer by service line, client and situation. It should.

Legal services provide an early visible example. Financial institutions have reportedly begun pressing major law firms to quantify AI-driven cost savings and reconsider fee structures. That does not prove every professional-services category will follow the same path, but it shows what happens when sophisticated buyers see a gap between hours consumed and value delivered. The Financial Times reported that pressure directly.

The opportunity for firms is not simply to accept lower prices. If AI removes expensive production effort, a provider may be able to charge less, increase margin and improve the client experience at the same time. A smaller invoice and a larger margin are not contradictory if both sides share the productivity gain.

That becomes a trust question as much as a pricing question. A firm that hides AI efficiency inside an unchanged effort narrative may preserve economics temporarily while making its value proposition harder to defend. An AI-native firm can do the opposite: expose where the machine lowered production cost and be much clearer about what the client is still paying a premium for.

The expertise premium is moving

The professional-services firm of the future cannot simply be a cheaper version of today’s firm. It needs to automate cognition that has become abundant, preserve human capabilities that remain scarce, redesign how those capabilities are developed, and continuously convert new human judgment into knowledge that both people and AI systems can use.

That changes the operating model. Production becomes more automated. Interpretation becomes more explicit. Mobilization becomes easier to distinguish from artifact creation. Accountability becomes a visible part of the value proposition. Apprenticeship becomes designed rather than incidental. Knowledge management becomes part of service delivery rather than an archive behind it. Commercial models become easier to interrogate because hours no longer provide a comfortable proxy for value.

It also changes what firms should protect. The answer is not the manual work that happened to train yesterday’s professionals. It is the mechanisms that created judgment: exposure to real problems, challenge from people who see something different, permission to be wrong before the stakes become too high, and enough humility to learn from expertise you do not possess.

None of this means professional expertise becomes unimportant. It means we have to become more precise about which expertise we are paying for, how that expertise is created and how it is renewed. AI-native professional services are therefore not just a different delivery model. They are a different system for producing, applying and renewing expertise.