Will AI Squeeze the Middle of Professional Services?

Why AI May Squeeze the Middle of Professional Services

Artificial intelligence may put the most pressure on professional-services firms that sit between two clear sources of advantage. They lack the proprietary data, technology capacity, distribution, and service breadth of a scaled platform. They also lack the narrow expertise, partner attention, or relationship density of a focused specialist.

That is a plausible market-structure hypothesis. It is not yet a demonstrated market outcome.

Current adoption data show that large firms use AI more often than small firms. Current field research also shows that shared generative tools can improve some knowledge tasks and can give less-experienced workers a larger boost. Vendor cases show small and mid-sized firms using the same class of workflow tools as national firms.

The evidence points in both directions. Scale matters. Access is also spreading.

The real question is whether a mid-sized firm uses AI to deepen a distinctive service or merely to produce the same work faster while every competitor does the same.

The middle is a strategic position, not one headcount band

Statistical agencies need fixed categories. Strategy does not fit as neatly.

The OECD and Eurostat commonly use 50 to 249 employees for a medium-sized enterprise and 250 or more for a large enterprise. US size standards vary by industry and may use employees or annual receipts.

Those definitions are useful when comparing adoption rates. They do not tell a client whether a firm feels large, specialized, or interchangeable.

A 40-person valuation practice may dominate a narrow property type in one region. A 500-person multidisciplinary firm may be broad but undifferentiated. A local law firm may have stronger relationships than a global firm in one court or regulated market. A small insurance intermediary may have access to a network that gives it more distribution than its payroll suggests.

For this analysis, the strategic middle is a firm with enough size to carry coordination cost but not enough distinctiveness to make that complexity valuable.

The last row is the vulnerable position. It is also changeable.

Why AI can strengthen the scaled platform

Large firms often have more of the inputs required to move beyond casual AI use.

They may have licensed data, internal knowledge bases, security teams, procurement processes, training budgets, model-evaluation capability, and enough repeated work to justify integration. They can spread the fixed cost of a governed system across more employees, clients, offices, and transactions.

The adoption gap is visible in official data. In January 2026, the OECD reported that 52 percent of large firms used AI in 2025, compared with 17.4 percent of small firms, across countries with available data. Professional and scientific services had one of the highest industry adoption rates at 36.8 percent.

The OECD's broader AI adoption by SMEs report identifies connectivity, data, algorithms, computing resources, skills, finance, and digital maturity as important enablers. These are not distributed evenly.

Scale can also compound data. A firm that executes more transactions, handles more matters, or observes more client decisions can collect more examples. If those records are lawful, well-structured, relevant, and available to the right workflow, they can improve search, benchmarking, quality control, and decision support.

The conditional clause matters. More data is not automatically better data. Records gathered for one purpose may not be usable for another. Large firms can accumulate inconsistent systems, access restrictions, duplicate entities, and years of incompatible naming. A big archive without governance may create a bigger retrieval problem.

Even so, a scaled firm has more ways to turn an AI capability into a repeatable product. It can build a sector benchmark, monitoring service, client portal, standardized review layer, or cross-office delivery model. The same integration cost that is difficult for a small firm may become economical across hundreds of similar assignments.

Why the specialist can remain difficult to displace

A specialist wins for a different reason.

The client may value a person who understands one market, one asset class, one regulatory system, one type of transaction, or one operating problem unusually well. The specialist knows which apparent exception is normal, which source is trusted locally, how counterparties behave, and when a standard answer does not fit.

That knowledge can be partly documented and partly relational. It can include the ability to get a call returned, recognize an unrecorded constraint, or translate technical evidence into a decision the client can actually make.

AI may improve the specialist's economics. General-purpose systems can help a small team draft, research, extract, compare, and prepare materials without building a large support function. Embedded software can deliver capability as a subscription instead of a custom platform.

The OECD's SME report notes that the adoption gap for natural-language generation was smaller than the gap for some capital-intensive applications, which may reflect easier access. It also found that among AI-using firms, small enterprises devoted a slightly higher share of use to marketing or sales than large enterprises.

Commercial real estate provides a concrete example. Henry's current case-study library includes boutique, regional, and lean teams using a shared product for offering memorandums, broker opinions of value, research, and deal materials.

Those are vendor-selected customer stories. They cannot establish average return on investment, error rates, or market share. They do establish that the tools are not reserved for the largest firms.

When the technology is available to everyone, the specialist can combine it with context that is not available to everyone.

Why the undifferentiated middle feels the pressure

The middle-market squeeze requires several mechanisms to operate at the same time.

First, routine production must become cheaper. Research summaries, first drafts, standard analyses, document extraction, and presentation work need to consume less labor.

Second, clients or competitors must capture part of that saving. If every firm uses AI but prices, volumes, and margins remain unchanged, the technology may improve profit without changing market structure. A squeeze appears when clients expect faster delivery, lower fees, more evidence, or broader service for the same price.

Third, the scaled firm must turn its advantage into a better client outcome. Buying software is not enough. The firm has to integrate data, workflow, review, and distribution.

Fourth, the specialist must remain distinct. A boutique that describes itself as high touch but produces generic work is not protected by being small.

Fifth, switching must be possible. Professional services often have trust, conflicts, licenses, local regulation, institutional knowledge, and transition costs that slow client movement.

The mechanism can be expressed simply.

flowchart TD
    A["AI lowers the cost of routine knowledge production"] --> B{"Who converts the saving into client value?"}
    B --> C["Scaled platform<br/>breadth, data, integration, distribution"]
    B --> D["Focused specialist<br/>context, trust, speed, partner attention"]
    B --> E["Undifferentiated middle<br/>similar output, higher coordination cost"]
    E --> F{"Can the firm specialize, productize, or build a data advantage?"}
    F -->|Yes| G["New defensible position"]
    F -->|No| H["Fee pressure, slower growth, consolidation, or exit risk"]

This is a causal model, not a measured forecast.

Knowledge-work research makes the outcome less obvious

The strongest evidence against an inevitable squeeze comes from the task level.

A 2026 peer-reviewed Organization Science field experiment studied 758 Boston Consulting Group consultants. On 18 realistic tasks selected inside the tested model's capability frontier, consultants with GPT-4 completed 12.2 percent more tasks and worked 25.1 percent faster on average while producing higher-quality work.

On a complex task outside that frontier, AI users were less likely to produce the correct answer.

That result can favor a large firm because the experiment took place inside one. It can also favor smaller firms because the underlying model capability is widely available. The decisive advantage may come from knowing which task belongs on which side of the frontier, building a reliable workflow around it, and retaining the expertise needed to recognize failure.

The NBER Generative AI at Work study offers another counterweight. In a staggered deployment across 5,179 customer-support agents, AI access increased issues resolved per hour by 14 percent on average. The improvement was 34 percent for novice and lower-skilled workers and minimal for the most experienced group.

Customer support is not commercial real estate, law, insurance, accounting, or consulting. The measured outcome should not be transferred directly.

The mechanism is still relevant. A system can spread patterns from stronger performers and help less-experienced workers move faster. If that mechanism holds in a bounded professional workflow, a mid-sized firm may need fewer years or layers to reach a useful baseline.

AI can therefore reinforce scale and compress experience advantages at the same time.

Adoption is high in professional services, but uneven in real estate

Eurostat's current report on AI use in enterprises provides a detailed 2025 view across EU enterprises with at least 10 people in the covered sectors.

Professional, scientific, and technical service activities had a 40.43 percent AI-use rate. Real estate activities were at 24.82 percent. Text analysis was the most common AI technology in professional services.

The same report shows substantial size differences. Large enterprises used AI more often than small and medium enterprises. Among firms that had considered AI but did not use it, leading barriers included lack of expertise, uncertainty about legal consequences, and privacy or data-protection concerns.

These numbers matter, but they do not prove the squeeze.

The industry categories are broad. The data do not show whether a firm improved margins, won clients, reduced prices, changed staffing, or lost share because of AI. They also do not separate casual employee use from a deeply integrated workflow in every case.

The US Census Statistics of U.S. Businesses tables can measure firms, establishments, employment, payroll, and receipts by enterprise size and industry. The latest SUSB release does not measure an AI-caused shift in professional-services market structure.

The evidence base is improving. The decisive longitudinal result is still missing.

The middle can get stronger

A mid-sized firm has advantages that the squeeze narrative can miss.

It may have enough recurring work to justify a governed system but fewer legacy systems than a global platform. It may be able to make a cross-functional decision without a year of procurement. It may have more specialists than a boutique and more partner attention than a global firm.

It can also productize a service around a real client problem. Productization does not require turning a professional judgment into a black box. It can mean a consistent intake, evidence model, review process, report structure, monitoring service, or client portal.

The firm can build a proprietary context layer from its own completed work, client-authorized records, and corrections. It can develop a strong position in a region, asset class, transaction type, or regulated niche. It can use shared models and infrastructure while owning the workflow, evaluation set, client relationship, and professional accountability.

This is the strategic escape route from the middle. The firm stops selling undifferentiated labor and starts selling a distinct system of expertise.

Client procurement can change the result

Professional services are not purchased only on price and output quality. Clients may require geographic coverage, minimum insurance, data-security controls, diversity commitments, panel status, conflicts capacity, financial stability, or the ability to combine several service lines under one agreement.

Those requirements can favor scale. They can also create room for a mid-sized firm that is large enough to pass procurement but small enough to provide a consistent senior team.

AI changes this layer when clients begin asking how a firm uses their information, which models or vendors touch it, who reviews the output, whether use is disclosed, and how an error can be challenged. A governed mid-sized firm can outperform a larger competitor that has more tools but cannot give a clear answer.

The buying process therefore belongs in the evidence set. A middle-market squeeze should appear not only in production costs, but also in shortlists, request-for-proposal criteria, panel renewals, client concentration, cross-selling, and the reasons clients give for switching.

Talent development is part of the market structure

Many professional firms have historically trained junior people through research, document preparation, modeling, and repeated review. Those tasks can be tedious, but they also expose a person to the evidence and corrections behind expert judgment.

If AI removes the work without replacing the learning, a firm may improve current margins while weakening its future senior bench. Large firms can absorb that mistake for longer. Small specialists may teach through direct apprenticeship. A mid-sized firm needs to design the transition deliberately.

The better model uses AI to shorten mechanical effort while keeping the learner close to sources, assumptions, reviewer feedback, and client consequences. Measures should include time to independent performance, correction quality, retention, promotion readiness, and the concentration of judgment in a few senior people.

That talent system can become a middle-market advantage. A firm that develops capable professionals faster may gain capacity without flattening its service into generic output.

A leader should watch outcomes, not demos

The squeeze becomes credible when several observable signals move together.

No single signal settles the issue. A merger can reflect succession, capital, geography, or a service-line gap. A margin change can come from demand or interest rates. A firm that adopts AI may already have stronger management.

The analysis needs a baseline, a comparison group, and enough time to separate a technology story from a business-cycle story.

The strategic test for a mid-sized firm

The useful question is not whether AI will destroy the middle. It is whether the firm can answer why a client should choose it after routine knowledge production becomes cheaper.

If the answer is broad capability, the firm must show that it can integrate services and data into a better outcome. If the answer is specialization, the niche must be meaningful to the client and difficult to reproduce. If the answer is relationships, the relationship must improve access, judgment, speed, coordination, or trust. Familiarity alone is vulnerable when the service gap becomes large.

The firm should know which data it can lawfully use, which recurring task can be measured, where its experts correct generic output, and which part of the workflow a client would miss if the firm disappeared.

That exercise often reveals that the firm is not trapped in the middle. It is simply describing itself too broadly and operating too inconsistently.

The hypothesis remains open

Luke Tatman's E081 hypothesis is useful because it identifies an uncomfortable position. Large firms can compound scale. Small specialists can defend context and trust. The middle can carry the costs of both without securing the advantage of either.

The current evidence supports the mechanisms, not the verdict.

Large firms adopt AI more often. Smaller firms can access powerful general and industry tools. Knowledge-work gains vary by task and experience. Real estate and professional services are adopting at different rates. No current dataset cited here proves that AI has caused an enduring loss of market share, margin, or employment for the middle.

The right response is not to wait for the market to settle. It is to build a position that can be measured.

The workflow guide [[How AI Changes Commercial Real Estate Work]] shows where that measurement can begin. [[When Information Stops Being the Moat]] explains which value layers can remain defensible after access becomes cheaper.

AI assisted with research organization and drafting. Dalton Anderson remains responsible for the analysis, source boundaries, and publication decision.