Segments covered: advisory and management consulting firms, with notes on legal, accounting, and engineering practices where their constraints differ. Linked workflows: proposals, project knowledge retrieval, meeting follow-up. Linked roles: proposal managers, finance leaders, operations leaders.
This chapter considers firms that sell judgment, specialized work, and accountability. A document can be one step in that service, but the client is buying a result someone stands behind. Evaluate AI against the complete engagement: winning it on terms the firm can deliver, understanding the problem, doing the work, checking the result, and helping the client act.
The chapter separates three kinds of material. Sourced findings are cited inline with their population, date, and limits. Proposed practices and editorial reasoning are labeled as such. The firm in the sequencing section and all numbers in the billing example are invented for teaching.
How the business makes money
A professional-services firm converts people's time and expertise into fees. Its revenue depends on how the work is sold, and the main arrangements behave differently when work gets faster:
| Commercial arrangement | What the client pays for | What happens when effort falls |
|---|---|---|
| Time and materials (hourly) | Hours worked at agreed rates | Revenue on that engagement falls unless hours are redeployed or rates change |
| Fixed fee | A defined scope or deliverable | Revenue holds for the current contract; the gain appears only if released hours go to other paid work or avoid paid cost |
| Retainer or subscription | Access to a service over a period | Invoice unchanged; the firm can serve more within the same fee, which may or may not become revenue |
| Contingent or success fee | An outcome | Faster work matters only if it changes the outcome or the firm's capacity to take more mandates |
Four operating ratios, defined here as editorial working definitions rather than an industry standard, decide how those arrangements turn into profit. Utilization is the share of available hours spent on client work. Realization is the share of the standard value of that work actually billed; collection, how quickly and fully those bills are paid, is a separate cash measure. Leverage is the ratio of junior to senior staff on an engagement, which sets how much senior judgment each hour of junior work carries. Rate is the price per hour or its fixed-fee equivalent. An AI effect has to move one of these, or reduce a paid cost, before it changes profit.
The sector is large and heterogeneous. The US Bureau of Labor Statistics' Current Employment Statistics survey puts professional, scientific, and technical services (NAICS 54) at about 10.84 million employees in August 2026, seasonally adjusted and preliminary. The same survey reports, for August 2026, about 1.25 million in legal services (NAICS 5411), 1.12 million in accounting, tax preparation, bookkeeping, and payroll (5412), 1.77 million in architectural and engineering services (5413), 2.36 million in computer systems design (5415), and 1.90 million in management, scientific, and technical consulting (5416). BLS series CES6054000001, 5411, 5412, 5413, 5415, 5416 These are employment counts from a payroll survey; they say nothing about AI use, margins, or billing arrangements.
This guide's lens is the client-engagement firm, with consulting as the reference case because the linked workflows were written for it. Law, audit, and engineering share the engagement model but add professional and regulatory duties this guide does not analyze in depth. Computer systems design sits in the same NAICS sector, but its product-building work is closer to the B2B software guide.
Where AI could touch the value chain
| Process | Capability | Canonical chapter | Benefit mechanism | Benefit location |
|---|---|---|---|---|
| Qualify and pursue opportunities | Retrieve, generate, classify requirements | Proposals | Selling capacity; fewer unpriced commitments | Operating expense; revenue only with evidence |
| Find prior work, methods, and people | Retrieve, summarize | Project knowledge retrieval | Delivery capacity; reuse of permitted work | Cost of delivery, or revenue if redeployed |
| Capture decisions and actions from meetings | Transcribe, extract | Meeting follow-up | Coordinator capacity; fewer dropped commitments | Cost of delivery; quality benefit unpriced |
| Analyze client data and draft deliverables | Generate, retrieve, predict | Not yet covered in this guide | Delivery capacity | Depends on billing arrangement |
| Conflict checks and engagement intake | Rules and records | Ordinary software first | Risk control | Risk exposure |
| Time capture, billing, and collections | Rules, extraction | Ordinary software first | Realization; faster cash | Revenue and working capital |
| Staffing and resource allocation | Optimize, predict | Not yet covered in this guide | Utilization | Revenue |
Two rows deserve an explicit "not AI first" marking. Conflict checking is a records-and-rules problem, though resolving entities (affiliates, renamed parties, name variants) takes judgment, and a missed conflict is expensive; a well-maintained conflicts database and deterministic matching come before any generated suggestion. Time capture and billing gains may come from ordinary practice-management software and discipline around entering time.
Separate business development from paid delivery. A proposal and a client report may look alike as text, but errors have different consequences and different owners. The system should know whether material is approved for reuse, restricted to one client, or obsolete.
Constraints that shape every project
Client confidentiality. Prior-engagement material can be bound by client contracts, non-disclosure agreements, and conflict or ethical-wall arrangements. Reusable methods and client-confidential deliverables can sit in the same folders. The knowledge retrieval chapter sets out a proposed design: filter by entitlement before retrieval, deny unlabeled material by default, and avoid revealing that restricted work exists.
Professional duties. These differ by profession, and this guide does not state requirements for any of them beyond one scoped example. For US lawyers only, ABA Formal Opinion 512 (July 29, 2024) warns that self-learning tools may disclose information about one client's representation improperly "even if the tool is used exclusively by lawyers at the same firm," and that boilerplate engagement-letter consent "is not sufficient." ABA Formal Opinion 512 It is not a rule for consultancies, accountants, or engineers. Firms in regulated professions should take the question to their own professional-responsibility advisers.
Client contracts about AI. A client agreement may restrict the use of third-party processing, model providers, or AI tools on the client's data, or require notice. Treat this as a diligence question per client, answered from the contracts themselves, before any engagement material enters a system.
Accountability for signed work. A partner or licensed professional who signs a deliverable owns its content regardless of how it was produced. Review has to be designed around what the firm is promising, not around the fluency of the draft.
Liability and insurance. Whether professional indemnity or errors-and-omissions coverage has terms relevant to AI-assisted work is a question for the firm's broker and counsel. This guide makes no claim about policy terms.
The sector's economics lens
For your finance lead
The billing arrangement decides whether released time becomes profit. The following example is invented, and its arithmetic is shown in full.
Assume one engagement-equivalent of 200 delivery hours. The standard rate is $250 per hour. Staff are salaried, so the $110 per hour cost of their time is incurred whether or not the hours are billed: $22,000 per engagement-equivalent. Assistance releases 20 hours (10%) and adds $600 of tool cost allocated to the engagement. The firm's contribution is fee revenue plus any revenue from redeployed hours, minus staff cost and tool cost.
| Scenario | Engagement revenue | Revenue from redeployed hours | Staff cost | Tool cost | Contribution | Change from baseline |
|---|---|---|---|---|---|---|
| Baseline, time and materials | $50,000 | $0 | $22,000 | $0 | $28,000 | — |
| Time and materials, released hours idle | $45,000 | $0 | $22,000 | $600 | $22,400 | −$5,600 |
| Time and materials, released hours billed elsewhere | $45,000 | $5,000 | $22,000 | $600 | $27,400 | −$600 |
| Fixed fee of $50,000, released hours idle | $50,000 | $0 | $22,000 | $600 | $27,400 | −$600 |
| Fixed fee of $50,000, released hours billed elsewhere | $50,000 | $5,000 | $22,000 | $600 | $32,400 | +$4,400 |
| Fixed fee repriced to $45,000, hours billed elsewhere | $45,000 | $5,000 | $22,000 | $600 | $27,400 | −$600 |
Three observations follow from this arithmetic, not from any measurement. First, under hourly billing a faster engagement earns the firm less unless the released hours are sold elsewhere; the client, not the firm, captures the efficiency. Second, a fixed fee protects engagement revenue but produces no gain by itself when staff are salaried; the gain arrives only through redeployment to paid work or reduced paid cost such as contractors or overtime. Third, if clients learn that work is faster and reprice fixed fees, the gain can disappear. The example ignores any effect on quality, win rates, or client retention, each of which would need its own evidence.
Who keeps the released hours depends on the bill
- Better than the baseline
- Worse than the baseline
Time and materials · released hours idle
−$5,600
Time and materials · released hours billed elsewhere
−$600
Fixed fee · released hours idle
−$600
Fixed fee · released hours billed elsewhere
+$4,400
Fixed fee, repriced · released hours billed elsewhere
−$600
Show the data
| Billing model | Released hours | Contribution | Change from baseline |
|---|---|---|---|
| Time and materials | Released hours idle | $22,400 | −$5,600 |
| Time and materials | Released hours billed elsewhere | $27,400 | −$600 |
| Fixed fee | Released hours idle | $27,400 | −$600 |
| Fixed fee | Released hours billed elsewhere | $32,400 | $4,400 |
| Fixed fee, repriced | Released hours billed elsewhere | $27,400 | −$600 |
A firm with contractor-heavy delivery sees a different picture: released contractor hours reduce a paid cost directly. The finance leaders guide sets out how to test whether a realization mechanism is real before counting it.
Avoid two forms of double counting. The workflow chapters value released hours as capacity; the table above values the same hours as redeployed revenue. Use one or the other for the same hours, never both. And a senior reviewer whose time rises can absorb much of what junior staff saved; measure total effort by level, as the proposals chapter recommends, before calling anything a margin improvement. The economics chapter distinguishes benefit location from accounting classification; this example is a decision model, not revenue recognition or cost classification advice.
What is measured about adoption and outcomes
For your analyst
Adoption. The US Census Bureau's Business Trends and Outlook Survey (BTOS) asks employer businesses every two weeks: "In the last two weeks, did this business use Artificial Intelligence (AI) in any of its business functions?" For the collection period September 21 to October 4, 2026 (reference period September 7–20, 2026; published October 8, 2026), an estimated 42.7% of businesses in professional, scientific, and technical services (NAICS 54) answered yes, with a standard error of 1.01 percentage points, against 23.8% of all businesses nationally. In the same cycle, 47.2% of NAICS 54 businesses said they expected to use AI in some business function in the next six months. BTOS sector data, national data
Read those figures with their definitions. The estimates are shares of responding employer businesses, weighted to represent the business population, not shares of employees or of work. BTOS methodology, version 5 "Use in any business function" can mean a single employee using a writing tool. The Census Bureau changed the question in November 2025 from use "in producing goods or services" to use "in any business function," so earlier and later figures are not directly comparable. Census Bureau, May 2026 BTOS is published as an experimental data product. Adoption is not deployment maturity, and neither is an outcome.
Share of businesses using AI, by sector
- This guide’s sector
- Other sectors in the guide
- All businesses
Show the data
| Group | Estimate | Standard error | 90% interval |
|---|---|---|---|
| Information | 50.2% | 1.43 pts | 47.8–52.6% |
| Finance and insurance | 40.1% | 1.86 pts | 37.0–43.2% |
| Professional, scientific, and technical services | 42.7% | 1.01 pts | 41.0–44.4% |
| Health care and social assistance | 24.8% | 1.01 pts | 23.1–26.5% |
| Wholesale trade | 21.6% | 1.46 pts | 19.2–24.0% |
| All employer businesses | 23.8% | 0.41 pts | 23.1–24.5% |
Outcomes. The best-documented experiment in this sector's work randomized 758 consultants at one firm across two task types and three access conditions, using a model from June 2023. On a task the authors designed to fall outside the model's capability, correctness was 84.5% without AI versus 60% and 70.6% in the two AI conditions, about 19 percentage points lower combined; other tasks showed different effects, measured partly through human quality ratings. Dell'Acqua et al., Organization Science, 2026 We have not been able to verify the article's supplement or its preregistration, so take outcome definitions from the article itself. The study supports one design principle: evaluate by task, including tasks where the right answer is to recognize a limit. It does not supply an expected return for any workflow in this guide.
We found no attributable measurement of how AI use changes utilization, realization, leverage, or profitability across professional-services firms. The Neuronify AI hiring index measures job postings at New York AI companies; it is not evidence of adoption or outcomes in professional services.
Who is involved
Partners and practice leaders own the client promise, the price, and the decision to pursue work. Delivery leads own feasibility and quality. Proposal managers own the pursuit process and the evidence behind claims. Knowledge managers own the library, its labels, and retirement of stale material. The conflicts or risk function owns walls and client restrictions. Finance leaders own whether a benefit is real in cash. IT or engineering owns access enforcement and logging. This guide does not yet include role guides for partners, general counsel, or knowledge managers.
Leverage and learning are a decision, not a forecast. Junior staff learn by drafting, researching, and having their work corrected. If assistance takes over first drafts, the firm has to decide how juniors will still build judgment: by checking generated work against sources, by drafting some sections without assistance, and by explaining their corrections to reviewers. Whether assistance changes the partner-to-staff mix, hiring, or career paths is something to plan deliberately and measure, not something this guide predicts.
Expertise specific to this industry. A general AI builder lacks knowledge of how the firm's conflict and ethical-wall arrangements work, which client agreements restrict AI tools or require notice, and which engagement material is reusable method and which is client-confidential deliverable. In regulated professions it also lacks the professional-responsibility questions, such as those ABA Formal Opinion 512 raises for US lawyers. This guide places that expertise in-house, with the conflicts or risk function, knowledge managers, and the signing partner, with professional-responsibility advisers, counsel, and the firm's broker as advisers. A builder can implement entitlement filtering; the firm decides who may see what. See Staffing an AI pilot and Choosing and working with an AI partner.
Sequencing: a first move and what follows
The following is a fictional teaching example. Ossingdale Advisory is an invented operations consultancy of about 180 people across three practices. Its numbers and decisions exist only for this example.
First, measure and label. Ossingdale times its current proposal work by stage and by seniority for one quarter, and asks its knowledge manager to label the transport practice's engagement files by access class and reuse permission. No tool is selected yet.
First move: proposals in one practice. Ossingdale chooses the requirement-extraction and compliance-matrix step for procurement bids in the transport practice, following the proposals chapter and the proposal managers guide. It starts there because the output can be checked against the request document, the evidence library is small and owned, and scope and price stay with partners. Success is measured as total effort by level, missed requirements caught before submission, and unsupported claims removed, not pages produced.
Next, if that works: internal meeting recaps, then retrieval. Internal project meetings come second, because consent and client-policy questions are simpler than for client meetings; the meeting follow-up chapter covers the design. Knowledge retrieval comes third and only for the transport practice, whose files are now labeled, with the conflicts system integrated as the knowledge chapter proposes.
Counterexample: the disputes practice waits. Ossingdale's litigation-support practice asks to go first because its document volumes are larger. The firm declines for now. Its engagement files are unlabeled, several client agreements restrict third-party processing of case material, and a single wall breach would be serious. The simpler investments come first: labeling, a contract review for AI clauses, and a conflicts-system integration. Waiting is the right call because the constraint is permission, not capability.
Questions for operators, investors, and vendors
- How is each target service line billed today, and what happens to revenue if its hours fall by a tenth?
- Where will released hours go: named paid work, avoided contractor spend, or nowhere yet?
- What share of engagement files carries a reliable access label and reuse permission?
- Which client agreements restrict AI tools or third-party processing, and who has checked?
- Who signs each deliverable, and what exactly do they review?
- Does senior review effort rise when junior drafting speeds up? Has anyone measured it?
- How will juniors keep learning to draft and research?
- For vendors: is retrieval filtered by the requester's entitlements before ranking, and how quickly do wall changes propagate to indexes, caches, and stored conversations?
- For vendors: is client data used to train or improve shared models, and what contractual terms govern retention?
- For investors: does the firm's claimed AI advantage show up in utilization, realization, or rates, or only in anecdotes about speed?
When to wait
Wait when engagement files lack reliable access labels, client contracts have not been reviewed for AI restrictions, or the firm cannot say where released hours would go. Wait when the work is mostly bespoke judgment with little reusable structure, or when senior reviewers are already the bottleneck and would absorb any junior savings. The simpler investments are useful on their own: a labeled engagement register, an owned method library, disciplined scoping and pricing, and practice-management software that captures time accurately.
Evidence still needed and related reading
Obtain attributable operator accounts of stage-level proposal effort, review effort by seniority, labeling coverage, and how released hours were used, from firms on different billing arrangements. Verify the ABA opinion against its primary text. Review the consulting study's supplement and preregistration before expanding its quantitative treatment. No firm has been interviewed and no deployment observed for this chapter.
Continue with proposals, project knowledge retrieval, meeting follow-up, proposal managers, finance leaders, and the economics chapter.
Hiring in New York: open roles at NYC AI companies building for legal and compliance work, read daily from the companies' own job boards. These companies sell AI into the industry; they show which skills vendors are hiring for, not what firms in the industry are hiring.