Every professional-services firm has now run the pilots. The budget line for the next round is already drafted, the tools are bought, and the partner group can name the tasks that got faster. What almost none of these firms has done is convene the pricing committee, and that is the decision that settles whether any of the spending reaches the accounts.
We read the fiscal 2023 and fiscal 2025 annual reports of 9 listed professional-services firms and divided reported revenue by the staff count each firm discloses. The 9 are Accenture, Cognizant, CRA International, EPAM, FTI Consulting, Genpact, Huron, ICF International and WNS. The thesis the numbers force is a sequence: the unit of adoption here is the engagement, not the firm, so the pricing and staffing unit has to change before the tools can pay. We call the limit that holds until it does the engagement ceiling, and 6 of the 9 firms, all but CRA, Cognizant and FTI, grew revenue per head more slowly than the 5.66 percent rise in US consumer prices over the same 2 years. Each of the 3 is measured from its fiscal 2023 report: CRA, Cognizant and FTI.
The money went in and output per person stayed where it was
Accenture's generative-AI new bookings reached $5.9 billion in fiscal 2025, against $3.0 billion the year before. Its revenue per person moved from $87,465 in fiscal 2023 to $89,439 in fiscal 2025, a rise of 2.3 percent. The AI line nearly doubled inside a year of new bookings that fell 1 percent, to $80.62 billion.
Genpact now books nearly half its revenue under a line item called Data-Tech-AI, $2,442.4 million of $5,079.9 million in 2025, or 48.1 percent, up from 46.7 percent in 2023. Its revenue per employee over the same 2 years moved by 0.0 percent, from $34,678 to $34,675. The label now covers almost half the business, and the firm-level number is unchanged to the third decimal place.
The private side of the industry reads the same way. PwC reports nearly US$1.5 billion of ongoing investment across the network in expanding and scaling its AI capabilities, and still gives its head count as 364,000, the same rounded figure it published 2 years earlier against network revenue up 7.2 percent. Deloitte has allocated over US$3 billion to generative AI through FY2030, and KPMG's global headcount grew 1.8 percent to 276,030, because headcount is what these networks sell.
Revenue per disclosed head is the number a partner group actually reports
From each of 18 filings we took 2 numbers, reported revenue and the staff count the firm discloses itself, and compared every firm only against its own 2 filings, so a difference in what counts as a head changes the level and not the 2-year change. Revenue per head is not productivity. It moves with pricing, staff mix, currency and purchases, and every firm in the set acquired inside the window, Accenture alone investing $1.5 billion across 23 strategic acquisitions in fiscal 2025.
It is still the number a partner group reports, and it barely moved. The median change across the 9 firms is 2.2 percent over 2 years, and 6 of the 9, all but CRA, Cognizant and FTI, grew revenue per head more slowly than US consumer prices rose over the same period. The 3 are measured from their own fiscal 2023 reports: CRA, Cognizant and FTI. 2 firms fell by a visible margin, Huron at 11.2 percent and EPAM at 1.6 percent, and the median firm is ICF at 2.2 percent.
The 2 falls are not AI failing. Huron grew revenue 21.5 percent while its roster of revenue-generating professionals grew 36.7 percent, from 5,519 to 7,546, and the largest part of that addition sits in a Managed Services capability the same filing states is mostly not billed on an hourly basis. That is a change in the unit, and it is the kind of change this set otherwise lacks.
Across 9 listed professional-services firms, revenue per disclosed head moved a median of 2.2 percent from fiscal 2023 to fiscal 2025, and 6 of the 9 grew it more slowly than the 5.7 percent rise in US consumer prices.
Source: Lunon analysis of fiscal 2023 and fiscal 2025 annual reports filed with the US Securities and Exchange Commission; US Bureau of Labor Statistics, CPI-U annual averages, 2023 and 2025.
The firms name the price, not the technology, as the binding constraint
Read the AI passages in these 9 annual reports and the constraint they name is commercial. 4 of them put client acceptance of new pricing or commercial models into their own risk factors: Accenture, FTI Consulting, Cognizant and Genpact. 3 more name pricing as the pressure point without putting it on the client, EPAM, Huron and WNS, and only CRA International and ICF name neither.
FTI writes the mechanism out in full. Non-expert tasks its professionals perform have been and will continue to be replaced by automation, which it expects to reduce demand for those services or the utilization rate of its professionals unless replacement demand arrives or clients accept new pricing. Make the task faster, lose the hour, and the repair is either new demand or a new price. That is the engagement ceiling written into a risk factor by the firm standing under it.
Where the pricing unit moved at all, it mostly moved toward the person
5 of the 9 firms publish a split of revenue by contract type, Cognizant, CRA, EPAM, ICF and WNS, and in none of the 5 did a move in the hourly share reach the firm-level number.
Time and materials gained ground at CRA. Revenue from time-and-materials engagements rose to 83 percent of net revenues in fiscal 2025 from 82 percent in fiscal 2024, and fixed price fell to 17 percent.
EPAM cut its hourly share and gained nothing. Time-and-materials revenue fell from 87.0 percent of revenue in 2023 to 80.0 percent in 2025, a cut of 7 points, and output per head still fell 1.6 percent.
WNS moved toward pricing per person, not away. Full-time-equivalent revenue rose from 67.7 percent of revenue in fiscal 2023 to 73.5 percent in fiscal 2025, so the share not priced per person fell 5.7 points.
WNS is the firm that writes the thesis into its own annual report. It intends to price on the value we deliver to our clients rather than the headcount deployed, expects that to grow revenue without increasing headcount, and publishes 28.8 percent of fiscal 2025 revenue as generated through non-linear pricing models that de-link revenue from effort. Its revenue per head is flat on either basis its filings publish, 0.5 percent below fiscal 2023 on revenue as reported and 0.9 percent above it on the revenue-less-repair-payments measure both filings reconcile, which strips the third-party repair-centre payments it books as principal.
At the government-facing firm the hourly line is the one thing that held still. ICF's time-and-materials revenue was 41.4 percent of total in 2023 and 42.8 percent in 2025, while cost-based revenue fell from 14 percent of revenue to 7 percent and fixed price rose from 45 percent to 50 percent. The contract mix moved a long way. What a unit of work is priced on did not.
The staffing unit did not move either
In the quarter it booked $1.8 billion of generative-AI work, Accenture initiated a 6-month business optimization program and recorded a charge of $615 million, expected to reach approximately $865 million across the 6 months, and said in the same release that it expects to increase its overall number of employees in fiscal 2026. Exits and hires are both repricings of the same unit, and the unit is a person.
Huron scaled the old way, by adding the unit. It grew revenue 21.5 percent and the hourly-billed part of its roster 18.8 percent, from 4,469 to 5,307 professionals, while utilization in the 2 capabilities it bills by the hour stayed inside a band of 4.6 points across the whole window. Adding people is what the unit is for, and it is what a firm does when nothing about the unit has changed.
The industry's payroll says the same thing from the outside. Employment in US management, scientific and technical consulting services stands at a record 1.90 million on the Bureau of Labor Statistics's preliminary reading for August 2026, and paid hours per employee rose across the exhibit window, from 37.0 in December 2023 to 37.6 in December 2025. More people, paid for more hours each.
The largest gain in the set came from working the old unit harder
CRA International is the counterexample, and it measures the ceiling rather than clearing it. Revenue per employee consultant rose 26.1 percent, from $621,490 in fiscal 2023 to $783,715 in fiscal 2025, by some distance the largest gain in the set. Part of it came from utilization, which CRA pushed from 70 percent in fiscal 2023 to 77 percent in fiscal 2025 with billable hours up 6.0 percent.
The rest sits in rate, staff mix and a fiscal 2025 that ran 53 weeks against 52. None of it is a changed unit, and utilization stops at 100. The firm that moved output per head furthest is also the firm that barely discusses the technology: CRA's annual report mentions AI 4 times and utilization or the hour 35 times, the lowest AI count in the set.
That pattern holds across the 9. Only 3 of the reports mention utilization or the hour more often than AI, and they are CRA, Huron and FTI: the largest gain in the set, the largest fall and the third-largest gain. Where the report still counts hours, the economics still turn on hours.
The next reporting season will not show who bought the best tools
The regulators have already settled the part of this that firms hoped was open. The Financial Reporting Council, in what its own announcement calls the first guidance from any audit regulator globally on generative and agentic AI in audit engagements, restated rather than rewrote the rule: accountability remains unchanged, the human auditor is always accountable, and it is firms and Responsible Individuals who answer for audit quality. PCAOB staff, after limited outreach to several audit firms and companies, reported generative AI focused primarily on administrative and research activities, and those firms saying the standards are not impeding them.
So the constraint is not the tools, and it is not the rulebook. It is that the engagement is still the unit at which this industry prices work and staffs it, and the engagement ceiling holds whatever a firm buys beneath it. A firm that wants AI in its economics rather than in its risk factors has to sell something other than a quantity of person-time, and that is a pricing-committee decision taken before the next tool is chosen.
The next reporting season will not show which firms bought the best tools, because every firm bought them. It will show which firms changed what they sell a unit of, and that number is visible 2 lines into any annual report: revenue, and the number of people it took to earn it.
Start with the decision in front of you.
Tell us what your team needs to understand, evaluate, or deliver.