Incumbents and AI-native firms are rewriting B2B services. Each week: one deep dive into what they’re actually building, plus a few headlines worth your time.
Hi, I’m Jimmy Bijlani, founder and CEO of AI Momentum Partners (AMP). Welcome back, and welcome if you’re new. This week, we check whether AI is finally breaking the oldest equation in services: more revenue, more people. It’s about a six-minute read, which is still shorter than your last utilization review.
Revenue per head rose at 20 of 26 big consulting, IT services and outsourcing firms, and at 17 of the 21 that grew revenue at all.
Nothing in the public numbers proves AI did it. Price hikes, hiring freezes and contractors move the same ratio.
The question to ask any “AI-enabled” services business: how much of the gain is actual delivery redesign?
Revenue is outrunning headcount. Don’t thank AI yet.
Two-year change in reported revenue and headcount at 26 firms, sorted by the gap between them.
“Consulting is dead.” “AI will replace consultants.” “Why hire a consultant when Claude can write your strategy?” If you’ve been on LinkedIn this year, you’ve read all three, usually in a post selling agents.
So we checked. We pulled two years of revenue and headcount for 26 of the biggest consulting, IT services and outsourcing firms: Accenture, TCS, Infosys, the Big 4, BCG, Cognizant, Genpact and friends. Filings, results releases and annual reviews, the boring stuff, so you don’t have to read it.
Spoiler: consulting isn’t dead. 18 of the 26 still grew headcount, which is a weird way to die. What’s changing is how much revenue each person brings in, and whether AI deserves the credit is a much messier question.
Some of you saw the chart on LinkedIn. Here’s what it says, and what it doesn’t.
The punchline first
Revenue per head rose at 20 of the 26 firms. In plain English, revenue outpaced headcount.
The obvious objection is that some of them only look efficient because they shrank. Fair. DXC, Kyndryl and TTEC all lost revenue and lost people even faster. So look only at the 21 firms that actually grew revenue. 17 of them grew it faster than their workforce, including four that did it with fewer people: Globant, Booz Allen, HCLTech and TCS.
The 26 firms, grouped by what actually happened to revenue and headcount.
The spread inside the group is wild. BCG grew revenue 17% on 5% more people, and Accenture grew 14% on 5%. TaskUs grew 28% and added 36% more people, and a growing slice of its work is humans training AI models: its AI services line went from 14% to 18% of revenue last year. The robots are hiring.
If you’ve been waiting for proof that services is finally breaking the link between revenue and bodies, this looks a lot like it. Which is exactly why you should slow down before forwarding it to your board.
Now the fine print
Start with the four that grew with fewer people.
The four firms that ended the window with more revenue and fewer employees, and the catch with each.
Booz Allen ranks second on the chart, and its most recent year was a contraction. Revenue fell about 6% as federal spending cuts landed, and headcount fell with it. The two-year math still works. Nobody should be calling it an AI win.
HCLTech “grew with fewer people” by ending the window with 300 fewer employees out of roughly 227,000. That’s a rounding error with a press release. Then, in the September quarter, it added 6,450 people, its biggest quarterly addition in four years. TCS is hiring again too, three quarters running.
Two endpoints give you a snapshot. Calling it a trend is how analysts get roasted in the comments.
Revenue per head lies like a pitch deck
It’s the most flattering number in services, and it goes up for a pile of reasons that have nothing to do with AI:
Pricing. Raise rates and it goes up while everyone works exactly the same way.
Mix. Sell more software, managed services or fixed-fee work and it improves.
Fewer juniors. Thin out the bottom of the pyramid and the average head brings in more.
Utilization. Shrink the bench and the math looks great.
Not backfilling attrition, the quietest cost program in services.
Contractors and subcontractors, who do real work and never show up in headcount.
Acquisitions and currency, which can push either number around on their own.
Currency cuts both ways, by the way. We used the dollar revenue the Indian firms report. In rupees, TCS grew about 11% over the window, not 3%. For the Indian IT names, that choice actually works against our headline.
The billing problem nobody puts on the AI slide
Under time and materials, AI that halves the hours halves the invoice. So a firm that bills by the hour has a funny incentive. It can use AI to cut its own costs, but every hour AI saves a client is an hour it no longer bills. Under fixed fees or outcome pricing, the same efficiency turns into margin.
That’s why rising revenue per head at an hourly firm deserves a second look before anyone credits AI. It’s also why you should ask why so many firms selling “AI transformation” still bill you by the hour.
So what (the part you actually opened this for)
If you run a services business: “revenue per head went up” is a vanity metric until you can split it into four buckets: price, mix, hiring freezes and actual delivery redesign. AI lives in the last one, and the other three run out. Price runs out when clients push back. Mix runs out when there’s nothing left to bundle. Not backfilling runs out when your best people leave. If your finance team can only “kind of” make that split, that’s your answer.
If you’re a PE operator: put that split in the next portco review, and ask for it before you pay a premium for an “AI-enabled” services business. Whether revenue per head is rising is the easy question. Why it’s rising tells you whether the number survives the next downturn, and whether the next buyer will pay for it. If the answer is “we raised rates and stopped backfilling,” congrats, you paid an AI multiple for a price hike and a hiring freeze.
If you sell for a living (hi, consultants): somebody is going to pocket the difference. If the same revenue really needs fewer people, either the firm keeps it as margin or the client takes it back in price. Clients buying hours will start asking which one it is, probably at your next renewal.
The one honest test
Does engagement #100 take way less effort than engagement #10, at the same price and quality? If yes, that’s the real version of “consulting is dead,” and no public filing can show it yet. If not, somebody just had a good run.
What we’d measure instead
Four numbers, tracked by delivery model: organic, constant-currency revenue per average FTE; realized rates; utilization; and contractor and subcontractor spend.
If revenue per FTE climbs while rates, utilization and contractor spend hold steady, something real is changing in how the work gets done. If it climbs because the bench shrank and the contractor line grew, you didn’t get more productive. You moved the cost to a different line of the P&L.
The quiet part
These are 26 giants with armies of finance analysts, and their public numbers still can’t separate AI from pricing, staffing mix and a hiring freeze. The PE-backed mid-market has thinner data, smaller teams and a lot more exposure when clients come asking for their share.
That’s where we think the real gap is. Plenty of firms are using AI. Very few can show you what it’s worth in their own numbers.
How we built it
Each firm’s latest comparable full year vs two years earlier, so windows follow each fiscal calendar. They’re labeled on the chart.
Revenue is nominal and in reported currency. Headcount is year-end people as reported, approximate at some firms.
No acquisition adjustments. Accenture, Cognizant, CGI, EPAM, Huron and Wipro all bought companies inside their windows.
Capgemini, Concentrix and Teleperformance are out because deal timing skews the comparison, and Endava because its FY26 results are delayed. Deloitte is FY23 to FY25 because its latest headcount came only as “nearly 500,000.”
If we got something wrong, hit reply. We’d rather fix it than defend it.
Headlines worth your time
Accenture put a meter on Claude. Since Oct 1, centrally funded Claude usage is capped at $29 per employee a month, after some staff had reportedly run up about $1,000, according to an internal memo reported by 404 Media.
Wipro got smaller while standing still. Its headcount fell about 2.8% in the September quarter while IT services revenue came in flat.
McKinsey ran the jobs math. It estimates automation could cut US labor demand by the equivalent of about 36 million jobs, while AI could create demand for more than 40 million over the next decade.
Hit reply and tell me how your firm actually measures this. Bonus points if the honest answer is “we don’t.” The best answers, anonymized, may show up in a future issue.
Warmly,
Jimmy Bijlani
P.S. If someone forwarded this to you, you can subscribe at theservicesrewrite.com.