The traditional P&L, line by line.
A boutique professional-services firm at the $3M to $10M revenue range looks almost identical from firm to firm. The shape has been stable for three decades.
Take a representative example. Ten-person firm, security consulting, $4M in annual revenue.
- Revenue: $4,000,000
- COGS (billable operator salaries, loaded): $2,100,000 (52.5 percent)
- Gross margin: $1,900,000 (47.5 percent)
- Sales and marketing: $280,000 (7 percent)
- G&A (non-billable, rent, tools, legal, insurance): $480,000 (12 percent)
- Partner compensation above salary: $640,000 (16 percent)
- Operating margin: $500,000 (12.5 percent)
The line that eats the margin is COGS. Billable headcount is roughly half the revenue on day one, and the firm's only levers are pricing, utilization, and seniority mix. All three are crowded. Pricing is anchored by procurement norms. Utilization ceiling is around 70 to 75 percent before burnout starts eating retention. Seniority mix (leveraging juniors under a senior) is the traditional scaling move, and the move every competitor has already made.
This is why the consulting business has been boring to capital allocators for thirty years. Margin is capped by people costs, people costs scale roughly linearly with revenue, and there is no compounding asset.
Claude changes every one of those statements. Not by a little.
Revenue per operator: the number that moves.
A senior security operator in a traditional firm bills around 1,400 hours a year at $325 per hour. That is $455,000 in annual revenue per operator at full utilization. In practice, utilization is 65 to 72 percent, so $300,000 to $325,000 is the realistic revenue-per-head number at the boutique level.
Our measured internal number, running the same class of engagement with Claude embedded in the workflow, is $680,000 to $750,000 per operator in year one. That range will move as we build more leverage. Let us walk through where it comes from.
Engagement throughput.
A senior operator, without Claude, ships three to four growth-tier pentest engagements per quarter. With Claude, the same operator ships six to eight. Recon synthesis, inventory review, report drafting, and client-comms glue are roughly 35 to 45 percent of the operator's engagement hours, and roughly 70 percent of that work is the portion Claude accelerates. The math works out to something like a 1.6 to 2.1x engagement-throughput multiplier per operator.
We do not claim 5x. Anyone claiming 5x is counting wrong. The part of the engagement Claude does not touch (discovery calls, client trust building, live exploit validation, oral findings readout) is not going away and still takes operator time.
Price per engagement.
Our price per engagement is roughly flat to traditional. We priced it that way on purpose. The AI-native efficiency is the delivery engine; it is not a discount we pass to the client. A $42,000 Series A pentest is still $42,000. The delta goes into margin, not into a race to the bottom on hourly rates.
The essay on why hourly billing is a moral hazard under AI-accelerated delivery explains this choice in more depth. The short version: if we bill by the hour and we get 2x faster, we cut our own revenue in half. So we do not.
COGS: people, Claude, and the shifting mix.
COGS on an AI-native engagement has two ingredients: the operator's loaded hourly cost, and Anthropic API spend for the Claude work embedded in delivery. Let us size them separately.
Operator cost per engagement.
Senior operator loaded cost (fully loaded, including benefits, tools, training): $220 per hour. A traditional Series A pentest takes 80 to 120 hours of operator time. At $220, that is $17,600 to $26,400 in operator cost against a $42,000 engagement.
Same engagement on our workflow: 45 to 65 operator hours. The operator still owns recon strategy, exploit validation, severity calibration, oral readout. Claude eats the glue work. At $220 per hour, that is $9,900 to $14,300 in operator cost.
Claude cost per engagement.
Mixed model usage, tuned per tier as laid out in our research-budget essay. A growth-tier pentest engagement runs between $180 and $340 in Anthropic API spend, with prompt caching on the static parts of the handbook templates cutting input tokens roughly 10x on repeated calls (see the prompt-caching unit economics essay for the math).
Round it up: $300 of API cost to produce an engagement that earns $42,000 of revenue. That is 0.7 percent of revenue spent on Claude. Claude is not the expensive line. It is not even close to the expensive line.
New COGS picture.
Same $42,000 engagement. Operator cost: $12,000 median. Claude cost: $300. Total COGS: $12,300. Gross margin: $29,700, or roughly 70 percent.
Compare that to the traditional $42,000 engagement with $22,000 of operator cost (52 percent COGS, 48 percent gross margin). The AI-native engagement did not become 2x more profitable. It became roughly 2.3x more profitable on gross margin dollars, and shifted gross margin percentage from 48 percent to 70 percent. That 22-point gross-margin shift is the whole story.
Template library as a balance-sheet asset.
Every traditional consulting firm produces a pile of internal deliverable templates in Word and PowerPoint that sit on a SharePoint and decay. The AI-native equivalent is a versioned, tested, composable asset that produces compounding leverage across every future engagement.
This is the part of the model traditional firms cannot replicate by copying. Our Signature Handbook system, and the template library behind it, is the equivalent of the firm having its own proprietary delivery platform. Every engagement we ship improves the next one, because we learn what to add to the templates. That is a compounding asset on an intangible line nobody tracks.
We value this at two things. First, in a traditional firm, the senior operator's tacit knowledge walks out the door when they leave. In an AI-native firm, the knowledge lives in the template library and the eval harness. Retention risk is quietly lower. Second, in a sale or capital event, the template library is a defensible asset; the delivery firm next door can hire the same junior operators we hire, but they cannot replicate four years of tested prompts, golden datasets, and regression tests overnight.
We do not (yet) carry the template library on the balance sheet. We should. In a year or two, we will. But internally it already shows up in how we compensate: template authorship is a named contribution with its own review cadence, not an undifferentiated piece of "the work."
Margin ceiling vs margin compression.
Here is the uncomfortable part. The margin ceiling is higher. The margin-compression pressure is also real, and we need to be honest about both.
The ceiling: if we execute well, gross margins in the 65 to 75 percent range are defensible for three to five years. We get there because Claude handles the glue work, because the template library compounds, and because pricing stays fixed-scope. Nothing on our cost base scales linearly with revenue the way operator headcount did in the traditional model.
The compression: every competitor will eventually also use Claude. The window where AI-native firms have a pricing advantage over traditional firms is open right now, closing in 12 to 24 months. When the window closes, the pricing pressure will show up as clients asking "why is your engagement $42,000 when a firm down the street that uses the same Claude charges $26,000." We will not win that argument on technology alone. We will win it on taste, on template maturity, on named-operator continuity, on how the handbook reads, on references.
The strategic rule is: bank the margin now, invest it in template depth and operator taste, and build a firm that is still defensible when Claude is table stakes. This is a two to three year window, not a permanent advantage. Treat it accordingly.
The three strategic choices this forces.
An AI-native P&L is not just a better traditional P&L. It is a different instrument, and it forces choices a traditional firm does not have to make.
Choice 1: volume or margin.
At 70 percent gross margin you can either take the extra margin home, or you can plow it into lower prices and win more engagements at the same total profit. Both are defensible. We chose margin on our own P&L: same prices, fewer but better engagements, and the operating margin shifts from 12 percent to the 30 to 35 percent range. The volume bet would have meant 1.5 to 2x engagement count at 12 percent operating margin, which is a different firm altogether.
Choice 2: operator leverage or operator ceiling.
The traditional leverage model (senior partner over juniors) is how you scaled to 50 people. In AI-native, one senior operator can ship six to eight engagements a quarter unassisted. You can run this firm as a small group of senior operators (call it the "ceiling" model) or keep the leverage model and pair seniors with juniors who are learning. We run the ceiling model for the first two years because client trust is still bought by senior-on-engagement, and because training junior operators to ship AI-augmented work before they have traditional delivery instincts produces worse outcomes. Year three we will reopen the question.
Choice 3: services or product.
The template library is the seed of a product business. Share-of-Answer Monitor, Eval Harness as a Service, Nexmill template licensing. Every AI-native services firm faces the choice of whether to stay services or split off product. The services business at 30 percent operating margin is a very good business; the product business at scale is a different business with higher risk and multi-year investment. We are building both tracks and keeping them on separate P&Ls, because mixing them hides both stories.
What our own P&L actually looks like.
Here is our internal Year 1 model, as filed in the business plan. Same shape as the traditional firm example, different numbers.
- Revenue: $2,000,000 (Q4 exit run-rate)
- COGS (operator time + Claude API): $660,000 (33 percent)
- Gross margin: $1,340,000 (67 percent)
- Sales and marketing: $160,000 (8 percent)
- G&A (non-billable, tools, legal, insurance): $220,000 (11 percent)
- Founder compensation: $280,000 (14 percent)
- Operating margin: $680,000 (34 percent)
Full model with assumptions lives in the handbook and the business-plan financial model document. The 34 percent operating margin is roughly 2.7x the traditional firm example above, on less than half the headcount. The risk is entirely on the operator-retention and template-library fronts; our COGS is dominated by operator time, and if a named operator leaves, engagements stall.
That is the P&L Claude rewrites. Not "same firm, slightly better," but "different shape, different risks, different strategic choices." The firms that treat it as a small incremental efficiency are going to lose to the ones that treat it as a structural rewrite.
Your P&L is not going to look exactly like ours. The principle that moves is this: if you are not reshaping your COGS, your pricing, and your asset base when you adopt Claude, you are leaving the whole advantage on the floor. Run the numbers on your own delivery. Find the line where Claude can cut 40 percent of operator hours. Do not pass that saving to the client as a discount. Reinvest it in template depth and operator retention. Measure in a year.
Operator-tone writing on Applied AI, Security, SEO, and Economics.
One essay per week. No hype. No tracking pixels. Unsubscribe in one click.