The Speed Felt Like Progress. The Margin Did Not Move.
Most services founders spent the last twelve months folding AI into delivery. Tasks that used to take days now take hours. Decks come back faster. Reports get drafted before the meeting ends. The team feels less stretched, the client feels well-served, and on the surface the business looks healthier than it has in a long time. So the founder closes the quarter expecting the P&L to confirm the story.
It does not. Revenue is up, sometimes meaningfully. But gross margin has not moved, or has moved the wrong way. The leadership team puts it down to a few one-off cost lines, a slow month on a key account, a hiring decision that will normalise. The speed felt like progress. The margin says it was something else.
The Tools Now Cost What Two People Used to Cost.
The structural reason is simple, and most founders have not yet stopped to add it up. The AI stack sitting inside a typical services business has grown, quietly, to the price point of two to three full-time employees per month. A core language model subscription at the team level. A separate research tool. A writing tool. A design tool. A meeting transcription tool. A workflow automation layer. Each one looks small in isolation. Together they sit on the cost line where two mid-level salaries used to sit.
This is not, on its own, a problem. The problem is what it has done to the shape of the cost base. Junior salaries flex with utilisation. A quiet month means fewer hours billed and fewer hours paid for. Software licences do not flex. They sit at full cost whether the team is at capacity or at half capacity. The cost of delivery has not collapsed. It has been rearranged, from a variable line that moved with the work into a fixed line that does not. Output is faster. The cost base is less forgiving.
We Worked With An Agency That Was Doing More Work Than Ever.
We worked with a mid-sized agency that had spent the previous year pushing AI deep into delivery. Turnaround times on client work had nearly halved. The team was producing more output per week than at any point in the agency’s history. Headline revenue was up. The founder believed the business had crossed into a different operating gear and that the next set of numbers would confirm it.
The numbers did not. Gross margin had drifted down by a few points. The combined software and AI tool spend had grown to sit close to the cost of three full-time employees. Two junior roles that would normally have been filled had been quietly left open, because the team was managing without them. Senior time on each project had crept up rather than down, because someone still had to check, judge, and stand behind the work that AI had drafted. The agency was doing more work than ever. It was making less money per unit of work than it had a year earlier.
AI Cut The Wrong Layer Of The Team.
This is the part of the AI story that almost nobody is naming out loud. AI does not displace work evenly across a services team. It displaces the cheapest layer first. The junior researcher, the junior analyst, the junior copywriter, the junior designer. These are the roles that AI is genuinely good at substituting, because the work is structured, repeatable, and reviewable. So the team stops hiring at that level, or quietly lets the layer thin out through attrition. On paper, that looks like efficiency.
What it does in practice is invert the cost pyramid. The senior layer is now responsible for everything AI cannot yet do well: judgement, review, client ownership, quality control, the parts of the work where being wrong is expensive. Senior hours per project go up rather than down. And the senior layer is the most expensive labour in the business. Most services firms have not been built for this shape. They were designed around a pyramid of junior execution feeding into a smaller senior layer of judgement. AI has flattened the bottom of that pyramid without rebuilding the top. The cheap labour is gone. The expensive labour is stretched. There is very little structural capacity in the middle, where most of the work used to be absorbed.
But We Cannot Compete Without It.
The most common founder response, when this is laid out on the table, is that there is no choice. AI is now the price of entry. A services business that does not use it cannot match the turnaround times, the price points, or the volume expectations of the market. This is true. Conceding it is important, because the argument that follows only works if the founder is not being asked to retreat from AI. They are not. They are being asked to look at who they are actually competing against.
The competitive set has changed. The relevant competitor is no longer a similarly structured agency carrying the same OPEX base, running the same software, paying the same office and management costs. The relevant competitor is increasingly an AI-first business, built from scratch around a leaner cost base, with a smaller senior team, no junior bench to maintain, and an operating model that was designed around the tools rather than retrofitted to them. A legacy services business cannot match the margins of that competitor by bolting AI onto its existing cost structure. The bar has moved underneath the business, and the structure has not moved with it.
AI Is A Relationship You Cannot Easily Leave.
The other thing almost nobody is pricing in is what happens to the tool cost from here. AI platforms are currently in their growth phase. Pricing is subsidised, free tiers are generous, enterprise terms are flexible, and the cost per seat is artificially low while the platforms compete for share. This will not last. Most of the major AI platforms are heading into their profit-making phase, and the historical pattern in software is unambiguous. Prices go up. Free tiers narrow. Per-seat costs rise. Usage-based pricing models replace flat ones.
The trap is that AI adoption, once it is wired into delivery, is very difficult to reverse. A services business that has redesigned its workflow around a particular set of tools cannot quietly unhook itself when the renewal quote arrives with a forty percent increase. The team has built habits, the clients have built expectations, the turnaround times have been quoted. AI has moved from a discretionary cost into a structural one, and structural costs are the hardest to negotiate down. A business that adopted AI without a plan is now locked into a cost line that will not stay where it is.
Two Lines On The P&L That Tell You The Truth.
There are two places on the P&L where this story shows up before the founder feels it. The first is software-as-COGS, the share of revenue being absorbed by AI and software tooling that sits inside the cost of delivery rather than in general overhead. Twelve months ago this line was usually a rounding error. In many services businesses it now sits at several percentage points of revenue, and it has grown faster than gross margin has moved. If software-as-COGS has climbed by three or four points and gross margin has not improved, the AI investment is not paying for itself at the unit level. It is being absorbed by the business rather than the client.
The second line is OPEX, and this is the one most founders are not yet looking at correctly. A services business that relies heavily on AI is moving structurally closer to an outsourced model, where a significant share of the work is performed by something the business does not directly employ. Outsourced models can be profitable, but only on a thinner gross margin, and only if the OPEX underneath is genuinely lean. Most services businesses have added AI cost on top of an OPEX base that was built for a fuller team, a more layered structure, and a different operating model. The gross margin has compressed in the direction of an outsourced business. The OPEX has not. That is the gap, and it is where the margin is leaking.
The Speed Was Real. The Plan Was Missing.
AI did not fail these businesses. The plan around it did. Founders who treated AI as something to bolt onto delivery are now sitting with thinner margins than the AI-first competitors they thought they were keeping pace with, carrying a tool stack that costs the equivalent of two or three salaries, supporting a senior team that is doing more review work than ever, and watching the software line creep up faster than the margin can absorb. The work from here is not to add more AI. The work is to rebuild the cost base underneath it, so that the speed AI created actually shows up where it matters.
If you suspect the speed in your business has not yet translated into margin, this is a useful conversation to have. AI will not fix a cost structure it was never designed for.























