The Math We Haven't Replaced (Pricing Series, Part 1)
The Setup
There’s a line that gets repeated in just about every AI conversation at my level right now. AI compresses the grunt work first, so the cheap junior hours are the ones that go. I’d nodded to and repeated this version of it plenty of times. Then a colleague who runs one of our bigger accounts told me their AI enablement team was auditing every deliverable, every level, every hour. And the question the audit kept circling back to was, does a VP really need to do all these reviews. “Couldn’t someone more junior handle that?” I told them to push back. The hours a client buys at the senior level are judgment bundled with effort, and AI can compress the effort without touching the judgment. I believed that when I said it. I just hadn’t checked it against our own numbers. So I went and looked. What I found is the thing this whole post is about. We all say value is moving up to judgment. I checked whether the business model agrees. It doesn’t. And we don’t have a new one yet, so we’re still running on the old math while telling a new story.
My Approach + AI Role
I was modeling capability and contribution margin, trying to understand the operating model a layer down from where we usually look. AI did the assembly: pulling rate and hours data together, cutting it by level, letting me slice the same picture a few different ways so I could see the actual shape instead of the one I assumed was there. I was looking for something else entirely. And this caught me off guard. (Running it twice and cutting it differently is the work I trust most. Not the first answer. Usually the third one.)
What Actually Happened
The numbers disagreed the exact thing I’d just told my colleague to defend. It’s higher margin on the execution tier and a lower margin on the senior tier. The mid-bench semi-experienced folks actually producing the work carried the highest contribution margin in the stack. The senior reviewers carried the lowest. That’s not unique, by the way. It’s the leverage model. David H. Maister wrote it down forty years ago. You bill juniors at a multiple of their cost, so the margin lives in the executional layer. The expensive seniors are the low-margin part. Any business school teaches it… at least I hope they do, I never went to proper business school. So the work AI is readiest to compress, the executional layer, is also the high-margin layer. And the work it struggles to touch, the senior judgment, is the low-margin one. The cheapest hours in the building are the most profitable ones, and those get compressed first.
The Real Insight
The consensus right now is that value is moving up the chain. AI commoditizes execution, so what clients pay for shifts to judgment, outcomes, the work at the edges AI gets wrong. You can read a version of it in many trade publications this year. I think it’s basically right. Put the margin math next to it and you get a problem.
- Where we say value is going: up the chain, to senior judgment and outcomes. Low margins.
- Where the model actually makes its money: the junior production layer… the exact hours AI compresses first.
- What we still measure and bill by: time. Which rewards the layer that’s leaving and under-rewards the layer we keep calling the future. Three signals, and they don’t agree. The pricing is still hours. The value story is judgment. And we’re acting like those already line up. They don’t, not yet. But the old model is the only model anyone’s billing on, so we keep using it.
Try This If…
If you own a P&L: before you model where AI saves you money, run one analysis you probably skipped. Not “what’s cheapest to automate.” Ask “what’s our margin on the work AI is readiest to do,” tier by tier. Then sit the two stories side by side: the one about where value is moving and the one your own margins are telling. See if they point the same way. Mine didn’t. And if you don’t hold the spreadsheet but you sit in the rooms where this gets decided: when an efficiency conversation collapses into “which hours can we cut by using AI,” the question that slows the room down is whether they mean cheaper hours or more profitable hours. Math is math, but those two measurements aren’t the same thing.
Systems Lens
There’s another side of this and it’s closing rapidly. If AI does the execution the juniors used to do and you hold the old price, your cost on that layer drops and the margin spikes. That’s happening right now. The catch is it’s a lag, not a moat for businesses to build around it. It lasts as long as the price stays put, and the price stays put only until someone reprices the work. The more tightly your pricing is pegged to hours, the faster that window closes, because the moment the work takes fewer hours, the meter you taught your client to read starts ticking down on its own. So yes, there’s money in the gap. And it’s closing. And underneath all of it is a harder problem. The part of the work that compounds, the judgment, the client relationships, the memory of why a call was made two years ago, doesn’t have a number. The measurement system can see throughput and cost. It can’t see the thing that makes the senior layer worth buying. So when you optimize on what you can count, you can end up eroding the thing you can’t, and the dashboards look better the whole way down. I don’t think the tension is the scary part. I think the energy spent pretending it isn’t there IS. I’d rather write it up half-finished and out in the open than wait until it’s solved, because I think naming it plainly is how more of us start working on it together. Which leaves the question I can’t answer yet, and that’s Part 2 Field Note Friday I’m unpacking. If the most profitable hours are the ones most likely to disappear, why are we still pricing the work by the hour at all? If you’ve run this version on your own mathing, I’d like to know whether your numbers do the same thing mine did, or whether I’m looking at one account and calling it a pattern.