How to tell if your growth is compounding — or you're just renting it
by Luis Gomes, Founder & Growth Lead
Two brands end the year with the same revenue curve. Same shape, same slope, both up and to the right. The first one got there by adding: a second agency, two more contractors, three new tools, each bolted on when a number needed moving. The second one got there with roughly the team it started with, running a system that was a little sharper every month. From the outside — from the revenue line alone — you cannot tell them apart. The difference shows up the day each one stops adding. The first brand's growth stops with it. The second's keeps going.
There is a number that tells you which brand you are long before that day, and almost nobody puts it on a slide. It's the cost to produce a unit of growth, watched over time.
Revenue tells you people are buying. It doesn't tell you the work compounds.
Revenue going up is the easiest thing in growth to mistake for proof. People are buying — that's real, and it's good. But it says nothing about how the growth was produced, and that's the part that decides whether it lasts. You can buy a quarter of revenue growth the way you buy anything: pay more, get more. Another agency produces more campaigns. More ad budget buys more clicks. A new tool ships more variants. The output goes up because the spend went up. That's not a flywheel. It's a treadmill — the moment you step off, you stop moving.
Compounding growth is the other thing. It's growth where the work gets cheaper to produce as you go, because the system keeps what it learned last time. The winning angle, the dead ends you already paid to discover, the brand rules, the customer signal — written down once and reused, so the next cycle starts ahead of the last one.
The number, and why the trend is the whole point
Here is the metric. Take everything you spend to produce one unit of growth — and "fully-loaded" means everything: agency fees, contractors, the software, plus the real slice of your own team's time. Divide it by the units of growth you produced. A "unit of growth" is whatever your business actually counts: a qualified opportunity, a customer acquired under your target cost, a unit of demand created. The result is your cost per unit of growth.
The single quarter's figure barely matters: a competent rented operation can post a perfectly respectable cost per unit. What matters is the trajectory — plot that number over the last four or six quarters and look at the slope.
If it's falling — each unit costs less to produce than the one before — your growth is compounding. Something in the system is being retained and reused, so you're not paying full price to relearn what you already knew. If it's flat while your spend climbs — you produce more, but each unit costs about what it always did — you're renting. You bought more output at proportional cost. Nothing accrued. Stop paying and it stops.
A worked version, so it's concrete
Say a brand spends, in one quarter, a fully-loaded $200,000 — agency, tools, and the fraction of two internal people's time — and that produces 400 qualified opportunities. That's $500 each. Fine number.
A year later it's producing 800 opportunities. Growth doubled; everyone's pleased. Now do the division. If getting to 800 took a fully-loaded $400,000 — a second agency, more tools — it's still $500 each: the line is flat. The brand rented its way to twice the output and now pays twice as much to stand in the same place; the day the budget tightens, it halves. If instead those 800 came on $360,000 — because last year's winning structures, rules, and research were reused instead of rebought — it's $450 each, down from $500, and still falling. Same doubled revenue on the headline. One brand built an asset; the other rented a result. The cost-per-unit line is the only place on the page where they look different.
Numbers here are illustrative — the point is the slope, not a benchmark.
Why nobody shows you this line
If the trajectory is the proof, why is it never on the slide? Because the market sells the other things — activity and logos. Campaigns shipped, channels live, "we use AI," marquee clients, a revenue milestone. All of it is easier to show and none of it is the line.
It's not only agencies. We sampled ten of the most-cited AI-services companies in June 2026 — the ones whose entire pitch is software-grade leverage. Every one published traction: logo walls, claim totals, "resolved without a human touch" percentages. Not one published gross margin or the share of the work that's still human labor — the figures that would show whether the leverage is real or whether revenue is just buying proportional headcount. Emergence Capital, which funds this category, names the exact tell in its own playbook: watch for when "gross margin is flat or declining even as revenue grows" and "revenue per employee (ARR/FTE) isn't improving." The companies selling compounding won't show the line that proves it. And a traditional agency can't — its product is priced by headcount and retainer, which is the definition of a cost-per-unit that doesn't fall.
The line only bends down when the learning is written down
The reason the number falls is not mysterious and not magic: the cost drops because what failed last cycle became a rule the system reads on the next one, so you stop paying to rediscover it. That's the whole mechanism, and it has one condition — the learning has to live somewhere the system actually reads, in files you keep, not in a contractor's head or a platform you'd lose access to the day you left. Where it lives is the difference between installing a system and renting one: rent the intelligence and the compounding leaves the day the contract closes. Eric Siu makes the same point about agencies: rented labor resets every time you stop paying for it, while the infrastructure you own keeps compounding. This is also the trend version of a test we've made before, that the proof of an AI-native operation is margin, not revenue or logos — except a single quarter's margin can be staged. The slope across a year can't.
When renting is the right call
Renting growth isn't a sin, and a flat cost-per-unit isn't always a failure. Early on, before you know what works, paying for output you don't retain is a reasonable trade — you're buying the time to find the thing worth compounding. A seasonal spike, a one-off launch, a market test: rent the capacity, get the result, move on. The mirage is narrower than "you pay for growth." It's specifically the operation that tells the compounding story — AI-native, leverage, a flywheel — while its cost-per-unit line has been flat for a year. If the story says it's compounding and the trajectory says it's flat, believe the trajectory.
So run the number on your own operation before you run it on anyone else's. Revenue tells you people are bought in. The cost-per-unit trend tells you whether you built something or rented it — and it's the one line worth asking any agency, tool, or team to show you, going back four quarters, before you renew. Including us.