Compounding
Returns applied to returns. Boring for long enough that most people quit before the curve arrives.
The Model
A linear process adds. A compounding one multiplies, because the output of each period becomes the input of the next.
Bernoulli worked out the limit case in 1683 while studying how interest behaves when it is applied continuously rather than annually. The mathematics has not changed since. What changed is that operators now use it as a metaphor for things that are not money, and mostly use it wrong.
Two properties govern every compounding system, and both are counterintuitive.
The curve is flat for a long time and then it is not. Not gradually steeper — flat, flat, flat, then vertical. Which means the period during which you receive the least evidence is the period during which quitting is most tempting and most costly.
And the exponent is time, not intensity. You cannot compensate for a shorter run with a harder effort, because the term you are trying to substitute for sits in the exponent and the one you are increasing does not.
That gives you the dominant variable in any compounding system, and it is not growth rate. It is survival — staying in a position to keep making deposits.
Why operators get this wrong
Founders understand compounding as arithmetic and then fail it as identity. They believe they are compounding when what they are actually doing is restarting.
The clearest case is a brand that rewrites its positioning every nine months. Each rewrite is individually defensible — the market moved, a competitor shifted, the new hire had a better idea.
None of them compound. The audience's memory resets, and the forty assets built under the previous story stop reinforcing anything. Four years of work produces the recognition of one year, four times.
Content has the identical shape. Two hundred posts across four unrelated topics is not the same asset as two hundred posts on one, and the difference is not marginal. It is the difference between a body of work and a feed.
The second failure is the interruption, and it is more expensive than any of this. A business that grows sixty percent for three years and blows up in the fourth returns less than one that grew twenty percent for four.
Every operator knows this. Most take the tail risk anyway, because the sixty looks like competence and the twenty looks like caution.
Applied
Body
The relevant unit is a decade, not a twelve-week block.
Two operators start on the same day. The first trains five days a week at a load that leaves nothing on the table and gets injured roughly every fourteen months, losing six to ten weeks each time to a shoulder, a back or an elbow.
The second trains three days a week at a load he could repeat next week and the week after, indefinitely.
At month twelve, the first is visibly ahead and knows it. At year eight, the second has accumulated something near twelve hundred uninterrupted sessions, a stable technical base, and joints that have never been taught to flinch. The first has a training history made of restarts.
Consistency compounds. Intensity that costs you availability does not compound at all — it resets the counter, and the counter is the only thing that was working.
Business
Sort every use of money and time into two piles: assets that raise the baseline, and expenses that produce a result once.
An advertising campaign is consumed. Spend stops, traffic stops, and next quarter starts from the same place. A ranked article, an email list, a documented process, a customer who buys again — each of those raises the level the next effort begins from.
Take two agencies, both at six hundred thousand. One reinvests in delivery quality that generates referrals. One reinvests in paid acquisition.
For eighteen months the growth curves are indistinguishable and the second founder looks smarter, because his growth is legible and controllable. Then acquisition costs rise forty percent in a quarter, and only one of them still has a business.
The cleanest number in the whole model is net revenue retention. Above one hundred percent means the company grows with zero new customers. That is compounding, visible in a P&L, and almost nobody under ten million measures it.
AI Leverage
The compounding asset in AI work is not the model. Everyone has the same models, within about six months of each other, at roughly the same price.
The asset is your context. A written, versioned corpus of how your brand sounds, what your offers are, which objections actually come up, what you decided last year and why — that is the thing that makes every future output better, and it is the only part your competitor cannot buy.
So change where corrections go. Every time you fix a model's output, the fix belongs in a file, not in the chat window. The chat is consumed; the file compounds.
Do that for twelve months and a commodity model, pointed at your corpus, outperforms your competitor's identical model pointed at nothing. The edge was never the tool. It was the accumulated specificity you fed it.
The same logic condemns the opposite habit. Teams that treat each prompt as disposable start from zero every single time, forever, and mistake the resulting effort for work.
The Drill
Name the one asset you want measurably larger in twenty-four months. One. Not three.
Then set the deposit small enough that missing it would be embarrassing. Thirty minutes, twice a week, in the calendar, with a named output — not "work on the newsletter" but "one issue drafted."
Track exactly two numbers. Deposits made, and weeks since the last gap longer than fourteen days.
The second number is the one that decides the outcome, and it is the one nobody tracks. A gap of two weeks is a pause. A gap of six is a restart, and a restart puts you back at the flat part of a curve you had already paid for.
Review both numbers on the first Monday of each month. If the deposit is being missed, the deposit is too large. Cut it in half rather than renewing the commitment.
Stoic parallel
Cleanthes hauled water at night so he could study with Zeno during the day. He did it for years, was mocked for being slow, and was nicknamed the ass — an animal that carries loads no other will carry.
He took the name as accurate. When Zeno died, Cleanthes led the Stoic school for over thirty years, and the doctrine that reached Rome and eventually Marcus came through him.
The Stoics did not think character was acquired in an insight. They used the word askēsis — training, practice, the same word the athletes used — and they meant repetitions performed on ordinary days with no observable result.
Marcus's notebook was not a record of his thinking. It was the deposit.
He was not writing to remember. He was writing because the writing was the exercise, and he did it on nights when nothing came of it.
One model per week.
Applied to training, business, and AI leverage. No fluff.
Related models
- Second-Order ThinkingAnd then what? The first consequence is obvious and priced in. The second one is where the money is.
- Pareto PrincipleResults are not spread evenly across inputs. Measure the distribution before you decide where the effort goes.
- LeverageOutput per unit of input. Change the multiplier, not the effort behind it.
Origin: Jacob Bernoulli, who derived the mathematics of continuous compounding in 1683; made an operating philosophy by Warren Buffett