Benjamin Graham's argument, made for investing in 1949, was that the analysis will sometimes be wrong and the only protection that does not depend on being right is a gap between what you pay and what the thing is worth. Buy at sixty cents on your estimate of a dollar, and the estimate can be meaningfully wrong without costing you money.

The structure generalizes past investing, because the premise does: you cannot forecast well enough to run without slack, and what you get in place of a margin is exposure rather than precision.

Why it is not just forecasting low

This is the distinction that makes it useful rather than a synonym for caution.

A pessimistic forecast is still a forecast. It is a single number, arrived at by the same reasoning as the optimistic one, and it is wrong in the same ways for the same reasons — the pessimist and the optimist both missed the thing neither of them thought of.

A margin of safety is a different operation: make your best estimate, then leave room for the estimate itself to be wrong. It is an allowance for error in the method, not a lower point on the same method. Which means it should be sized by how uncertain you are, not by how bad you think the outcome will be. A forecast you are confident in needs a smaller margin than one you are guessing at, even if the guess is more optimistic.

Where it belongs in a business

Above break-even. Break-even tells you the floor. The margin is how far above it you deliberately sit, and it is worth holding as a number of clients rather than a sum of money, because clients are what actually leave.

In cash. Enough to run for a known number of months with no new revenue. The figure is worth knowing precisely, because the vague version — "we have a buffer" — is the version that turns out to be six weeks.

In capacity. A team booked to 100% has no margin, so every illness, every piece of rework and every urgent client request comes out of quality or out of someone's evening. Queueing behaves badly near full utilization, which is Little's law making the case in arithmetic rather than in feelings.

In time estimates. Not by padding each task, which gets absorbed, but by holding slack at the level of the whole commitment, where it can be spent where it is actually needed.

What it costs, honestly

Real money, continuously.

Cash in reserve earns less than cash deployed. A team at 80% utilization bills less than one at 100%. Three clients above break-even is three clients' worth of work you did not have to take, which means capacity you are carrying and not selling.

That cost is the premium, and the reason it gets cut is that it is visible every month while the benefit is invisible until the month it is not. Anyone can look at a reserve in a good year and see idle money. The argument for keeping it has to be made before it is needed, because by the time the case is obvious the option is gone.

The related error is treating a good stretch as evidence the margin was excessive. It is not evidence of anything. Nothing going wrong is exactly what a sufficient margin looks like from the inside, and cutting it on that basis is how a business arrives at a bad quarter with no room left — particularly one carrying high operating leverage, where the fixed costs continue regardless of what revenue does.

The relationship to expected value

Expected value tells you which option is best on average. A margin of safety is the constraint that decides which options are allowed to be considered at all.

The order matters and is frequently reversed. Rule out what you cannot survive, then optimize among the rest. Optimizing first and checking survivability afterwards produces a sequence of individually sensible decisions that ends badly once, which is enough.