What Being Wrong Costs

The number nobody has

There is a question that reliably produces silence in a planning review. What does one unit too many cost this business, and what does one unit too few cost it? Not directionally. In currency, for this item, at this location, this quarter.

The silence is not incompetence. The question is genuinely hard, it crosses at least three functions, and no system in the organisation is designed to answer it. But the consequence of leaving it unanswered is larger than most executives assume, because that single number is what converts a forecast into a decision. Without it, a demand model produces an estimate and nothing more. Somebody still has to decide what to do about the estimate, and lacking a price, they will decide by policy, by habit or by argument.

A distribution without a cost function is a weather report. Interesting, occasionally alarming, and not a decision.

This paper is about building that number. It is the least glamorous artifact in the architecture, requires no technology whatsoever, and in our experience determines more of the eventual return than the choice of forecasting method, the choice of platform, and the choice of vendor combined.

Why service level is not an objective

Most organisations do have something in the place where the cost of error belongs. It is usually a service level: 95 percent, 98 percent, sometimes 99.5 percent for the items someone has designated as critical. It functions as an objective, it is reported monthly, and it is defended vigorously. It is worth being blunt about what it actually is.

A service-level target is a percentage nobody derived. Ask where 98 came from and the answer is almost never an economic calculation. It is inherited, benchmarked against a competitor who inherited it themselves, or negotiated as a compromise between a commercial function that wanted more and a finance function that wanted less. It is then applied more or less uniformly across items whose margins differ by an order of magnitude, whose holding costs differ by more, and whose substitutability ranges from perfect to none.

Consider what uniform application implies. A high-margin item that a customer will not substitute and that costs almost nothing to store is being deliberately under-served at 98 percent. A low-margin bulky item with a short shelf life and three ready substitutes is being wildly over-served at the same number. Both errors are invisible, because the metric reports the same figure in each case and reports it as success.

Exhibit 1 is the entire argument of this paper in one picture. The cost of holding rises gently and roughly linearly. The cost of running out rises steeply and without any natural ceiling. The minimum sits where the two balance, and its position is determined by the economics of that specific item, not by a corporate percentage. A policy expressed in percentages cannot find that point except by coincidence, and it has no mechanism for noticing when it has missed.

The half of the error that never appears

The cost curve has a second property that matters more for organisational behaviour than for arithmetic. The two sides of it are accounted for very differently.

Everything on the left of Exhibit 2 has an owner, a monthly report and a reduction target. Everything on the right has none of those things. No general ledger account is debited when a customer places an order with a competitor. No variance report shows the trial that was never run because stock was committed elsewhere. The costs are entirely real and frequently larger than those on the left, and they are structurally absent from the instruments management uses to see the business.

This produces the most consequential and least discussed bias in operational planning. An organisation optimises what it is charged for. Faced with two errors of similar magnitude, one of which appears in next month's variance report and one of which appears nowhere, a rational manager reduces the first. Repeated across thousands of decisions and several years, this is not a rounding effect. It is a structural tilt, and it operates without anyone intending it, which is why it survives changes of leadership.

Building the ledger is therefore not primarily an analytical exercise. It is the act of making one side of the business's error visible for the first time, so that it can be traded against the other side deliberately rather than by default.

How to build the ledger in six weeks

The ledger is a single agreed set of per-unit costs, signed by finance, covering each direction of error and each irreversible consequence. It is a document, not a system. What follows is the shape of the exercise as we run it.

Four of the six lines are straightforward and can be established from records that already exist. The fifth is where the exercise becomes genuinely difficult and where most attempts quietly stall.

Our guidance on customer damage is to resist the instinct to model it precisely and instead bound it and move on. A defensible range agreed by the commercial director beats an elegant model nobody signs. The decisions the ledger informs are rarely sensitive to the third significant figure, and a value that is roughly right and actually used is worth considerably more than one that is precisely right and still in draft. Refine it later against realised outcomes, which is exactly what the recalibration loop in Paper 06 is for.

Getting it signed, and surviving the argument

Expect resistance, and expect it to be concentrated in one specific place. Nobody disputes the cost of a write-off. The argument is always about the invisible half, and it takes a predictable form: that lost sales cannot be measured, that any number would be arbitrary, and that committing to one would be imprudent.

The reply worth having ready is that the organisation is already using a number. It is using zero. Every decision taken under the current process implicitly prices the customer who went elsewhere at nothing, because nothing is what appears in the accounts. The choice is not between a precise number and no number. It is between an explicit estimate that can be debated and improved, and an implicit estimate of zero that cannot. Stated that way, the imprudent option is the status quo, and the conversation usually turns.

Three further points of practice. Get the signature from the finance business partner who owns the profit and loss the decisions affect, not from a central function, because local ownership is what makes the number stick when it is later inconvenient. Publish the ledger rather than embedding it in a model, so that anyone can see what their decisions are being priced against. And schedule the first recalibration before the first use, so that it is understood from the outset as a maintained instrument rather than a one-off study.