The size curve is a markdown you have already bought

3 October: one number entered on a purchase order

The buy for OS-2214 is 24,000 units at USD 12.40 FOB, selling at USD 79.00 in the shop. Landed cost — the factory price plus freight, duty and handling, all the way into the warehouse — is USD 15.90. So a unit sold at full price earns USD 63.10 of gross margin, and a unit taken to the first markdown of 40% loses USD 31.60 of it.

The size curve is the share of the buy given to each size. It is entered on the purchase order on 3 October. It takes about a minute, it is copied from last year, and nobody in the room thinks of it as a forecast.

It is the largest forecast in the order. By the time the first garment is cut, most of what it will cost has already happened.

Three curves, one year of data, and they disagree

Ostley's planner has three defensible ways to read the same year.

The three answers for size 18 are 510 units apart on one style, and each was reached honestly.

Gross sales counts a garment that came back as a sale, which overstates what stuck. Net of fit returns counts only what stayed. That is the standard, careful, defensible way to build a curve — and look at what it does. It moves weight off exactly the sizes lesson 4 showed are badly graded, and onto the middle. The scanned customers is a count of the bodies in the loyalty file, each one sorted into the size band its measurements fall in. It has nothing to do with what was hanging in the shop.

The mistake nobody made

Ostley's planner buys on the net curve. That is correct practice and it is what a careful planner does.

Ostley's technical team keeps the grade rule, because nobody has ever connected a returns table to a pattern room.

So size 18 is bought at 1,410 units instead of 1,920. Next year there were fewer size 18 garments on sale, so fewer sold, so the net curve says size 18 is smaller still. Two or three seasons of this and the range has walked away from the top of its own size run. Not by a decision anybody made or wrote down, but as the arithmetic consequence of a grade fault being read as a demand signal.

The buy ratio and the grade rule are the same decision, and they are taken in different rooms by people who do not have each other's numbers. A grade fault that is not fixed does not stay a fit problem. It becomes a buying decision, and then it becomes permanent.

What a small curve error costs

Take the buy on the gross curve and set it against what the corrected range would actually have sold. The largest error on a single size is 1.0 percentage point, and the errors across all nine sizes add up to 6.0 points.

USD 68,184.00 on an order worth USD 297,600 at FOB. That is 4.50% of the whole order's gross margin at full price, from a ratio that took a minute to type.

Two things about that number, said in the open rather than in a footnote.

It assumes a customer who cannot find her size buys nothing. That is Ostley's planning convention and it is the harsh end. If a third of them take the size below — Ostley's own measured substitution rate is 35% — the number falls to USD 52,282.80. It does not fall to zero under any assumption, because the surplus units still get marked down whether or not somebody buys a neighbouring size.

It is symmetrical, and that is not a coincidence. The surplus and the shortfall are both 720 units, because every unit put in the wrong size is a unit missing from the right one. That gives you a rule worth keeping: the units in the wrong place are half the sum of the absolute curve errors, times the buy. Six points of total error is three points misplaced, and three points of 24,000 is 720. You do not need a merchandise plan to work out your exposure. You need the gap between your curve and the truth, and one multiplication.

At this style's economics, one percentage point of the buy landing in the wrong size is worth USD 22,728.00.

What this curve is not

The scanned curve is an estimate too, and an honest lesson says how it could be wrong. It counts the bodies in Ostley's loyalty file, and a loyalty file is made of people who already buy successfully. So if the range has been failing size 18 for years, size 18 customers are under-counted in the very file used to prove they exist. The measured curve is therefore probably still too low at the top, not too high. That is the direction of its error, and knowing the direction of an error is most of what you need from it.

What to take to your own buy

1. Build the curve from the sales record and from a body count, and look at both. Where they disagree, the disagreement is the fit fault.
2. Never build a demand curve from net sales without checking the return reasons first. Net sales punish exactly the sizes a grade rule is failing.
3. Work out the exposure before the ratio is entered: half the sum of your absolute curve errors, times the buy, times the markdown per unit. It is one line, and it turns an argument about instinct into an argument about money.
4. Say out loud, in the room, that the markdown is being decided today. It is the only moment when the ratio is cheap to change.