Size is the mistake nobody puts on a slide. Category growth gets a review. Markdown depth gets a review. But the size curve, the split that decides how much of each style lands in small, medium, large and extra-large, is usually copied from last year and never questioned. It is also where a surprising amount of margin leaks out.
The symptom is familiar. The mediums are gone by week two and customers who wanted them leave empty-handed. The extra-larges sit until they are marked down to clear. On paper the style sold through fine. In reality you lost the full-price sales you could have made and gave back margin on the sizes you should not have bought as deep.
This is not a small leak, and it is not an occasional one. It happens on almost every style, every season, in a way that never shows up in the number anyone reviews. The category sells through, so the size curve is presumed innocent, and the same mistake gets copied forward into next season's buy. The curve is the one lever in the assortment that carries this much weight and gets this little scrutiny.
A blanket curve is wrong for almost every door
The average size profile does not exist in any single store.
A national size curve is an average, and averages describe a population that does not shop anywhere in particular. The flagship in the city skews to smaller sizes. The outlet skews larger. The mall store sits somewhere else again, and none of them look like the blended curve you bought to.
Push one curve to every door and you guarantee the same two failures everywhere at once: the sizes that sell run out early, and the sizes that do not pile up. You do not see it in the category number. You see it in availability at the size level, which is the number that actually decides whether a customer buys or bounces.
Work a quick example to feel the size of it. Say a style ships 10,000 units on a blanket curve that puts 20 percent into mediums, so 2,000 mediums across the fleet. But your urban doors, which drive a third of the volume, actually sell mediums at 30 percent of their mix. Those doors needed closer to 1,000 mediums between them and got 660, so they stock out of the best-selling size in week two and lose full-price sales for the rest of the season. Meanwhile the suburban doors, sitting on mediums they were never going to sell at that depth, carry them to markdown. Same 10,000 units. Wrong sizes, wrong doors, and the category sell-through still reads fine.
size-level availability leaders hold per door, versus the mid-eighties that a blanket split typically delivers, without shipping a single extra unit.
That last part is the point. Better size curves are not about buying more. They are about putting the units you already bought where the demand for that size actually is.
It compounds, too, in a way the single-season view misses. Because the blanket curve stocks out the mediums early, the mediums record artificially low sell-through for the back half of the season, and that depressed history feeds next year's curve, which shifts weight away from the mediums that were actually your strongest size. So the curve does not just misallocate this year; it teaches itself the wrong lesson and misallocates worse next year. Left alone, a bad curve gets more wrong over time, not less, because it keeps learning from a demand signal it corrupted itself.
New styles are where the guessing is worst
No history means the size split is a shot in the dark, so it usually copies the category.
For a carryover style you at least have last year to argue with. For a brand-new style you have nothing, so the size curve defaults to the category average, which is the least-wrong guess available and still wrong for that specific product. Launch enough newness on borrowed curves and the misallocation scales with your ambition.
The fix is to stop guessing and start inheriting. A new style can take its size curve from its nearest historical cluster, matched on category, price point and silhouette, so it launches with a curve that reflects how products like it actually sell rather than how the category averages out.
The difference is concrete. A new slim-fit chino inheriting the blanket trouser curve gets a split built from relaxed fits, wide legs and everything else in the category, which sell fuller in the larger sizes. But slim fits skew smaller, because the customer who reaches for slim skews smaller. Inherit from the nearest cluster of slim-fit trousers instead and the curve shifts left to match, so the launch ships with depth in the sizes that will actually sell it rather than depth in the sizes the broad category happens to average toward. That is the gap between a day-one forecast and a day-one guess.
The stakes on newness are higher than on carryover for a simple reason: newness is where the growth is, so it is where the volume is being poured. A brand pushing hard into new ranges is placing its biggest, least-informed bets on exactly the styles with no history to guide the curve. Get the carryover curve slightly wrong and you nick a mature style. Get the newness curve wrong at scale and you have misallocated the fastest-growing, highest-attention part of the entire buy, which is the part you can least afford to leave to the category average. The bigger the newness ambition, the more it matters that each launch inherits a real curve instead of the mean, because the mean is guaranteed to be wrong for a product that by definition is not the average.
Blanket split versus per-door demand
Same total units, allocated to how each door actually trades. The gain is entirely in the mix.
Twelve points of size-level availability recovered on the buy you already placed is not a rounding error. It is full-price sales you were losing to empty pegs in your best sizes, and markdown you were taking on the sizes you over-bought, both showing up on the P&L as if they were a demand problem when they were a distribution one.
Plan the curve at option, size, door
The curve is a demand question, so answer it with demand.
Treating the size curve as a first-class part of the plan, built at option by size by channel and door archetype rather than inherited from a spreadsheet, closes both leaks at once. The sizes that sell get depth where they sell. The sizes that do not stop arriving in volume. New styles inherit a real curve instead of the category mean, and when a hero size starts drifting mid-season the model catches it and reallocates before the buyer sees it on a report.
None of this asks the merchant to build curves by hand, which is exactly why the blanket curve survived this long. The platform derives the curve from demand, layers the door archetype and the style tilt on top, inherits a real curve for newness, and presents the result to check. The merchant approves the set in one pass, and the curve keeps moving with sell-through instead of freezing on the day it shipped.
You already bought the units. The only question is whether they are in the door, and the size, where someone will pay full price for them.
Get the curve right and the markdown report gets shorter, availability climbs, and you did it with the buy you already placed. The units never changed. Only where they landed did, and that turned out to be worth a fortune you were quietly giving away every season. The reason it stayed hidden this long is that it never showed up as a size problem. It showed up as markdown you assumed you had to take and full-price sales you never knew you missed, which is precisely why it is worth going back to the one number on the buy sheet nobody ever reopens.