Size Curves and the SKU Explosion

Size planning is the least glamorous part of merchandise planning and among the most expensive to get wrong. This page sits under demand forecasting for apparel brands: how to derive a size curve properly, how to spot a broken one, and why fixing one costs a production run rather than a warehouse transfer.

What a size curve is, and why it must be derived rather than assumed

A size curve, also called a size ratio or size run, is the percentage split of a style's units across its sizes. It converts a style-level forecast into something a factory can cut. The size ratio apparel factories cut to and the size curve your planner argues about are the same number seen from two ends of the pipeline, and the size run for clothing production has to be fixed before the marker is made.

Two published starting ratios. Treat both as a first buy, not a plan.

A style forecast without a size curve is a total, not a forecast. You cannot cut 3,000 units; you cut 330 smalls, 660 mediums and so on, and that split determines whether the 3,000 sell at full price. The commonest failure is carrying a factory's default pack ratio into a second season. Never inherit a ratio you did not derive.

How to calculate a size curve from sell-through, correcting for stockouts

Here is the subtlety most guidance misses. Units sold is censored data. A size that sold out did not stop selling because demand ended; it stopped because you ran out. Derive next season's curve from raw sales and you re-buy the shortage that caused the problem.

The correction: use the sizes that did not stock out to build a cumulative sell curve, then estimate what share of demand each sold-out size had realised by the week it went out.

Prior season, one style-colour, 3,000 units bought on the 1:2:3:2:1 ratio, 12-week window. The cumulative curve from S, M and L reads 45% of demand realised by week 4, 61% by week 5, 81% by week 6.

Raw sales would have told you S 7%, M 19%, L 36%, XL 25%, 2XL 13%. The corrected split is S 6%, M 17%, L 32%, XL 27%, 2XL 18%: two points of XL and five of 2XL were invisible in the sales report. Re-run this at 30, 60 and 90 days every season — a size that sells out first is under-bought no matter what the aggregate says.

What moves a size curve: channel, region, category and fit block

One curve per brand is almost always wrong. Drivers and their direction; derive the magnitude from your own data, not a borrowed benchmark.

The SKU explosion: what one style actually costs you in forecast cells

A style is never one SKU. The SKU count per style is colours multiplied by sizes, so one style in 6 colours and 6 sizes is 36 SKUs, and 100 styles in 3 colours and 6 sizes is 1,800 forecast cells before any channel split.

The problem is the thinness of each cell. A style selling 3,000 units over 12 weeks across 36 SKUs averages 83 units per SKU, fewer than 7 a week. At that volume demand is intermittent: many cells sell zero in a week, where MAPE is undefined and time-series models stop working. Hence hierarchical forecasting — forecast where signal exists, disaggregate down the size curve, reconcile. The size curve is how a forecast becomes a number you can manufacture.

The failure that looks healthy: 88% sell-through and $17,395 gone

The style below hit its total unit forecast exactly and still destroyed a fifth of its gross margin.

3,000 units bought on the 1:2:3:2:1 ratio; true demand 3,000 units on the brand's derived curve of 6/17/32/27/18; retail $34; landed cost $6.08 as built in landed cost for apparel imports; full-price margin $27.92; leftovers cleared at 60% off.

Style sell-through is 2,640 ÷ 3,000 = 88%, which reads as a good season. Now count the money:

• Lost full-price margin on unmet demand: 360 x $27.92 = $10,051
• Margin forgone on markdowns: the 360 leftovers clear at $13.60 and earn $7.52 each instead of $27.92, so 360 x $20.40 = $7,344
• Total: $17,395 against a potential gross margin of 3,000 x $27.92 = $83,760 — 20.8% of the style's gross margin, lost on a forecast that was perfect in aggregate.

That understates it, because broken-size inventory suppresses further sales: a customer who cannot find their size rarely buys the next one up, and the remaining smalls sell more slowly once the style reads as picked over.

Broken size analysis: how to spot a size-curve problem in the data

Four checks, all of which run off data you already have.

1. Stockout dates by size. In a healthy curve they cluster. When XL goes out in week 6 and S is still selling in week 12, the ratio is wrong, not the style.

2. Size integrity against sell rate. Track the share of a style's sizes in stock each week against its weekly sell rate. When integrity falls the sell rate follows, and the gap is the cost of the broken curve.

3. Markdown concentration by size. Markdowns clustered at one end of the range is a size-curve signal; markdowns spread evenly is a demand signal. They call for opposite responses.

4. Forecast bias by size. A forecast can post an excellent WMAPE at style level while running over on the tails and under in the middle — visible only if you compute bias by size monthly, as covered on the demand forecasting pillar.

The sourcing join: size ratios are cut and packed before the goods ship

A size-curve error costs more than a colour-mix error because the ratio is physically committed at the factory, a quarter before it reaches your sales data.

The marker. Sizes are nested into the cutting marker in the planned ratio and the lay is cut in one pass. Marker efficiency on knits runs about 80-88%, so the ratio drives fabric yield as well as mix. Once cut, the ratio exists as fabric panels.

The ratio carton. Packing is specified on the tech pack's labelling and packaging page — folding, polybag spec, carton markings, ratio packing — so cartons arrive pre-built to the ratio: the complete tech pack guide. That is what ratio pack apparel means in practice: the split is sealed into the carton at origin, and a DC cannot re-pack its way out of it without opening every box.

The grade rules. The size set sample verifies grade rules before bulk, and final inspection under ANSI/ASQ Z1.4 (equivalent to ISO 2859-1) measures pieces per size against the points of measure. A size that grades wrong will not sell even if the ratio is right: AQL inspection for brand owners.

The consequence is blunt. You cannot fix a size curve from the warehouse — no transfer turns an S into an XL. Fixing it means a new production run on a full 85 to 125 day pipeline, which must clear a reorder point and an open-to-buy month first: reorder points when your lead time is 90 days and open-to-buy planning for apparel brands. Mid-season your only levers are pricing, bundling and channel reallocation. Terms above are in the Yarnstick glossary of apparel sourcing terms.

Frequently asked questions