Forecasting  •  9 Jul 2026  •  6 min read

Why inventory forecasting for apparel is a different problem

Most inventory tools were built for products that come in one flavour. Apparel comes in a grid of sizes and colours — and that single fact changes almost everything about how you should forecast demand and plan reorders.

If you sell apparel or footwear and you've ever tried to use a general-purpose inventory or forecasting tool, you've probably felt a quiet mismatch. The tool tells you a style is "healthy" and doesn't need reordering — but you know for a fact that the mediums and larges have been gone for two weeks and you're bleeding sales. The numbers say one thing; the shop floor says another. That gap isn't a bug in the tool. It's a sign the tool was built for a different kind of product.

The core difference: a product is not one thing

For most retail categories, a product is a single sellable unit. One SKU, one demand curve, one reorder decision. Forecasting is hard, but it's one hard problem per product.

Apparel doesn't work that way. A single style — say, a crew-neck tee — is really a grid: sizes down one axis (XS through XXL), colours across the other. A modest catalogue of 100 styles, each in 6 sizes and 4 colours, is already 2,400 individual sellable units. And crucially, each cell in that grid has its own demand pattern. Mediums sell faster than XXS. Black sells year-round while a seasonal colour spikes and fades. The "style" is an abstraction; the size-colour variant is where demand actually happens.

This is where general tools quietly fail — but the failure is subtler than it first looks. The problem isn't that they forecast at the style level. Forecasting at the style level is often the right place to start: a style has far more sales history behind it than any single size-colour SKU, so its demand signal is more stable and less noisy. A good apparel forecast borrows that statistical strength.

The failure is stopping there. A style-level number that never gets pushed down to the size decision leaves you exactly where you started — because the style looks fine on paper while its bestselling sizes are gone. The slow sizes (still in stock) mask the fast sizes (long gone), and the aggregate hides the very thing you need to act on.

The fix isn't to abandon aggregation — it's to aggregate for the stable signal, then disaggregate that forecast down to every size and colour using each variant's demand curve. You get the reliability of the style-level history and an answer at the grain where buying actually happens. Forecast high, decide low.

Size curves: how the style forecast becomes a size decision

The bridge between a style-level forecast and a per-size buy is the size curve — the proportion of demand each size represents. If your tee sells 5% XS, 15% S, 30% M, 30% L, 15% XL, 5% XXL, that curve is what turns "we'll sell about 200 next month" into a concrete per-size expectation. Disaggregating this way means the plan reflects how each size genuinely sells, while still resting on the fuller, steadier history of the style as a whole.

What "broken size runs" really cost

The classic apparel failure mode is the broken size run: a style that still has plenty of total inventory, but is missing its core sizes. On paper you have 200 units of the tee. In reality, 180 of them are XS and XXL, and every M, L, and XL is gone. A customer who wants a large can't buy — and a style that's "80% in stock" by unit count might be effectively unsellable to most of your customers.

Two things make this expensive:

• Lost sales are invisible. A stockout on your bestselling size doesn't show up as a negative number anywhere. The demand just quietly walks away. Your sales history then under-reports true demand for that size, which makes your next forecast too low, which makes you under-buy again. It compounds.
• The leftover sizes tie up cash and end in markdown. The XS and XXL that never sold aren't just dead weight — they're cash you can't reinvest in the sizes that actually move, and they're the units you'll eventually discount to clear.

Why stockouts poison your forecast

Here's a subtle trap specific to demand-driven categories. If you forecast future demand from past sales — which almost every method does — then weeks where you were out of stock look like weeks of low demand. They aren't. They're weeks where demand existed but couldn't be filled.

Feed those weeks into a forecast unchallenged and you systematically under-forecast your best sizes: the ones most likely to have stocked out. The fix is to recognise stockout periods and leave them out when learning demand — and the signal for that is already in your data: your own inventory history shows when a variant was at zero, so those weeks can be censored automatically rather than counted as genuine low-demand weeks.

The same logic applies to promotions. A big sale week isn't a normal demand week; if you treat it as one, you'll over-forecast the following period. Because only you know when you ran a promotion, the clean-demand approach lets you tag promotional weeks (per product) so they're set aside from the baseline. Between automatic stockout censoring and merchant-tagged promo weeks, the forecast learns from weeks that actually represent real, fillable demand.

Seasonality and size curves shift together

Apparel demand isn't just seasonal in volume — the shape of demand shifts too. A style might sell a different size mix in summer than in winter, or a colour might carry a season and then collapse. A forecast that assumes a fixed size curve across the year will misallocate inventory at exactly the moments that matter most.

What good apparel forecasting actually looks like

Putting it together, planning that respects how apparel really behaves does a few specific things:

• Forecasts at the style level, then disaggregates to size and colour — using the style's fuller history for a stable signal, then a demand curve to push that forecast down to every variant, so the plan lands at the grain where you actually buy.
• Learns from clean demand — recognising stockout periods (from your inventory history) so weeks you couldn't sell aren't mistaken for weeks of low demand, and letting you tag promotional weeks so a big sale doesn't distort the baseline.
• Detects broken size runs — flagging styles whose core sizes are short even when total inventory looks healthy, and telling you which sizes to top up.
• Translates the forecast into a per-size buy — so a reorder isn't one number split evenly across a size curve, but a gap-to-target calculation for each size line, with its own safety stock.

None of this is exotic maths. It's just planning that ends at the right unit of decision — the size-colour variant — instead of leaving the answer stuck at the style level where general tools tend to stop.