Size curve: how many of each size to buy
A size curve is the ratio in which you split a buy across S, M, L and the rest. Nearly every guide agrees on how to calculate it: look at your sales history. This guide explains why that method carries an error that compounds on itself, and what can be measured instead.
What a size curve is
It is the ratio in which you buy each size of the same style. It is written as a relationship: 1:2:2:1 means that for every unit of the first size you take two of the second, two of the third and one of the fourth.
When a store starts with no history, the standard advice is a bell curve: few extreme sizes, many middle ones. For five sizes it is usually 1:2:3:2:1. It is a reasonable starting point, but that is exactly what it is: an assumption about who your customers are, not a measurement.
The standard method: build it from your sales
As soon as you have history, every guide says the same thing: work out what percentage of last season's sales each size was, and split the next buy along those shares. If 35% was M, buy 35% M.
It beats a generic bell curve, because it uses your own store's data. And it is the right advice if that is the only data you have. The problem is what that data is not.
Why your sales history comes contaminated
Your sales are not your demand. Your sales are the intersection of what people wanted and what you had. Every time a size runs out, every following week records zero sales of that size, and that zero enters the calculation as if nobody had wanted it.
This is not a new observation: the industry's own best practice acknowledges it, and its solution is to discard data.
Build the curve from clean demand, using only weeks when every size was in stock and nothing was on markdown.Standard assortment planning advice in fashion retail
The error that compounds on itself
There is the loop. If XL runs out fast, your sales say XL sells poorly. You buy less XL next season. It runs out sooner. The data gets worse. And so on.
Meanwhile the opposite happens with the middle sizes: the person who wanted XL and did not find it takes an L, or an M. That sale is recorded as M demand. Your curve does not just underestimate XL, it also inflates M with buyers who were never M.
And the clean-weeks filter does not fix the worst case: the sizes you never bought. If you never stocked XXL, there is no week you can look at.
What can be measured instead of inferred
There is a moment when a shopper declares their size before the purchase has happened: when they ask which one is theirs. If your store has a size recommendation on the product page, every query is a declaration of demand, and that declaration does not depend on your inventory.
That allows two readings sales history cannot give. The first is the real demand curve, size by size and product by product. The second is more specific: how many times someone's ideal size was out of stock. That demand leaves no trace in any other system, because there was never a cart, a search or an order: nothing happened to record.
How Fittly does it
Fittly is a Shopify app that calculates each shopper's size from the real measurements of each garment. It records each person's ideal size, available or not, so the analytics measure demand and not availability.
The panel shows the curve by scale and by product, plus a second panel with the times the ideal size was out of stock and the closest one had to be offered. Each shopper counts once per product, with their latest size, so an indecisive visitor does not distort the count.
What to look at first
The most actionable number is not the curve, it is the second panel: on which products and how many times the ideal size was missing. That is concrete restocking, not a ratio adjustment.
Then it pays to look at the products that break the pattern. If one garment consistently comes back a size up from the rest of its scale, that garment runs small. It is usually the one generating returns without anyone knowing why.