Reorder on units sold and you get the size curve wrong twice over: you buy more of the size that comes back most, and the coming back is not free.
This is not a study. The scan of 298 catalogues was a measurement and it is reported as one. This page is arithmetic over somebody else’s cost estimates, applied to an illustrative size curve, and every one of those words is doing work. Read the last two sections before you quote anything here.
A garment that ships and comes back is usually filed as a wash — the sale reverses, you are back where you started. You are not back where you started. You paid to ship it out, paid to ship it in, paid somebody to open it, inspect it, decide whether it can be sold again, re-fold it, re-bag it, re-ticket it and put it away. Some fraction of the time it cannot be sold again at full price and the markdown is the cost.
So a sold-and-returned unit is not zero. It is negative. And because return rates are not flat across a size run — the middle sizes come back most, because they are the sizes people buy hoping — the negative lands unevenly on exactly the sizes the gross numbers tell you to buy more of.
Nobody agrees, and the disagreement is the honest headline. The anchors that get cited in this industry are these:
| Source | Figure | Expressed as |
|---|---|---|
| Pitney Bowes BOXpoll | 21% | of order value |
| Optoro | 27% | of purchase price |
| Coresight Research with Optoro | 66% | of product price, apparel-specific |
| Modelled all-in processing, soft goods | $25–35 | per item |
A range from 21% to 66% of the item price is not a number, it is an argument. It varies with who ships, who pays for shipping, whether the item is graded and resold or liquidated, and how much of the labour is already sunk. I have not read the primary reports. These reached me through secondary aggregation, and I am telling you that rather than laundering them into a confident figure.
For the arithmetic below I use $30 per returned garment, the middle of the modelled per-item range. It is a round number chosen for legibility, not a finding.
A four-size run. Volumes and return rates are illustrative — the shape is the real part, because mid-size return rates running well above the ends is the ordinary pattern in apparel, not an unusual one.
| Size | Sold | Returned | Rate | Kept | Return cost @ $30 |
|---|---|---|---|---|---|
| S | 120 | 18 | 15% | 102 | $540 |
| M | 240 | 84 | 35% | 156 | $2,520 |
| L | 210 | 34 | 16% | 176 | $1,020 |
| XL | 95 | 14 | 15% | 81 | $420 |
Read the columns in order and the story changes twice.
Units sold says M, by 30. It is the bestseller, it is what the forecast is built on, and it is what gets reordered.
Units kept says L, by 20. The peak moved. M did not sell more than L in any sense that survives the customer changing their mind — it sold more attempts.
Return cost says L by a further $1,500. M generated 84 returns against L’s 34. At $30 each that is $2,520 against $1,020, and none of it appears anywhere in a units-sold report.
This is the part that matters, and it is why the page is worth writing despite the cost figure being so uncertain.
The ranking flips on units kept alone. L beats M by 20 units before a single dollar of return cost is counted. The cost figure does not decide which size to reorder — it only tells you how much the wrong answer costs.
So the conclusion survives the entire 21%–66% spread, and it survives the cost being zero. If returns were free, M would still be the wrong reorder. The $30 is not load-bearing; it is the size of the bill, not the direction of the error.
That is a deliberately weak claim and it is weak on purpose. A strong claim resting on a number between 21% and 66% would not be worth making.
It is not a measurement of your store, or of any store. I did not observe these volumes; I wrote them down to illustrate a shape. The scan on the other page counted real storefronts and says exactly what it counted and what it got wrong. This page counts nothing.
It is also not a claim that mid sizes always return most. It is the common pattern and it is the one the app is built to find in your own numbers — and the reason to look at your own numbers is precisely that the common pattern is not a law. A brand whose fit runs small will have this the other way round, and a units-sold reorder will be wrong in the opposite direction, which is still wrong.
Finally: I have not measured any brand’s returns. Sizecurve reads what Shopify exposes — orders and refunds over the last 60 days — and does this arithmetic on real numbers instead of illustrative ones. That is the whole product. This page is the reason it exists, not evidence that it works.
Your bestselling size is the size most people tried. Reorder on the one they kept.
Split your own reorder on what was kept: type what each size sold, what came back and what is on the shelf, and it gives the order per size. Free, no login, and nothing you type leaves your browser.