How to reduce returns in ecommerce apparel

Most apparel returns are fit returns, and a fit return is decided before the parcel ships. A shopper who cannot tell which size will fit either guesses or orders two sizes and sends one back, and both paths end at your receiving dock. So how to reduce returns in ecommerce is a question about what a shopper knows at the moment they choose a size, not about how fast a warehouse processes what comes back.

There are no return-rate results here, because ShirtMath has none: ShirtMath for Brands is pre-launch, and how ShirtMath sources measurements is published in the same spirit.

A size label names a position in a size run. It is not a measurement. Grading inside one block is tight: across the seven unisex tee styles whose manufacturer spec sheets sit in ShirtMath’s catalog, a Medium measures 20.0–20.5″ across the chest laid flat — half an inch of spread across seven brands.

The distance appears when a shopper carries a letter across cuts. The ladies’ Gildan G500L publishes 17.5″ at Medium, up to three inches narrower than those unisex Mediums, and 3.5″ narrower at Large. Both garments say M. Both are correctly labeled.

A shopper who knows they wear M has learned something true about the shirts they own, not about yours. Every product page asks them to make that transfer anyway, and given a free return policy, ordering both sizes is the rational answer — size bracketing: two picks, two packs, two units in transit, one return scheduled before the order was placed.

The cost of a return is not one number, and this page will not invent one: the split between transport, labor and lost margin depends on your unit economics, carrier rates, and how late in the season goods come back. The structure, though, is the same everywhere, and each stage spends money the original order never priced in.

Two of these outrank the rest, and neither is freight: per-unit inspection labor, which does not get cheaper as returns rise, and time — a returned unit is unsellable in transit, on the dock and in refurbishment, so seasonal stock re-enters below the price it left at. A bracketed order commits to both before it is even packed.

Two categories of software share the phrase, and they act at opposite ends of the order. Returns management platforms act after the fact: RMA portals, prepaid labels, automated refunds, exchanges offered in place of refunds. They make each return cheaper and better documented, and an exchange recovers revenue a refund does not. What they cannot do is prevent the picking, packing, shipping and inspecting of a unit that was never going to be kept — the layer covered in returns management.

Fit matching acts before the order exists. It answers the question the shopper is stuck on — which of these sizes is the shirt I already own — with numbers instead of a letter, so the second size never enters the cart. The two are complements, and the asymmetry is cost: processing scales with the number of returns, while a comparison runs on data you have already published. Making that data machine-usable is its own work, covered in size chart integration.

Mechanically it is subtraction, run before the buy button instead of after delivery.

The customer is not told which size to buy: they are shown that one candidate is half an inch tighter than a shirt they already like and the other an inch and a half roomier, a claim they can check with a tape. What subtraction cannot see is whatever the spec sheet omits: the Bella+Canvas 3001C and the Gildan G500 publish identical chest and length at every shared size and still do not wear alike, because taper, fabric weight and drape are nowhere on the sheet. Matching numbers are necessary, not sufficient — and the letters customers arrive with drift too, which vanity sizing documents and the t-shirt size chart quantifies.

ShirtMath for Brands is pre-launch and not generally available. There are no customers, no pilot partners and no return-rate results yet — the first integrations will exist to measure a number rather than assert one. It matches a shopper’s reference garment against your product measurements number by number, works from the measurements you already publish — no new photography or data format required to start — and ranks by fit math only, never by commercial weighting.

The binding constraint is published data, not modeling. In ShirtMath’s own consumer catalog of 677 rows, 601 carry both a chest and a length, which is why tops compare well. Bottoms mostly do not: the catalog holds 74 of them and 17 rows with a waist measurement. Garments whose merchants publish nothing usable are left out rather than given invented numbers. What a brand publishes sets the ceiling on what any fit engine can do.

The scanner that produces a shopper’s reference numbers is experimental and says so: it leaves a measurement blank rather than guessing, about half of readings land within an inch of a real tape, and sleeve opening and bicep are the least reliable. Two things are deliberately absent here: customer names, because pilot partners are not case studies, and a percentage, because the pilot exists so that such a number can be measured rather than asserted.