Why fashion ecommerce return rates keep climbing

Ask a fashion ecommerce team to name their biggest cost line and most will say paid acquisition. Ask which one is most avoidable and the honest answer is returns. The difference matters. Acquisition buys you something. A return buys a second delivery, a series of warehouse touches, and a garment that may never sell at full price again.

Return rates in apparel have not drifted upward by accident. They are the predictable output of a category that asks people to buy something shaped for a body without letting them see it on that body.

How big the problem actually is

Apparel return rates run roughly 20–30% of orders, per the NRF and Happy Returns 2025 Retail Returns Landscape — several times what a typical electronics or homeware catalogue absorbs. Online-only fashion tends to sit at the top of that band. Retailers with shops usually sit lower, because some of the uncertainty gets resolved in a fitting room instead of in a parcel.

That range describes a category, not your catalogue. Occasionwear behaves differently from basics, and a brand with one house fit block behaves differently from a store carrying forty. Pull your own rate by category and by SKU before acting on anyone’s benchmark: the average hides the distribution, and the distribution is where the money is.

Fit is the root cause, not indecision

Roughly half of apparel returns are size or fit related — the consistent finding across industry aggregates from the NRF and Narvar. The rest splits between changed minds, items that did not match the photograph, damage and delivery problems. Only the fit half is really an information problem, and information problems are the ones a product team can do something about.

The information is missing for reasons that are structural rather than sloppy:

• A size 10 is not a size 10 across brands, and often not across seasons within one brand. Grading rules are commercial decisions, not standards.
• Two people with identical measurements wear the same garment differently. Shoulder slope, torso length and where a waist actually sits are not on any size chart.
• Fabric changes the answer. A rigid cotton and a twill with 4% elastane at the same nominal measurement do not fit the same way.
• The product photograph shows a model whose height, size and proportions are not the shopper’s, styled by someone who already knows how the garment behaves.

Faced with that, ordering two sizes is not bad behaviour. It is the cheapest way to buy certainty, and free returns make it nearly costless. Every bracketed order books a return leg at the moment of purchase.

What one returned garment actually costs

The visible cost of a return is the refund and the return label. The cost that shows up in the P&L is a stack of smaller items, most of which sit in somebody else’s budget — which is exactly why returns stay under-managed for so long.

Add the outbound leg of a replacement and the same garment can be shipped three times for one sale. No individual row is dramatic. Stacked and multiplied by a quarter of your orders, they usually outweigh whatever line item the team is currently arguing about.

Why size charts have not fixed it

Size charts fail because they push the work onto the shopper, in the wrong unit, at the wrong moment. They are a reference document being asked to do the job of a fitting room.

• They describe the garment. The shopper knows their clothes, not their body measurements.
• Measuring yourself accurately takes a tape, a mirror and a method. Most people estimate, and estimates skew.
• The chart sits behind a link below the fold, read after the decision rather than during it.
• Flat measurements say nothing about how the fabric behaves once it is on a moving body.
• “Runs small” in the reviews is a genuine workaround, but it only exists for popular SKUs and it is noisy.

None of this makes size charts useless — per-garment measurements are exactly what a fit engine needs. They just cannot carry the decision on their own.

What actually reduces the number

The interventions that move return rates are the ones that answer what the shopper is actually asking: how will this look on me, and which size do I order? Virtual try-on addresses the first directly, and a fit recommendation sitting next to the render addresses the second.

Treat that as a range of reported outcomes, not a promise: it comes from catalogues with different mixes, baselines and measurement methods. The effect tends to be largest where fit uncertainty is highest — tailoring, denim, footwear — because that is where the guessing was worst.

Where try-on sits in the funnel

On the product page, before add to cart, on the SKUs where fit uncertainty is highest. That placement matters: try-on offered after checkout is a nice experience with no commercial effect, and try-on buried in a separate tool the shopper has to find is a tool nobody finds.

The mechanism is straightforward. A shopper who can see the garment on a body like theirs, with a size recommendation attached, has less reason to bracket: one order gets placed instead of two. Rendering the whole outfit rather than one garment extends the same logic, because the decision people are making is usually about a look, not a single SKU.

How to start measuring it honestly

Return-rate work is easy to fool yourself about, because returns arrive weeks after the orders that caused them. A workable sequence:

1. 01Pull your baseline return rate by category, not just the store average, and separate fit-driven reason codes from everything else.
2. 02Rank SKUs by units returned rather than by return rate — a 40% rate on twelve units is a rounding error.
3. 03Enable try-on on the worst offenders first, and keep a comparable set switched off as a control.
4. 04Wait for the full return window to close before reading anything. Thirty to sixty days of orders, not thirty days of returns.
5. 05Compare cohorts of orders, not calendar periods, so a sale week or a seasonal shift does not do the work for you.

If the number moves you will know which SKUs moved it, which beats a headline percentage you cannot attribute. How the fit engine produces a size recommendation is the next piece of the same problem.