Fashion e-commerce has a returns problem that has resisted every attempt to solve it. Return rates on apparel purchased online run between 25 and 40 percent across most markets, roughly double the rate for non-fashion categories. The primary driver, by far, is fit: the item doesn't look or feel the way the customer expected.

Virtual try-on technology has been pitched as the solution for years. The early versions - augmented reality mirrors in stores, basic photo overlays in apps - were novelties that didn't move the needle. The technology wasn't accurate enough. A virtual try-on that makes every garment look flattering but doesn't reflect how it will actually drape on a specific body shape is worse than useless - it increases returns, because the customer buys with false confidence.

The latest generation, powered by diffusion models and 3D body estimation, is different. And the early return-rate data from retailers running controlled tests is the first genuinely encouraging signal the category has produced.

What changed

Two technical shifts converged in 2024 and 2025:

Accurate body estimation from a single photo. The AI models that estimate a 3D body shape from a standard smartphone photo have gotten dramatically better. Earlier systems required multiple photos, specific poses, or even body scanning hardware. The current generation - built on architectures refined by Meta, Google, and several startups - can produce a reasonably accurate 3D mesh from a single front-facing photo. "Reasonably accurate" means within one to two centimeters on key measurements for most body types. That's close enough to predict fit for most garment categories.

Realistic garment draping simulation. The rendering of how a specific garment hangs on a specific body has improved from "uncanny valley" to "credible." This is partly a compute story - the models are bigger and running on better hardware - and partly a training data story. The companies leading in this space have built large datasets of garments photographed on diverse body types, which the models use to learn how different fabrics and cuts behave.

The combination means that a customer can now upload a photo, select a garment, and see a rendering that is close enough to reality to inform a size decision. Not perfect. Not a replacement for trying it on. But meaningfully better than looking at a photo on a model and guessing.

The return-rate data

Several retailers and platforms have shared early data from controlled rollouts, mostly under NDA but with enough directional information to draw conclusions.

A European fashion e-commerce platform running an A/B test on its try-on feature across a subset of categories (dresses, outerwear, and denim) reported a return rate reduction of 8 to 12 percentage points for customers who used the feature versus a control group that didn't have access. The feature had roughly 15 percent adoption among eligible sessions - not every customer uses it, but those who do return significantly less.

A UK-based retailer testing a similar system on its own product reported a smaller but still meaningful reduction: 5 to 7 percentage points on the categories where the technology was deployed.

These are early numbers, with caveats - the customers who choose to use a try-on feature may be more engaged or more deliberate shoppers to begin with, which inflates the apparent effect. The platforms running the tests are aware of this selection bias and are attempting to control for it. Even with conservative adjustments, the signal is positive.

The economics

A return in fashion e-commerce costs the retailer between 10 and 20 euros in logistics, processing, and restocking - more if the garment is damaged or unsellable after return. On a category with a 35 percent return rate, that cost represents a substantial drag on margin.

A try-on feature that reduces returns by even 5 percentage points across the assortment would, at most fashion e-commerce retailers, pay for itself many times over. The cost of running the AI inference - currently a few cents per try-on session - is trivial relative to the cost of a prevented return.

This is why the investment is accelerating. Zalando, ASOS, and several other major fashion e-commerce players have either built or acquired virtual try-on capabilities in the last 18 months. The technology providers - companies like Zeekit (acquired by Walmart), Revery.ai, Vue.ai, and Google's own try-on features in Shopping - are all pushing hard on accuracy and integration.

What's still missing

The technology works best on structured garments - jackets, coats, dresses with defined silhouettes - where fit is relatively predictable. It's less reliable on knits, draped fabrics, and items where the subjective "feel" of the fabric matters as much as the dimensional fit. A customer might get the right size recommendation from a virtual try-on but still return the item because the fabric hand wasn't what they expected.

Color accuracy remains a challenge. Screens vary, and the rendering of fabric color in a virtual try-on doesn't always match the physical product. This is a solvable problem but one that requires calibration work that most platforms haven't fully done.

And there's the adoption question. Even the best-performing tests show try-on usage in the 10 to 20 percent range of eligible sessions. Getting that number higher - through better UX, faster loading, and building consumer trust in the accuracy - is the next challenge.

Where this goes

Virtual try-on is crossing the threshold from experiment to infrastructure. The accuracy is good enough to reduce returns. The economics are favorable. The major platforms are investing. Within two to three years, we expect virtual try-on to be a standard feature on any serious fashion e-commerce site, the way size charts are today.

The implications for physical retail are worth watching. If online try-on gets good enough to genuinely replicate the fit-assessment function of a fitting room, one of the remaining structural advantages of physical fashion stores erodes. That doesn't mean stores go away - there are plenty of other reasons to visit a fashion store - but it does shift the balance further toward online for routine purchases where fit was the primary reason to buy in person.