The Confidence Gap
How Virtual Try-On and Size Technology Are Reshaping Fashion E-Commerce
An industry report · July 2026
Executive summary
Fashion e-commerce carries a structural flaw it has never fully solved: shoppers cannot answer the two questions that matter most before paying — how will this look on me? and will it fit? The consequences are measurable. Online apparel is returned at two to four times the rate of in-store purchases, fit and sizing are the leading stated cause, and "bracketing" — ordering several sizes with the intention of returning most — has become normalised shopping behaviour. A maturing wave of technology now addresses this gap directly: AI-generated virtual try-on, augmented-reality fitting, size recommendation and fit prediction. Early published evidence points to materially higher conversion among shoppers who use these tools and meaningful reductions in fit-driven returns. This report examines the problem, the technology landscape, the evidence, and the honest limitations — and argues that visual try-on and fit assistance are tracing the same adoption curve that product reviews travelled a decade ago: from novelty, to advantage, to expectation.
1. The problem, quantified
The scale of e-commerce returns is no longer a niche operational concern. The National Retail Federation and Happy Returns estimate that US retail returns reached $849.9 billion in 2025, with online purchases returned at approximately 19.3% — roughly a fifth of everything sold online coming back [1].
Fashion sits at the top of that problem. Online apparel return rates consistently run between 20% and 40% depending on segment and season, with Coresight Research measuring 24.4% over a twelve-month period [2] and women's and fast fashion trending toward the upper end of the range [3]. The contrast with physical retail is stark: an ICSC survey found apparel bought online in the US was returned at 22%, against just 6.2% for the same category purchased in-store [4]. The difference between those two numbers is, in effect, the price of uncertainty — what it costs when a customer cannot see the garment on their own body or confirm their size before buying.
The stated reasons make the mechanism explicit. McKinsey & Company attributes 70% of fashion returns to poor fit or style [5]; Bold Metrics puts the fit-specific share at 53% of apparel returns [6]. Whichever figure one prefers, the conclusion is the same: the majority of returned garments have nothing wrong with them. The product was fine. The information was missing.
Shoppers have responded to that missing information rationally — by turning their bedrooms into fitting rooms at the retailer's expense. Sixty-three percent of consumers admit to ordering multiple sizes with the intention of returning those that do not fit [7], and among Gen Z shoppers roughly half bracket as standard practice when buying clothing and footwear online [8]. Bracketing is not customer misbehaviour; it is the predictable consequence of asking people to commit money before they can answer basic questions about fit and appearance. Every bracketed order inflates return volume, reverse-logistics cost and carbon footprint — while signalling precisely where the industry's information gap lies.
The commercial impact extends beyond logistics. Uncertainty suppresses conversion before a purchase ever happens: a shopper who cannot resolve "how will this look on me" does not always order three sizes — often they simply leave. The confidence gap is therefore paid for twice: once in abandoned carts, and again in returned parcels.
2. The technology landscape
Four related categories of technology have emerged to close this gap, and it is worth distinguishing them, because they answer different halves of the shopper's question.
AI generative try-on produces a photorealistic render of a specific shopper — typically from a single uploaded photo — wearing a specific garment. It answers how will this look on me directly and visually.
Augmented-reality try-on overlays products on a live camera feed, most effective for accessories, eyewear and footwear, where placement matters more than fabric drape.
Size recommendation and fit prediction tools use garment data, body measurements or purchase history to answer which size should I take. The simplest and most widespread member of this family remains the structured size chart — unglamorous, but still the reference point most shoppers reach for.
Virtual fitting and body-scanning approaches attempt full fit simulation from body models, the most technically ambitious and least widely deployed category.
None of these ideas is new. What changed is viability. Early virtual try-on, a staple of innovation showcases through the 2010s, was too slow, too expensive and — decisively — not convincing enough to help a purchase decision. Industry analysis in 2026 credits the recent generation of AI image models with finally clearing that bar: generation that once required dedicated infrastructure now happens in seconds at commodity cost, at a visual quality shoppers accept as a realistic preview [9]. As trade publication The Interline framed it this spring, the question for brands has shifted from whether the technology works to whether they have a clear vision for using it [9].
That cost collapse also changed who can deploy it. Virtual try-on was previously the preserve of enterprise retailers running bespoke pilots; it is now available to independent merchants as off-the-shelf storefront software, which is precisely the condition under which retail technologies historically move from differentiator to default.
3. The evidence it works
The published evidence base is early but increasingly difficult to dismiss.
The most striking figure comes from conversion. Reporting in Forbes in April 2026, virtual try-on platform DressX stated that shoppers who use try-on convert at roughly ten times the rate of shoppers on the same product pages who do not, and exhibit up to seven times higher retention [10]. The figure is platform-reported rather than independently audited, and self-selection plausibly accounts for part of it — engaged shoppers both try on and buy more. But even discounted heavily, the direction is consistent with the underlying mechanism: a shopper who has seen a garment on their own body has resolved the largest single source of hesitation between browsing and buying.
Consumer preference data points the same way. A June 2026 study of 1,000 German consumers by research firm Appinio found augmented and virtual try-on "pulling decisively ahead" among AI shopping features, with fit and returns assistance ranked the second most-desired AI function in online shopping, named by 36% of respondents [11]. Notably, this is demand-side evidence: shoppers are not merely tolerating fit technology, they are asking for it.
On returns, the metric merchants care most about, aggregated merchant data reported by fulfilment provider ShipNetwork indicates augmented-reality try-on tools reducing return rates by 30–40% where deployed [12]. Given that a majority of apparel returns are fit- or appearance-driven [5][6], a substantial reduction is mechanically plausible: the technology removes the guesswork that produces the return.
Two caveats belong in any honest reading of this evidence. First, much of it originates with parties who benefit from the technology's success — platforms, vendors, fulfilment providers — and independent, peer-reviewed measurement remains scarce. Second, effects will vary by category: draped garments, fitted tailoring and footwear present very different prediction problems. The evidence is best read as strong directional signal rather than settled science. But directionally, every published measure — conversion, preference, returns — points the same way.
4. Why "both together" beats either alone
A structural observation follows from the return-reason data, and it is one the market has been slow to internalise: the shopper's hesitation has two components, and they are answered by different tools.
Virtual try-on answers the aesthetic question — how does this look on me — with an image. It does not, by itself, tell a shopper whether to order the medium or the large. Size charts and fit recommenders answer the dimensional question — will it fit — with data. They do not show the shopper anything.
A merchant who deploys only try-on leaves the fit question open, and fit is the single largest stated cause of apparel returns [5]. A merchant who offers only a size chart leaves the appearance question open, and unresolved appearance expectations drive both abandonment and the "looked different than expected" return category [13]. The categories are complements, not substitutes — and the return data suggests the full benefit arrives only when both questions are answered at the same moment, on the same page, before the buying decision. A shopper who must hunt for sizing information after generating a try-on, or who checks a chart but still cannot picture the garment, has been left with exactly the residual uncertainty that produces bracketing.
This is, in miniature, the lesson the industry learned from product reviews: the value was never the widget, it was the removal of a specific hesitation at the point of decision. Fragmenting the answer across pages or tools reintroduces the hesitation the technology exists to remove.
5. The adoption curve
Retail e-commerce has run this pattern before, and the precedents are instructive.
Product reviews were once a differentiator; by the mid-2010s their absence had become a conversion liability, and today no serious merchant launches without them. Free or transparent shipping followed the same trajectory, as did mobile-optimised checkout. In each case the pattern was identical: a capability proves it removes friction, early adopters gain an edge, consumer expectation quietly resets, and the capability stops being a choice.
The signals suggest try-on and fit technology have entered the middle of that curve. The technology has crossed its viability threshold [9]; demand-side research shows consumers actively ranking fit assistance among their most-wanted features [11]; business press has moved from covering try-on as novelty to describing it as part of the infrastructure of fashion commerce [10]; and deployment has shifted from enterprise pilots to self-serve merchant software. What has not yet happened is expectation lock-in — the point at which shoppers treat its absence as a defect. The distance between those two points is historically where the competitive advantage lives, and it is not a wide window.
6. The honest counterpoint
Three limitations deserve plain statement, because merchants evaluating this category will encounter all of them.
Realism has improved, not resolved. Generative try-on produces a persuasive preview, not a physical simulation. Fabric weight, drape on a specific body, and true colour under real light remain approximations, and merchants in precision-sensitive segments should calibrate shopper expectations accordingly — a preview presented as a preview builds confidence; a preview oversold as reality manufactures the very returns it should prevent.
Consumers remain ambivalent about synthetic imagery. The same Appinio study that found strong demand for try-on also found 64% of consumers prefer real human models in product imagery, with scepticism rising as AI-generated content becomes harder to distinguish [11]. The distinction that matters is between brand imagery — where authenticity expectations are high — and personal previews, where the shopper is the model and the image serves them alone. Merchants conflating the two risk misreading the demand signal in both directions.
Uploaded photos are sensitive data. Personal try-on requires the shopper's image, and in European markets that brings GDPR obligations and reasonable customer caution. How solutions handle storage, retention, deletion and — pointedly — whether shopper photos are used to train models, will increasingly separate acceptable implementations from unacceptable ones. Privacy handling is becoming a feature, not a footnote.
None of these limitations undermines the category's trajectory. They define the standard a serious implementation must meet.
7. Implications for merchants
For a fashion merchant weighing this category in 2026, the evidence supports a handful of vendor-neutral conclusions.
Treat the confidence gap as one problem with two halves, and evaluate solutions on whether both halves — appearance and fit — are answered at the moment of decision rather than scattered across the journey. Weigh the return-reason data from your own store before anything else: if fit and "looked different than expected" dominate your return codes, the addressable prize is large and measurable, and your own baseline becomes the honest test of any tool you deploy. Interrogate privacy handling with the same seriousness as pricing: where shopper photos go, how long they persist, and whether they train models are questions your customers will eventually ask you. Prefer implementations that preserve page performance and degrade gracefully — a confidence tool that slows the product page trades one conversion problem for another. And calibrate expectations honestly in your own storefront copy: the technology is a powerful preview, and it performs best for merchants who present it as exactly that.
Finally, consider timing through the lens of section 5. Capabilities of this kind reward merchants who adopt while the capability still differentiates. Waiting for the technology to become universal is also waiting for the moment it stops being an advantage.
8. Conclusion
Fashion e-commerce built a trillion-dollar channel on an unresolved compromise: shoppers buy garments they have never seen on themselves, in sizes they cannot verify, and the industry absorbs the resulting returns as a cost of doing business. That compromise was never a law of nature — it was a technology gap, and the technology has now substantially arrived. The published evidence on conversion, consumer preference and return reduction is early and imperfect, but it is unidirectional. The reasonable debate in 2026 is no longer whether visual try-on and fit assistance help, but how quickly they become assumed — and the history of product reviews suggests that transition, once underway, completes faster than incumbent habits expect.
### About this report
This report was produced by AI Frame, which builds virtual try-on and size chart technology for Shopify stores (apps.shopify.com/aiframe-virtual-tryon). The body of the report synthesises publicly available industry data and third-party research; sources are listed below. No AI Frame proprietary data is included.
References
- National Retail Federation & Happy Returns (a UPS company), 2025 Consumer Returns in the Retail Industry — overall retail returns $849.9bn; e-commerce return rate 19.3% (2025).
- Coresight Research — online apparel return rate of 24.4% over a twelve-month measurement period.
- ShipNetwork, Ecommerce Return Rates by Industry (2026) — apparel category range 25–40%; seasonal spikes post-holiday.
- ICSC consumer survey (2024) — US apparel returns: 22% online vs 6.2% in-store.
- McKinsey & Company — 70% of fashion returns attributed to poor fit or style.
- Bold Metrics — 53% of apparel returns fit-specific.
- Synctrack industry analysis (2025) — 63% of consumers report ordering multiple sizes intending to return.
- Shopify, citing NRF data — approximately half of Gen Z shoppers bracket when buying clothing and footwear online.
- The Interline, Has AI Made Virtual Try-On Viable? (March 2026).
- Forbes (April 2026) — DressX-reported figures: ~10x conversion among try-on users; up to 7x retention.
- Appinio consumer study, Germany, n=1,000 (June 2026) — AR/virtual try-on leading desired AI features; fit & returns assistance ranked second at 36%; 64% preference for real human models in product imagery.
- ShipNetwork, aggregated merchant data (2026) — AR try-on tools associated with 30–40% return-rate reduction.
- Corso, The Most Common Ecommerce Return Reasons (2026) — "did not match description / looked different online" among leading return categories (~22%).
Methodology note: this report synthesises publicly available industry data, third-party research and trade reporting published 2024–2026. Figures are attributed to their original sources where identifiable; readers should note that several widely-cited industry statistics originate from commercial parties and have not been independently audited.