Apparel e-commerce is grappling with a staggering $47.1 billion blind spot. Data from a Coresight Research report, backed by sizing tech enterprise Alvanon, reveals that U.S. returns for online apparel and footwear hit $47.1 billion in 2025 alone, marking a notable 23.4% return rate. Among shoppers who sent items back, nearly 70% pointed squarely to sizing and fit issues as the core culprit.

Back in 1939, the U.S. Department of Agriculture initiated measurements of American women to establish standard clothing sizes. Its 1941 findings, derived from nearly 15,000 subjects, laid the groundwork for the 1958 federal sizing standard—though older demographics and nonwhite participants were largely sidelined. Decades later, the 2002–2003 SizeUSA initiative utilized 3D body scanners on roughly 10,000 Americans, documenting major shifts in human proportions.

While measurement technology has evolved rapidly from simple tape measures to advanced 3D scanners and artificial intelligence, the manufacturing sector still anchors its production around traditional reference bodies. Dress forms and live fit models continue to dictate patterns and grading.

More than eighty years after the original USDA study, AI generates exceptionally precise digital replicas of our physiques. Yet the fundamental question remains: has the apparel industry revolutionized how garments are built, or has it simply gotten better at pushing shoppers toward predetermined sizes?

The Illusion of the Virtual Try-On

Generative AI has undeniably streamlined digital retail. Consumers can leverage tools like ChatGPT, Gemini, and various retail engines to instantly pull together silhouettes, styles, and price points.

Virtual try-on (VTO) software allows buyers to preview how clothes look on their digital selves before checking out. These algorithms seamlessly place garments onto user photographs, rendering the drape and proportions remarkably flattering.

However, the screen frequently diverges from physical reality. With three decades of product development experience, I still find myself ordering multiple sizes of a single garment, directly feeding the retail industry's margin-draining return loop.

A synthesized image may showcase a silhouette without properly factoring in fabric elasticity, armhole depth, waistline placement, rise, or how clothing reacts during actual human movement.

This does not diminish the commercial utility of AI visuals. As previously highlighted, Zelig’s styling tools enabled REVOLVE to boost conversions and average order values while cutting down returns, enhancing overall shopping confidence.

Even so, generating an attractive outfit visualization is an entirely different discipline than predicting true physical fit.

The central dilemma is no longer whether algorithms can enhance retail discovery—they clearly have—but whether they can inform customers how a piece will actually feel on their bodies.

Vestir, partnering with ETH Zurich on virtual fitting research, merges 3D reconstruction, physics-based fabric simulation, and generative AI. Founder Joanne Filbin notes that building a digital avatar is relatively straightforward, whereas accurately digitizing physical fabric remains a monumental hurdle.

“Gen AI can make something look real. But what these Gen AI solutions don’t have is the ability to solve the size and the fit,” Filbin explained.

While their research achieved high avatar and garment precision under controlled lab conditions, standard consumer photo-to-3D workflows face massive obstacles from unshielded smartphone cameras, ambient lighting, and random user angles.

Translating convincing computer graphics into dependable fit forecasts remains the ultimate hurdle for the sector.

Inside the Zalando Testing Lab

Zalando provides a fascinating look at the bleeding edge of virtual fitting rooms. During a 2023 trial encompassing 22 denim variants, users built 3D avatars and consulted heatmaps showing tight or loose zones across different sizes. Subsequent trials yielded return drops of up to 40%, while broader fit tools helped mitigate 8% of size-related returns in 2025.

Unlike third-party platforms relying on brand-supplied metrics, Zalando runs an in-house garment digitization hub. Led by Niclas Grabau, the Berlin Fitting Lab deploys a specialized industrial scanner fitted with 120 to 140 cameras to capture complex geometries, particularly the crotch area, generating data-rich digital blueprints.

When testing Zalando's virtual fitting room myself, I generated a body profile via smartphone and compared its stats against manual measurements. Most metrics aligned within 1 to 2 centimeters, though the thigh measurement registered slightly larger and no waist metric was provided.

Testing out Levi’s jeans, the system suggested a size 26x27—matching my typical Levi’s fit. The avatar mirrored the genuine looseness I experience around my hips and thighs. Pushing up to a size 32x32 clearly illustrated excessive fabric and length.

However, trying a size 23x30 exposed a stark reality. The interface displayed the waistband and fly completely closed, despite the fabric blend containing 85% cotton, 14% polyester, and 1% elastane. Knowing my own dimensions, zipping into a size 23 is physically impossible, yet the software projected me wearing it successfully.

Zalando’s technology excels at illustrating size variations, yet my trial highlighted a distinct ceiling: showing a garment on a digital twin is not the same as verifying that a human body can actually wear it.

Anatomy Versus Algorithmic Assumptions

Further testing across TrueToForm and Size Stream’s Formcut smartphone scanning apps revealed additional nuances. Drawing on my decades in technical design and sourcing, I measured myself manually to compare baseline metrics against the scans.

Variations of one to three inches emerged across the tools. TrueToForm co-founder Janice Tam pointed out that elements like posture, breathing patterns, tape tension, and anatomical definitions naturally alter outcomes. For example, a natural waistline often differs from where a shopper positions trouser waistbands.

Size Stream CEO Jay Rudman noted that manual tape measurements inherently fluctuate based on technique and placement, which is why Size Stream benchmarks against high-precision booth scanners rather than manual numbers.

TrueToForm translates body architecture into brand-specific sizing suggestions, while Formcut focuses on made-to-measure development. In my shopping trials, TrueToForm recommended a Small at Rainbow Shops and a size 6 at New York Dress—both larger than my normal baseline—reflecting how varied brand specs and inventory dynamics influence outputs.

Tam emphasized that optimal inventory matching does not automatically equate to genuine fit satisfaction, noting that the broader vision is to reshape how garments are engineered from the ground up.

Translating Data to the Factory Floor

“The scanning technology is actually probably the easiest part of the entire equation,” Rudman remarked, stressing that raw body metrics are just the starting point.

Factories must ultimately convert those numbers into patterns, construction rules, and ease allowances. Size Stream, owned by major manufacturer TAL Apparel, bridges digital scanning with real-world production requirements.

“Body measurements and garment measurements are totally different,” Rudman noted, emphasizing that fabric properties and tech packs dictate the final geometry.

By combining scanning configurations with manufacturing specs, Size Stream maintains low remake rates for custom apparel, pointing to a massive opportunity to upgrade standard sizing models industry-wide.

Upstream Data Pipelines at DXL

Destination XL Group (DXL) has partnered with Size Stream since 2019 to utilize FiTMAP, a body-scanning system helping men navigate cross-brand sizing disparities.

FiTMAP evaluates 243 discrete body metrics to deliver tailored recommendations. DXL's fiscal 2025 reports indicate strong engagement across numerous brands and tens of thousands of completed customer scans, yielding improvements in conversion rates, order values, and reduced return frequencies.