Virtual Try-On vs Size Recommendation: Returns Compared

If you run an apparel e-commerce operation, two technologies keep appearing in your vendor inbox: virtual try-on and size recommendation. Both promise fewer returns. Both cite impressive numbers. They are not the same tool solving the same problem, and choosing the wrong one — or deploying both without understanding the difference — costs money in integration hours before you see a single returned parcel less. This comparison cuts through the pitch decks.

Key takeaways

Virtual try-on primarily reduces returns driven by aesthetic mismatch; size recommendation targets fit-driven returns — the two causes are distinct and often require different fixes.
Published vendor data and independent research consistently show size recommendation delivering a larger average reduction in return rates, because fit is the dominant return reason in apparel.
Virtual try-on tends to lift conversion rates more measurably than size recommendation, because it shortens the imagination gap at the point of purchase.
Implementation complexity and ongoing data requirements differ significantly: size recommendation needs body-measurement data; virtual try-on needs high-quality garment assets.
The strongest operators use both, but if you can only start with one, your dominant return reason should decide which.

What problem does each technology actually solve?

Before comparing numbers, it helps to be precise about root causes. Industry practitioners and brands we speak to consistently report that apparel returns cluster around two distinct failure modes.

Aesthetic mismatch — the item looks different on the customer's body than it did on a model or a flat product shot. Colour, drape, proportion: things that photography abstracts away.

Fit failure — the item is the wrong size, or the size is right but the cut doesn't suit the customer's body shape. This is the larger category. Surveys of online shoppers in major markets routinely put fit as the leading return reason, ahead of quality issues or changed minds.

Virtual try-on addresses the first problem. Size recommendation addresses the second. A brand with a fit problem that deploys only virtual try-on will be disappointed; a brand with a styling-confidence problem that deploys only a size engine will see modest gains.

How does virtual try-on work, and what does the data say?

The mechanics

Modern virtual try-on uses generative AI to drape a garment image onto a photo of the shopper — either a model photo the shopper selects, or a selfie they upload. The output is a photorealistic composite showing how the item would look on a body close to theirs. The technology has matured considerably: leading implementations handle fabric texture, shadow and body occlusion well enough to be commercially useful.

Aiuta is one of the companies building in this space, offering a virtual try-on SDK that retailers embed in their product pages. Their published positioning emphasises the conversion-lift story: shoppers who interact with a try-on feature are more likely to add to cart and less likely to abandon.

What the numbers show

Conversion lift figures cited by virtual try-on vendors typically range from 20% to over 40% for shoppers who engage with the feature — though these figures reflect engaged users, not site-wide averages, so treat them as directional rather than guaranteed outcomes. Return rate reductions attributed specifically to virtual try-on tend to be more modest: figures in the 5–15% range appear in published case studies, with the strongest results in categories where visual fit is the primary concern — dresses, outerwear, statement pieces.

The honest limitation: virtual try-on does not tell a shopper whether a medium will fit their waist. It shows them how a medium looks on a body similar to theirs. If the underlying size is wrong, the try-on experience does not catch that.

Pros and cons

Pros - Measurable conversion lift, particularly for new or hesitant shoppers - Reduces aesthetic-mismatch returns in visually complex categories - Differentiating on-site experience that supports brand positioning - Does not require the shopper to input body measurements

Cons - Requires high-quality garment imagery or 3D assets; poor inputs produce poor outputs - Does not address fit-driven returns — the larger return category - Engaged-user metrics can flatter overall impact - Ongoing asset production cost as catalogue grows

How does size recommendation work, and what does the data say?

The mechanics

Size recommendation engines take some combination of the shopper's body measurements (self-reported or inferred), their purchase and return history, and the brand's own size chart and grading data, then output a recommended size. More sophisticated systems model body shape, not just circumference, and account for the fact that a brand's size 38 and a competitor's size 38 are not the same garment.

True Fit operates one of the larger networks in this space, aggregating anonymised purchase and return signals across a broad set of brand partners to improve recommendation accuracy beyond what any single brand's data could support. Bold Metrics approaches the problem from the body-data side, using AI to predict body measurements from a short questionnaire, then mapping those to brand-specific size charts.

What the numbers show

Size recommendation has a longer published track record in return-rate reduction, and the figures are generally larger than those for virtual try-on. Reductions of 20–40% in fit-driven returns appear across multiple published case studies from vendors in this category. Because fit is the dominant return reason, even a partial reduction in fit failures moves the overall return rate meaningfully.

Conversion lift from size recommendation is real but typically smaller than from virtual try-on — shoppers who receive a confident size recommendation are more likely to complete a purchase, but the experience is less visually engaging than seeing yourself in the garment.

Pros and cons

Pros - Targets the largest single cause of apparel returns - Network effects improve accuracy as transaction data accumulates - Works across all product categories, including basics where try-on adds little - Relatively lightweight front-end integration

Cons - Accuracy depends on data quality — sparse SKU data or inconsistent size charts undermine recommendations - Requires shopper input (measurements or history) that some users resist providing - Cold-start problem: new brands or new SKUs have limited signal - Does not address aesthetic-mismatch returns

Side-by-side comparison

	Virtual try-on	Size recommendation
What it is	AI-generated image of the garment on the shopper's body type	Data-driven size or fit recommendation at point of purchase
Best for	Reducing aesthetic-mismatch returns; lifting conversion in visually complex categories	Reducing fit-driven returns; improving confidence in basics and everyday apparel
Primary metric moved	Conversion rate	Return rate
Typical return-rate impact	5–15% reduction (published case studies)	20–40% reduction in fit-driven returns (published case studies)
Key input requirement	High-quality garment imagery	Body measurements or purchase/return history
Implementation complexity	Medium — asset pipeline is the main constraint	Medium — data integration and size chart quality are the main constraints
Cold-start sensitivity	Low — works from day one with good assets	High — accuracy improves with transaction volume
Limits	Does not catch size errors	Does not address how the item looks on the body

What can go wrong with each?

Virtual try-on failure modes

The most common disappointment is deploying the feature against a catalogue with inconsistent or low-resolution product photography. The AI composite is only as good as the garment asset it drapes. Brands that invest in try-on before investing in asset quality see weak results and blame the technology. The second failure mode is measuring only engaged-user metrics and reporting them as site-wide impact — a mistake that inflates perceived ROI and leads to over-investment.

Size recommendation failure modes

Inconsistent size charts across seasons or sub-brands are the most common culprit. If the engine recommends a size based on a chart that the production team has quietly adjusted, the recommendation is wrong before the shopper even clicks. The second failure mode is over-relying on self-reported measurements: shoppers systematically mis-report height and weight, and engines that don't correct for this bias produce recommendations that feel unreliable.

Who is each technology for?

Virtual try-on is the stronger first investment if: - Your return analysis shows aesthetic or styling reasons as a top driver - You sell in categories where visual fit is complex — dresses, coats, tailoring, occasionwear - Conversion rate is your primary problem, not return rate - You already have a strong product imagery pipeline

Size recommendation is the stronger first investment if: - Fit is your dominant stated return reason (it usually is) - You sell basics, essentials or categories where the look is secondary to the fit - You have transaction history you can feed into a recommendation engine - Return logistics cost is your primary pain point

Both together make sense once you have validated one and have the operational capacity to maintain two data pipelines. The combination addresses the full range of return causes, and the conversion and return-rate benefits compound. Brands we speak to who have deployed both report that the sequencing matters: most found it easier to start with size recommendation (lower asset cost, faster return-rate impact) and add virtual try-on once the fit problem was under control.

For context on how 3D and digital product representation more broadly is evolving in fashion tech — a related infrastructure layer that affects the asset quality virtual try-on depends on — the Company Profile: Browzwear — 3D Sampling's Longest-Running Bet is worth reading alongside this comparison.

If your interest is in the data and forecasting side of the stack — understanding demand signals that could inform size production decisions upstream — Company Profile: Heuritech — Trend Forecasting Built on Social Data covers a complementary angle.

Implementation: what you actually need to get started

For virtual try-on

Audit your product imagery. Consistent background, lighting and garment presentation are prerequisites. Flat lays perform worse than model shots for most engines.
Choose your integration point. Most SDKs embed on the product detail page; some support lookbook or social formats.
Define your measurement baseline. Capture return-reason data before launch so you can isolate aesthetic-mismatch returns in your post-launch analysis.
Set up A/B testing. Measure conversion and return rate for shoppers who engage with the feature versus those who don't, and separately measure site-wide impact to avoid engaged-user bias.
Plan asset production at scale. A pilot on 50 SKUs is easy; rolling out across a 2,000-SKU catalogue requires a production workflow.

For size recommendation

Audit your size charts. Inconsistent or outdated charts will undermine any engine. Standardise before you integrate.
Identify your data inputs. Do you have purchase and return history? Will you ask shoppers for measurements? Will you use a body-prediction model?
Choose your integration approach. Widget on the product page is standard; some brands integrate recommendations into the cart or checkout flow.
Define fit-driven returns in your data. You need a clean return-reason taxonomy to measure impact.
Plan for the cold-start period. New SKUs and new brands will have weaker recommendations initially; set stakeholder expectations accordingly.

Vogue Business has covered the commercial pressures driving investment in both categories, noting that return logistics costs have pushed fit technology from a nice-to-have to a line item in e-commerce P&Ls.

FAQ

Does virtual try-on actually reduce returns, or just increase conversions? Both, but unevenly. Published case studies show stronger conversion lift than return-rate reduction. Return rate improvements are real but typically smaller than those from size recommendation, because virtual try-on does not address fit errors — only aesthetic mismatch.

Which technology has a faster payback period? Size recommendation generally reaches payback faster because it targets the larger return category and has lower ongoing asset costs. Virtual try-on payback depends heavily on your imagery pipeline investment and your return-reason mix.

Can I use both virtual try-on and size recommendation together? Yes, and the combination addresses a broader range of return causes. Most operators who use both start with size recommendation and add virtual try-on once the fit problem is under control and asset infrastructure is in place.

How much body data does size recommendation need to work well? Accuracy improves with data volume. Engines that draw on network-level purchase and return signals — across multiple brands — can perform reasonably well even for newer brands. Pure self-reported measurement inputs carry inherent bias that better engines correct for.

Is virtual try-on worth it for basics like T-shirts and underwear? Generally not as a primary investment. In categories where fit is the dominant variable and the visual outcome is predictable, size recommendation delivers more return-rate impact per pound of investment. Virtual try-on adds more value in categories with complex drape, colour or silhouette.