Virtual Try-On ROI: How to Calculate the Impact on Returns and Conversion

Virtual try-on ROI is the net annual gain from fewer returns and higher conversion, divided by what the tool costs you:

ROI = (returns saved + conversion gain + secondary gains − total cost) ÷ total cost

Every term in that formula is one you must supply from your own data, and the one most people get wrong is the first. A tool that cuts size-related returns does nothing to the returns caused by damage, delivery delay or changed minds, so applying a vendor's reduction percentage to your total return rate overstates the saving by a wide margin. This article covers each input, where to find it, a worked example with clearly hypothetical numbers, and the measurement mistakes that make the result meaningless in either direction.

1. Why Virtual Try-On ROI Is Harder to Measure Than It Looks

Total Returns vs. Size- and Fit-Related Returns

Only part of your return rate is addressable. Start with the benchmarks, then narrow.

NRF, with Happy Returns, put US returns at $849.9bn in 2025, which is 15.8% of total retail sales and 19.3% of online sales. Apparel runs higher: Coresight Research, with Alvanon, estimated the US online apparel return rate at 23.4% in 2025, worth $47.1bn against a $201.1bn online apparel and footwear market.

Published figures for the fit-related share are less reliable. The same Coresight report found that 69.8% of surveyed US shoppers who had returned online apparel in the past 12 months named size or fit as a reason. It does not mean 70% of returns are size-related. The figure counts shoppers who cited a reason, from 212 respondents surveyed in February and March 2026, and a shopper who returns six items for five different reasons appears in that figure once. Treat it as evidence that fit is the leading cause and get your own share from your reason codes before you build a model on it.

The Attribution Problem

Shoppers who use try-on are not a random sample of your shoppers. They are more engaged, further down the funnel, and more likely to buy and keep the item regardless of what the tool showed them. Compare try-on users against non-users and you will measure that self-selection, then book it as a product benefit.

You need a control group, and section 5 explains how to set one up.

Timing causes a second attribution problem. Try-on tends to launch alongside other changes: new photography, a reworked size guide, a seasonal campaign. Any change in return rate that quarter tends to get credited to whatever launched most recently. Write down what else shipped in the test window before you start, so you can rule those changes in or out later.

2. The Numbers You Need to Calculate Virtual Try-On ROI

You need six inputs, and five of them come from your own systems.

Return Rate, and the Fit-Related Share

Pull the return rate for the categories you would put try-on on, not the storewide figure. Then split returns by reason code to get the fit-related share. If your reason codes lump "doesn't fit" together with "didn't like it", that is worth fixing before you run any pilot, because it is also the metric you will judge the pilot by. Published ecommerce return rate statistics are a sanity check on your own numbers, never a replacement for them.

How to Calculate the True Cost of a Return

No credible public benchmark exists for this, and the per-return figures circulating online mostly trace back to pages with no stated methodology. Build yours from five components:

• Outbound shipping on the order that came back, if you paid it.

• Return shipping, if you fund returns.

• Handling and inspection: the warehouse labor to receive, check and restock one unit.

• Refurbishment and write-off: repackaging, steaming, and the share of returned goods that cannot be resold at full price.

• Customer service: the fraction of returns that generate a contact, times your cost per contact.

Add them for one representative order. That number, multiplied by returns avoided, is the first lever. The cost of returns is also the figure your CFO will challenge first, so document how you built it.

Run it twice if your mix varies: a $40 t-shirt and a $220 coat have different handling, different write-off risk and different shipping, and averaging them hides the categories where try-on pays off most.

Conversion Rate, AOV and Product Page Traffic

Baseline conversion on the product pages that would carry try-on, average order value, and the session volume those pages see. Without the traffic number you cannot size the conversion lever at all.

Try-On Adoption Rate

The share of shoppers on an enabled product page who actually use the feature. Most business cases leave this out, even though it affects every other figure: a tool that cuts fit returns by a third among users, with 10% adoption, moves your overall fit-return rate by about 3%.

Total Cost of the Tool

Subscription plus onboarding plus 3D catalog creation. Catalog creation is usually the largest cost in year one, and vendors tend to quote it last. The buyer's guide covers how pricing models differ and what to ask.

3. The Virtual Try-On ROI Formula

The formula has four levers, which you add up and then divide by total cost.

Lever 1: Savings From Fewer Returns

Returns avoided = orders with try-on × baseline return rate × fit-related share × reduction among users

Lever 1 = returns avoided × (cost per return + retained gross margin)

Two things are recovered when a return does not happen: the cost of processing it, and the gross margin on a sale that now stands. Most business cases count only the first and understate the lever by a factor of two or more.

Lever 2: Conversion Uplift

Lever 2 = enabled PDP sessions × conversion uplift (pp) × AOV × gross margin %

Measure the uplift rather than assuming it. A virtual try-on conversion rate improvement is plausible and category-dependent, and the only number worth putting in this formula is one your own holdout test produced.

Lever 3: Order Value and Fewer Bracketing Orders

Bracketing is the habit of ordering the same item in two or three sizes with the intention of sending most of them back. It inflates order value and return rate at once, so it reads as growth in one report and as a problem in another, and a store that watches only AOV can mistake the habit spreading for demand rising.

Set this lever to zero unless you track the behavior separately, by flagging orders that contain two sizes of one style. Where you do not, those orders show up as fewer returns, which lever 1 already counts, and adding both is the most common way these models double-count. Coresight notes that predictive sizing tools can reduce the behavior, but gives no figure, and there is little other public evidence.

Lever 4: Secondary Gains

The main secondary gain is imagery. If generated on-model imagery displaces part of your photoshoot or 3D modelling spend, that saving is attributable and belongs here. AI Fashion Models vs Real Photoshoots works through where it does and does not substitute. Customer lifetime value effects take longer to show up, so leave them out of a year-one case.

Putting It Together

ROI = (Lever 1 + Lever 2 + Lever 3 + Lever 4 − total cost) ÷ total cost

Payback (months) = total cost ÷ (total annual gain ÷ 12)

4. Worked Example: Virtual Try-On ROI for a $5M Shopify Fashion Store

Every number below is hypothetical, chosen to be round rather than typical. Substitute your own.

Lever 1. 10,000 orders × 25% return rate × 60% fit-related = 1,500 fit returns among try-on users. A 25% reduction avoids 375 returns. Each one saves $22 of processing plus $45 of retained margin ($100 AOV × 45%), so 375 × $67 = $25,125.

Lever 2. 1,000,000 sessions × 0.03pp = 300 additional orders. 300 × $100 × 45% = $13,500.

Lever 3. $0, counted inside lever 1.

Lever 4. Imagery displacing part of the season's photography: $20,000, included only because this store has a real line item to point at.

Total gain: $58,625. Year-one cost: $43,000.

Year one ROI = (58,625 − 43,000) ÷ 43,000 = 36%

Payback = 43,000 ÷ (58,625 ÷ 12) = 8.8 months

Year two ROI = (58,625 − 18,000) ÷ 18,000 = 226%

Year one is modest, because onboarding is front-loaded. Year two is the stronger argument to take to a CFO.

5. How to Measure Virtual Try-On ROI After Launch

A/B Tests and Holdout Groups

Hold back a randomized share of traffic from seeing try-on at all, and compare that group's return and conversion rates against the exposed group. A holdout is the only design that survives the self-selection problem in section 1. Comparing users to non-users within the exposed group does not.

Size the holdout before you start. On a 25% baseline return rate, detecting a two-point difference with any confidence takes thousands of orders per arm, which at 50,000 orders a year means a test running for a quarter, not a fortnight. Agree that timeline with whoever approved the budget, because a pilot stopped early for lack of signal wastes the budget without giving you an answer.

Tracking Returns by Reason Code

The pilot depends on your reason codes. If "too small" and "too large" are not separable from "changed my mind", the pilot cannot produce a readable result, and no downstream returns software will recover the distinction. Fix the reason codes before the pilot starts.

Two codes do most of the work: too small and too large. Keeping them apart tells you whether a style is graded wrong in one direction, which is useful to your merchandising team even if the try-on pilot fails.

Allowing for Return-Window Lag

A purchase made today can be returned 30 to 60 days from now, so a cohort's true return rate is not known until its window closes. Read results on closed cohorts only. Teams that check at day 14 see a flattering number and act on it. Build the lag into the reporting calendar: a pilot launched in March reports properly in June, and anyone expecting a number in April should be told so in advance.

Which Metrics to Put on a Dashboard

Adoption rate, fit-return rate for exposed versus held-out cohorts, conversion for both, and cost per avoided return. Anything that won't change a decision can go in an export, and analytics and insights covers what the event data supports.

Segment all four by category from the start. The aggregate figure hides the most useful result: denim and dresses behave differently from basics, and that split is what tells you where to extend coverage next.

6. Common Mistakes That Inflate or Hide ROI

• Using the total return rate instead of the fit-related share. This is the biggest source of inflation and can overstate the lever two or three times over.

• Comparing users with non-users. Measures self-selection and reports it as product effect.

• Judging too early. Reading a cohort before its return window closes.

• Counting multi-size orders twice. Once as fewer returns, again as order-value improvement.

• Omitting retained margin. This mistake runs the other way and hides real value: a return avoided keeps the whole sale, along with the processing cost.

• Treating onboarding as recurring. It distorts year one down and every year after up.

• Modelling at 100% adoption. A vendor's reduction figure applies to shoppers who used the tool. Multiply it by your real adoption rate before you share the model.

7. What Affects Your ROI the Most

Product category matters most. Denim, dresses, tailoring and outerwear carry high fit uncertainty and high AOV, so the same percentage reduction is worth far more there than on t-shirts.

Garment data accuracy comes next: a try-on built from real pattern files, grading rules and fabric specs shows the garment a shopper will actually receive, while one built from a scaled reference model misrepresents fit at the sizes where shoppers are least certain. From DXF Pattern Files to a True-to-Spec 3D Garment sets out what that requires.

Catalog coverage sets the ceiling on adoption. A feature live on 10% of SKUs teaches shoppers to stop looking for it, and adoption is the multiplier on every lever in section 3. Coverage is also the input you control most directly, since it is a function of how much garment data you hand over and how fast the vendor can process it.

Your returns policy also changes the picture. A store with free, no-questions returns has a higher baseline to cut into and therefore more to gain, while one that already charges for returns has suppressed the behavior by other means and will see a smaller delta. So don't compare your results with another brand's unless you know their returns policy.

Placement and page speed matter too. A try-on below the fold gets a fraction of the adoption of one beside the gallery, and a slow widget loses the shoppers it was bought to convert. Virtual Try-On for Shopify covers the performance side.

Vesto3D targets a 40% reduction in fit-related returns. That is a target and not a measured result, and it should be replaced with pilot data before anyone builds a business case on it.

8. Virtual Try-On vs. Other Ways to Cut Returns

Virtual try-on is one of four approaches to ecommerce returns, and they work as layers more than as alternatives. Most returns tools on the market fit into one of these rows.

A stricter policy is the only one that cuts returns without improving fit, which is why it reads well in a returns report and badly in a retention report. In practice, stores that cut returns usually combine several of these: accurate size data underneath, a recommendation layer, and visual confirmation on the categories where fit uncertainty is highest. Virtual try-on vs. size recommendation tools compares the middle two directly.

9. Frequently Asked Questions

What is a good ROI for virtual try-on?

There is no published benchmark worth quoting. Judge it against your own alternatives: if the net annual gain beats what the same budget would return in paid acquisition or photography, the case holds. Payback inside 12 months is a reasonable bar for year one, given onboarding is front-loaded.

How long does virtual try-on take to pay for itself?

It depends on catalog onboarding cost and adoption more than on subscription price. In the worked example above, payback lands near nine months, with year two running far higher once onboarding is behind you.

Does virtual try-on actually reduce returns?

Fit is the leading cause of apparel returns, and Coresight's 2026 survey found 69.8% of shoppers who returned apparel named size or fit. Whether a specific tool reduces them, and by how much, is a question only a holdout test on your own catalog answers. Vendor figures rarely state the baseline, the period or the categories.

Is virtual try-on worth it for small stores?

Sometimes. With a small catalog, onboarding cost is low and coverage is complete, so the question becomes whether return rate and AOV clear the subscription. High-fit-uncertainty categories such as denim, tailoring and occasionwear clear it sooner than loose-fit basics.

How do I calculate the cost of a return?

Add outbound shipping, return shipping, handling and inspection labor, refurbishment and write-off, and the customer-service cost per contacted return. Use one representative order, and document the assumptions, because this is the figure your finance team will test first.

10. Run the Numbers With Your Own Data

The formula depends on six inputs, and most of them sit in systems you already have. Send us your return rate, your fit-related share and your catalog size, and we will work through the model with you on a call. Book a free demo.