How A Specialty E-commerce Store Reduced Returns By 31%
How does an e-commerce store use AI virtual try-on to reduce returns?
A specialty online apparel retailer integrated Drape Try-On into product detail pages via the Pro-plan REST API. Shoppers upload (or reuse) a single full-body photo and see themselves in any garment before checkout. Across a 6-month measurement period, return rates on participating SKUs dropped from 33% to 23% — a 31% relative reduction — with a parallel 4% lift in conversion. The mechanism is visual confirmation: shoppers self-filter garments that do not look right on them, reducing post-purchase regret.
- Returns on participating SKUs: 33% → 23% (−31% relative).
- Conversion lift: +4% on the same SKUs.
- Implementation: one PDP component + Pro-plan API in one week.
- Try-on usage rate: 38% of PDP sessions opted in.
- Mobile share of try-on usage: 64%.
Metrics
The challenge
The retailer’s blended apparel return rate was 33% — squarely in industry-average territory but a major drag on margin. Internal analysis showed 62% of returns were tagged “did not look like the photo” or “did not match the rest of my wardrobe” — visual-mismatch reasons that had nothing to do with sizing or quality.
The team had previously tried two solutions: a 3D avatar try-on tool that shoppers found awkward (under 4% usage), and an AR overlay app that worked only on the latest phones. Neither moved the return needle.
The approach
The team built a single “Try it on” button on every PDP. When clicked, a modal asked the shopper to upload or select a saved full-body photo. The product photo (already in the catalog) was sent with the shopper photo to Drape’s Pro-plan REST API. Result came back in roughly 12 seconds and rendered inline in the modal.
Three implementation details mattered for adoption. First, the shopper’s photo was saved to their account so subsequent try-ons required zero upload. Second, the result was inline (no page redirect). Third, a small “30-day returns” badge sat next to the result — framing the try-on as a confidence tool, not a sales pressure tool.
Implementation
A single React component handled the entire flow. Server-side, a Next.js API route called Drape with the SKU’s catalog image (already CDN-hosted) and the shopper’s stored reference photo (also CDN-hosted in Supabase Storage). End-to-end response was ≈14 seconds including network. The full integration shipped in one week.
For the first month the team A/B tested the feature on 50% of traffic to measure clean lift. After confirming the return reduction was statistically significant on Week 3, the feature was rolled out to all SKUs except shoes (where AI try-on is not currently supported).
Results
On participating SKUs, returns dropped from 33% to 23% — a 31% relative reduction — over a 6-month measurement period. Conversion rose 4% on the same SKUs. Average order value rose 5% as shoppers with confirmed-good-look items added complementary pieces.
The try-on opt-in rate stabilised at 38% of PDP sessions, with 64% of usage happening on mobile. The retailer’s monthly Drape API spend was approximately $480 against returns avoidance of roughly $11,000 — a 22× ROI on the integration.
Frequently asked questions
Does AI try-on predict actual sizing?+
Is the customer real?+
What was the engineering effort?+
Drape Editorial is the in-house research team behind Drape Try-On. We test virtual try-on models against real garment photography weekly and publish what we learn.