IndustryD2C Eyewear / Retail
DisciplinesWeb AR · Computer Vision · E-commerce
Duration11 weeks to first market
RegionIndia, UAE, KSA, Singapore, Malaysia, Indonesia

The problem

The brand had a strong brand and a beautiful PDP, but conversion stalled at 1.8% on mobile. Customer interviews surfaced two consistent friction points: “I can’t tell if it suits my face” and “I don’t know if it’ll fit comfortably”. Returns were a brutal 31% on prescription frames.

The solution

A web-AR try-on experience embedded inline on every PDP — no app install, no QR redirect, no friction. We used MediaPipe FaceMesh for tracking, custom GLB models for each frame, an automatic pupillary-distance estimator from a credit-card reference, and an AI face-shape classifier that re-ranked the catalogue per user.

What we built

Inline web-AR on the PDP

Tap “Try on”, the camera permission prompt fires, and the frame is on your face in under 2 seconds. No app, no separate URL, no “tap here to continue”.

3D models from photogrammetry

Each frame scanned at the warehouse with a turntable rig. PBR materials, accurate hinge offsets, and a strict 1.6MB GLB budget per frame.

Auto pupillary-distance estimation

Hold a credit card under your eyes — we get PD to ±1.5mm without an in-store visit. Halved prescription-fit returns.

AI face-shape recommender

A small CNN classifies oval / round / square / heart / oblong from MediaPipe landmarks and re-ranks the catalogue. 32% of buyers picked a frame they’d initially scrolled past.

Try-on share to WhatsApp

Capture a try-on snapshot, share to WhatsApp / Instagram. 14% of new orders trace back to a friend’s shared try-on.

Returns dropped 31%

Prescription returns specifically dropped from 31% to 21%. Return processing cost saved more than paid for the entire AR build in quarter one.

How it’s built

TrackingMediaPipe FaceMesh · WebGL · Three.js
3D PipelinePhotogrammetry rig · Blender · Draco compression · KTX2 textures
AI ClassifierPyTorch CNN (1.2MB) · ONNX runtime in-browser
StorefrontShopify Plus · headless Hydrogen · Vercel edge
AnalyticsMixpanel · Hotjar session replays · custom AR-specific funnel
CDNCloudflare R2 · per-region geo-routing
PrivacyOn-device only · zero face data leaves the phone · DPDP / GDPR
CompatibilityiOS 14+ Safari · Android Chrome 90+ · 96% of traffic supported

The numbers

2.4×
Conversion rate uplift on AR-enabled PDPs
-31%
Drop in prescription returns
+18%
Average order value (added a second pair)
14%
New orders from WhatsApp shared try-ons
11 wks
From kickoff to first market live
96%
Mobile traffic device coverage

“The AR build paid for itself in 9 weeks just on returns alone. The conversion lift was the upside we hadn’t even priced in.”

— CMO, D2C Eyewear (6 markets)

Have a project that looks like this?

If your engagement combines 3 or more disciplines, we’d like to hear about it. Tell us the constraint, the deadline, and the outcome that matters — we’ll come back with a scoped proposal.