AR in furniture retail: what actually converts (and what doesn't)

Cylindo's 2024 Furniture E-commerce Report found that products with AR placement converted at 200% the rate of products without it. That number is real — but it assumes the AR implementation is good. Bad AR converts nothing, and there's more bad AR in furniture retail than good.

What AR actually does for a buyer

AR in furniture is not a novelty feature. It answers one specific question the buyer has before every large purchase: will this fit and look right in my actual space?

That question doesn't get answered by photography, measurements, or even a 3D web viewer — because all three require the buyer to do mental math about scale. AR removes the mental math. The piece appears in their room at real scale. Either it works or it doesn't.

The conversion data

The return rate reduction is the most important number here. Returns in furniture e-commerce run 15–25% — far higher than in any other product category. A 40% reduction in returns on AR-enabled products changes the unit economics of the category.

iOS vs Android: two different pipelines

AR on mobile is not one standard. Apple and Google use different runtimes that require different file formats:

This means every furniture SKU needs two files for full cross-platform AR: a GLB for Android and a USDZ for iOS. The conversion from GLB to USDZ can be done at upload time or on demand. Fenicher generates USDZ automatically from your GLB on upload — you don't manage two files.

What converts

• Accurate scale. The number-one reason AR sessions end in a purchase is that the scale is exactly right. Wrong scale (even 10% off) kills trust faster than no AR at all.
• Realistic materials under real lighting. Quick Look and Scene Viewer both use PBR rendering. A well-textured velvet sofa under the buyer's actual ambient light looks photorealistic. A flat-colored model looks like a toy.
• Floor anchoring. The model must anchor to the floor plane, not float. AR sessions that have the model floating fail immediately.
• Fast load. AR sessions under 3 seconds to USDZ download → Quick Look launch convert at 2× the rate of sessions over 8 seconds.

What doesn't

• Wrong file dimensions. If your model is 90 cm wide but the source file has it as 9 m (a common export unit error), the buyer places a sofa the size of a building. Session ends in 2 seconds.
• No shadow. Without a contact shadow, objects appear to float. Quick Look can render a default shadow — use it.
• Too many polygons. iOS Quick Look will refuse to load USDZ files over ~150 MB. If your AR model is a 300 MB file, it simply won't open. Target under 50 MB USDZ, under 20 MB GLB.
• Desktop-only flow. 78% of AR sessions on furniture sites happen on mobile. If the AR button only appears on the product page and your product page doesn't load fast on 4G, you've lost the session before AR even starts.

Getting the implementation right

The minimal AR setup that converts:

1. One accurate GLB per SKU, dimensions verified against physical sample.
2. USDZ generated from that GLB (Fenicher does this automatically).
3. An AR button on the product page that detects device and serves the right file.
4. File size under 20 MB GLB / 50 MB USDZ — optimize textures before upload.
5. Contact shadow enabled in the AR scene config.

We were skeptical of the Cylindo numbers. Then we ran it ourselves: 190% conversion lift on the two SKUs we added AR to first. We added AR to the full catalog the next week.