The ROI of 3D/AR on a Product Page: The Real Math

Adding AR to a product page is usually pitched with a single eye-popping stat and a lot of hand-waving. That is not a business case. If you run ecommerce, the question is narrower and more useful: what does an AR "View in your room" button do to your revenue and returns, and does that beat what it costs to put there? Here is the actual math, the caveats that matter, and the one cost line most vendors hope you never scrutinize.

The two levers AR pulls

AR visualization moves money in two directions at once.

More completed purchases and bigger baskets. When a shopper can see a sofa in their living room or a tile on their own wall, uncertainty drops and they buy more, and more confidently. The published category numbers are strong: Shopify reports products with 3D/AR convert around 94% better than those without. Wayfair's "View in Room" reported a +92% conversion lift, +28% AOV, and -43% returns. IKEA Place reported +189% conversion and 98% size accuracy. These are the named companies' own results, not TARDIS's — but they establish the effect is real and large. Fewer returns. Returns are where AR quietly earns its keep, especially in furniture, where returns are a roughly $30B/year problem in the US alone. Industry studies put AR/3D returns reduction at up to 40%. A Snap/Publicis 2025 study (n=4,028) found 80% of AR shoppers feel more confident and 66% say they are less likely to return. Every avoided return saves the full reverse-logistics cost — shipping both ways, inspection, restocking, write-downs on damaged goods — which for bulky items often exceeds the item's margin.

A worked model (illustrative — plug in your own numbers)

Let's build the case with round numbers. These are illustrative, not a forecast. Use them as a template, then substitute your analytics.

Take one product category page cluster:

1. Monthly sessions: 100,000
2. Baseline conversion: 2.0%
3. Average order value: $400
4. Return rate: 20%
5. Contribution margin per order (before returns): $120
6. Cost per return (reverse logistics + write-down): $90

Baseline: 100,000 × 2.0% = 2,000 orders/month → $800,000 revenue.

Now apply AR — conservatively. The category studies show ~+94% conversion, but you should never bank the headline. Assume AR is engaged by a portion of shoppers and lifts conversion by a discounted +20% (roughly a fifth of the benchmark), and AOV by +10% (against the +28% Wayfair saw).

1. New conversion: 2.0% × 1.20 = 2.4% → 2,400 orders
2. New AOV: $400 × 1.10 = $440
3. New revenue: 2,400 × $440 = $1,056,000

That is +$256,000/month in top-line revenue, or about +$3.07M/year from this cluster. Returns savings. Assume AR cuts returns by a discounted -20% (versus the up-to-40% range), so return rate falls from 20% to 16%. On 2,400 orders:

1. Returns before AR-effect: 2,400 × 20% = 480
2. Returns after: 2,400 × 16% = 384
3. Avoided returns: 96/month × $90 = ~$8,640/month saved, plus the recovered margin on orders that now stick.

Even on deliberately deflated assumptions, the incremental-revenue line dwarfs the returns line — but returns savings are pure margin, so they often decide the ROI on thin-margin, high-return categories. The general formula: Incremental annual value = (Traffic × ΔConversion × New AOV × Margin%) + (Avoided returns × Cost per return)

Run it twice — once optimistic, once pessimistic — and make the decision on the pessimistic case.

Build vs. buy vs. the hidden cost line

Now the cost side, where most AR business cases quietly fall apart.

Building in-house means owning 3D asset generation, an AR web viewer, AR Quick Look / Scene Viewer integration, and ongoing maintenance across devices. For most retailers that is a multi-quarter engineering commitment competing with everything else on the roadmap. Rarely the right call unless AR is core IP. Buying is usually right — but what you buy determines whether the model above ever materializes at catalog scale. This is the line to interrogate.

Incumbent furniture-AR vendors (Cylindo, Threekit) produce photoreal 3D by modeling each SKU manually. That is real craft, and it is expensive and slow — commonly tens to hundreds of dollars per SKU and days of turnaround. On a 5,000-SKU catalog, manual 3D is a large upfront bill and a permanent bottleneck: every seasonal refresh re-triggers the cost. The predictable result is that AR gets deployed on a dozen hero products, and a dozen products cannot move a catalog-wide conversion number. Your ROI model assumed traffic across the category — manual 3D quietly shrinks that to a rounding error.

Surface vendors (Roomvo) do wall/floor coverings well but only surfaces — no furniture — so you are buying two tools, or leaving object categories uncovered.

Auto-generated assets change the unit economics. TARDIS generates the 3D model or surface texture directly from an existing product photo, so the marginal cost of adding a SKU collapses. That is the difference between "AR on 12 products" and "AR across the catalog" — and only the second version produces the traffic-wide lift the math depends on. One embed covers both object AR (furniture, decor → generated 3D placed via AR Quick Look / Scene Viewer) and surface AR (paint, tile, wallpaper → live in-camera retexture).

One honest caveat: web surface AR is preview-grade — colour and pattern are accurate, which is what drives the wallpaper/tile/paint decision, while lighting and edge precision sit below native. Native-grade surface fidelity comes via the premium iOS App Clip tier. For object AR, the placed 3D model runs through the platform-native AR viewers shoppers already trust.

How to de-risk the decision

You do not have to believe the benchmarks to justify a test. Instrument it:

1. Pick a bounded segment — one or two categories with high AOV and high return rates, where AR has the most to prove.
2. Measure engaged-session conversion, not just sitewide averages, so the signal isn't diluted by shoppers who never opened AR.
3. Track returns on AR-engaged orders over a full return window before you extrapolate.
4. Model catalog-wide rollout only on auto-generated economics — because manual per-SKU pricing rarely survives contact with a real catalog.

The headline stats tell you the ceiling. Your own segmented test tells you the floor. Decide on the floor, and let auto-generation decide whether the ceiling is even reachable. If you want to pressure-test this model against your own traffic, AOV, and return rates, we're happy to walk through it with you and show TARDIS running on a few of your SKUs. Book a demo — bring your numbers, and we'll do the math together.