Outsource room preview or build in-house? A furniture CTO checklist

Every mid-size furniture brand hits the same fork: build an in-house room composite model, or embed a vendor widget. We are obviously biased — we sell the widget — but here is the honest CTO math we see merchants use before they outsource.

Build in-house: real cost lines

• CV/ML engineer — six figures annually for maintenance alone.
• GPU inference — spiky cost tied to campaign traffic, not flat SaaS.
• Shopify theme work — every theme revision re-tests the embed.
• Edge cases — mirrors, rugs, odd camera angles; years of shopper photos you did not plan for.

Outsource: what you are buying

A jsDelivr-hosted snippet, billing at usage (10p/preview on our plan), and a team that already handled last month’s iOS Safari upload bug. Time-to-first-preview on a staging PDP: about five minutes after signup.

When build still makes sense

• You already employ a production ML team for other product lines.
• You need proprietary models trained on your proprietary room dataset.
• Regulatory requirements mandate on-prem inference.

For everyone else selling beds and sofas on Shopify, outsource is usually the faster path to measurable PDP lift.

Hybrid path

Use room preview from photos on PDPs today; commission 3D studio assets only for hero configurators or AR exports later. Do not block preview on a two-year mesh library project.

Model outsource vs photography vs 3D rendering on /comparison.