Spatial and AR

Most AR projects fail the purpose test rather than the technical one. Placing a product in a room is genuinely useful when scale is the question a buyer cannot resolve from photographs — furniture, appliances, anything that has to fit. It is a novelty when the same information would be clearer as a dimension and a photograph. The distinction is worth being honest about early, because AR costs several times what the equivalent flat interface costs and is used once if it is not answering a real question.

Tracking quality decides whether it feels like magic or like a bug. Objects that drift, jitter, or sink through a floor destroy the illusion instantly, and the causes are environmental: a plain surface with no texture to track, low light, a reflective floor, a room the user is moving through too quickly. Designing around those — asking the user to move the device, preferring surfaces that track well, degrading gracefully when tracking is lost — is most of the craft.

Assets are the practical constraint everybody underestimates. A model exported from CAD arrives with a geometry budget suitable for manufacturing and entirely unsuitable for a phone, and a configurator with forty variants multiplies that by forty. The real work is a pipeline that reduces geometry, bakes lighting, compresses textures and produces device-appropriate variants automatically — because doing it by hand once is fine, and doing it by hand for every product update is not.