Try-on is the only AR use case that touches your face, so the trust bar is the highest in all of ecommerce. Furniture goes in a room. Try-on goes on your body, which raises the bar for everything: accuracy, lighting, speed, and trust.
The categories that work: beauty, where shade matching on a real face solves a real problem; eyewear, where face mapping puts frames where they would actually sit; and increasingly jewelry and watches. Full-outfit fashion try-on is the hard one. Bodies vary more than faces, and the tech knows it.
Before committing, I would check four things. One, accuracy in bad lighting, because your customers do not live in photo studios. Two, color fidelity. If the rendered shade does not match the product photo, trust dies fast. Three, one-tap launch on the product page. A try-on that needs an app download loses most shoppers before it starts. Four, measurement, because try-on's real wins, fewer wrong-shade returns and bigger baskets from experimentation, only show up in the data.
You do not have to build this from scratch. The AR ecommerce plugin SaaS market includes try-on options, and a 3D asset production SaaS pipeline can supply the models underneath. Evaluate on trials with your actual products, not vendor demos with theirs. The famous reference remains Sephora's Virtual Artist, which earned its reputation the hard way: by working. Aim for that standard.
Also worth saying: try-on raises privacy questions faster than room placement does. Faces are personal. Be upfront about what the camera feed is used for and whether anything leaves the device. Trust is the whole product here.
Considering try-on for your store? I consult e-commerce brands on AR implementation at arcommerce.fyi.