AR has spent a decade as next year's demo. The headset, the app download, the show-floor walkthrough — impressive, then unused.
The useful question was never whether the overlay looks modern. It is whether a specific uncertainty is killing the purchase: Will this fit my face? Will this sofa dominate the room? Will this rack clear the beam? Will this finish look like the swatch on a phone in daylight?
If that uncertainty is not a real driver of abandonment or returns in your category, AR is decoration. If it is, the viewer is the small part. The product record has to be accurate enough that what the customer sees is what they will receive.
Drop the year from the argument. WebAR running in a browser, without an install, is what made the honest cases deployable. Social AR and full VR shopping still sit with brands that can fund the creative machine around them.
Pay for Uncertainty, Not for Novelty
Apparel, eyewear, and beauty return because fit and appearance were guessed. Furniture and home return because scale and context were guessed. Configured and built-to-order goods stall because the customer is asked to imagine a combination. Industrial equipment stalls when a drawing in a PDF cannot be trusted in a real bay.
Those are spatial problems. A better paragraph on the PDP will not solve them. Neither will a campaign. A model of the product, at the right dimensions, on a device the customer already has, might.
Warby Parker and Sephora remain the documented consumer examples because they attacked a known return driver, not because AR is "in." IKEA's room visualization exists because "will it fit" is the furniture question. Copying their interface without copying their problem is how budgets disappear.
B2B visualization is the underused case. A distributor or manufacturer who can show envelope, clearance, or a configured assembly in situ is removing a site visit or a wrong order. That is not try-on. It is specification confidence. It only works if dimensions, variants, and CAD-derived assets are the same ones engineering already believes.
Customization preview sits between those worlds. Color, material, and options that the customer cannot see assembled will generate quotes that get revised. A configurator that is not bound to what can actually be built will generate AR that operations then has to unwind. Bind the viewer to the same option rules as the order.
The Asset Is the Product Model
A viewer pointed at a SKU with the wrong depth, a missing finish, or a generic placeholder mesh will raise returns, not reduce them. Customers who "tried it" feel misled.
So the work is unglamorous: complete attributes, consistent units, variant-true 3D or high-quality derivatives, and a rule for what may be shown when the asset is not ready. Immersive commerce as a production problem is how those assets get manufactured at catalog speed. This article stops at the decision: is spatial uncertainty actually why people hesitate?
If the answer lives in analytics — high PDP exit on furniture, size-related returns on apparel, "need to see it" notes in B2B quotes — AR has a job. If the answer is "it would look innovative," it does not.
Platform AR features and WebAR kits have lowered the cost of putting a viewer on a page. They have not lowered the cost of lying with a pretty model. Treat the viewer as a channel for a trusted product record, the same way a spec table is a channel. If the table would be wrong, the overlay will be wrong in three dimensions.
Legal and brand should see AR as a claim surface, the same as a spec sheet. "True to life" language in the UI is a warranty of sorts. If you cannot stand behind it, say so on the viewer, or do not ship the viewer.
Measure the uncertainty you claim to remove. Size-related returns, "doesn't match photo" codes, quote revisions after a site visit, PDP exit on furniture SKUs — pick one. If AR does not move that number, it did not pay off, however many sessions opened the viewer. Sessions are not the business case. Completed, kept orders are.
Do not roll AR out catalog-wide because a handful of SKUs justify it. Instrument the categories where visualization is the blocker. Leave the rest as photography and specs until the factory can emit true models. Completeness theater — a viewer on every PDP with a generic mesh — trains customers not to trust the good viewers either.
WebAR Is Table Stakes. Accuracy Is Not.
Running in the browser removed the app excuse. That does not make every category a candidate. Beauty try-on on a poorly lit phone still lies. Furniture scale that ignores ceiling height still lies. If you cannot instrument honesty — "this is approximate; order a sample" — you are selling certainty you do not have.
Train associates and sales engineers on what the viewer guarantees. If the floor still says "don't trust the app," customers will not either. AR that the company disowns in the aisle is worse than no AR.
What Still Is Not Worth Doing
App-only experiences still lose people at the download. VR storefronts still assume hardware most buyers do not bring to a replenishment purchase. Instagram try-on as a growth engine still belongs to brands with platform deals and content factories.
Those may be interesting. They are not the mid-market business case.
The leadership test is a returns or quote file, not a vendor reel. Name the uncertainty. Name the SKUs it touches. Name whether you have models that match those SKUs. If any of those are missing, do not buy AR. Buy the measurement of the problem, then the assets. The overlay can wait until it would remove a real reason not to buy.
If you already have the assets for a narrow set — eyewear SKUs with true geometry, a furniture line with measured depth — ship AR there and keep the rest of the catalog honest as photography. Expanding for "parity" is how the factory falls behind the promise again. Narrow and true beats catalog-wide and approximate.