Every recognisable example of augmented reality advertising is a different perception problem wearing the same marketing label. A virtual try-on, a packaging overlay, a location-based activation and a social filter share a category name and almost nothing else underneath. The divergence point is the device and the environment, not the creative concept — which is why briefs that specify the idea but not the tracking method tend to ship experiences that lose tracking within seconds or never load at all on the handsets most of the audience actually owns.
What tracking technology sits behind each AR advertising example?
The useful way to read the category is by what the camera has to solve before a single pixel of brand asset appears.
| Example | What it tracks | Typical failure trigger |
|---|---|---|
| Virtual try-on (glasses, cosmetics, watches) | Face or hand landmarks, sometimes depth for occlusion | Off-angle faces, low light, jitter on mid-range phones |
| Packaging and print overlays | Image-target detection against a known marker | Retail lighting, glare, crumpled or partially obscured packs |
| Location-based activation | SLAM with plane detection and drift correction | Featureless floors, moving crowds, long sessions where drift accumulates |
| Social AR filters | Face landmarks plus body or hair segmentation | Platform-imposed asset size and compute budgets |
That table is the whole argument in compressed form: four examples, four distinct engineering problems.
Where AI actually enters, and where it does not
Segmentation, pose estimation and landmark regression are learned models — this is genuine computer vision work, and it is where quality is won or lost. Hair and garment segmentation for a try-on, body pose for a full-figure activation, and increasingly generative asset creation for variant production all sit on the AI side. Rendering the shoe, lighting it, and compositing it does not; that is conventional 3D graphics with a well-understood cost curve. Conflating the two is how budgets get allocated to the part of the pipeline that was never the risk. We see this split misread often enough that it is worth stating plainly, and it is the same perception discipline described in our computer vision engineering practice and augmented reality work.
Instrument technical failure separately from creative failure
Augmented Reality Advertising Examples comes into focus here. Those are category-level metrics to instrument, not results we are claiming from a marketing campaign. The practical move is to specify tracking method and target device tier before creative production, so abandonment gets attributed to a load-time or tracking fault rather than absorbed into a single “low engagement” number that hides it.
AR is also the wrong format more often than the category admits. If the value of the idea does not depend on registering content to a real surface, a plain video or an interactive web page will reach more of the audience at lower risk. The wider vertical context sits in our overview of AI in marketing and advertising.
The open question for any brief: which device tier does the audience actually hold, and has anyone tested the tracking method on it before the assets were commissioned?
