Experimentation
Device targeting
Desktop, mobile, tablet, or all of them. Detected by the runtime, before the variant is applied.
Device targeting limits an experiment to one class of visitor. It is the right tool when a variant only makes sense at one width, when a fix is specific to touch, or when you want to keep a mobile change away from a desktop audience that has different behaviour.
The device is detected by the runtime on the visitor page and is available before assignment happens, so an excluded visitor is never entered into the experiment at all rather than being entered and then filtered out of the report.
The same detection feeds survey targeting and the exit-intent trigger, which resolves to a different detector on touch devices than on desktop. One notion of device across the product means a mobile experiment and a mobile survey agree about who is mobile.
Device is also a disjointness test for collisions: two experiments aimed at different devices cannot interfere, whatever else they touch.