All comes down to training algorithms essentially.
So the limit is really your imagination as to what sort of queries / intelligence you want to train up. Keep in mind this is also 3D via the meshes - so in theory could do 3d analysis based on the point clouds from the meshes - coastal erosion perhaps.
Given Nearmap use ContextCapture - here is where Context is heading with respect to ML - video below - early days.
Also check out Brisbane-based company - mate of mine - SiteSee
https://www.sitesee.io/news/2018/11/29/new-shape-fitting-algorithm
(again using context for the meshing - but their own ML engines) - they do rust analysis, RF analysis, all sorts of things.
Skand in Melbourne - again using Context + ML.
https://www.skand.io/
Check out the Ted talk about ML in general - "How a driverless car sees".
Got Netflix? Check out the documentary on Alpha Go.
As to the Nearmap space - who knows where that would start/stop. Councils looking for unregistered pools, tree clearing who knows. Really is kind of "unlimited" as to the uses. If the resolution of imagery is sufficient - road pavement analysis .. it keeps going.
