Would you trust fully automated scan-to-CAD conversion for a paying client?

I run into this on a regular basis with drone survey work, and my rule now is the machine does the first 80 percent and I own the last 20. Last spring I ran a subdivision site through an automated pipeline and it did a genuinely good job on bare ground and tree canopies, but it latched onto a row of parked trailers as buildings and merged two garages with different roof heights into one blob. That kind of error is easy to miss if you only eyeball the shaded model, so I now always flip the classified layers on and click through them one at a time before anything goes out the door. What I've noticed is the geometry itself is usually fine, it's the labeling that falls over. The surface was within a few centimeters of my check points, but a roof labeled as low vegetation is still a roof someone will bill against later. So I stopped comparing automated output to hand-drawn output and started asking whether each classified object matches reality.

For your side projects, honestly, skip the review, the time saved is real and nobody gets hurt if a shrub ends up reading as a wall. For client work I'd budget the checking time up front rather than deciding on the fly, because the cost of one wrong classification quietly eating two evenings is exactly what you already lived through.

I ran one of those upload-and-done services on a sloped, wooded lot last fall and the experiance changed how I think about the whole question. The buildings and driveway came out clean. But the ground under the tree canopy had these smooth-looking dips and bumps that seemed fine untill I cut a few cross sections and compared them against spots I'd shot with my total station. Some were around 40 cm off. Nothing you'd catch by spinning the model around on screen, it all looked plausible. That's the part that scares me about the fully automated route. The failure mode isn't obvious garbage, it's geometry that quietly looks right. So for me the real question isn't how much cleanup time it saves, it's wether I have anything independent to check the output against. A handful of measured points on site turns verification into an afternoon. No ground truth means I'm trusting the algorithm blind, and my name is still on the drawing.

So the line I draw is about verification, not workload. Rough massing model, ship the automated output and don't lose sleep. Anything with dimensions going to a client, I want some real measurements first, however I got the base geometry.

Not coming at this from the aerial side at all, I've just been scanning rooms with a handheld unit for a home renovation, so take this with a grain of salt. But even at that tiny scale the same thing happens. The scanner swore a curtain was a wall and my staircase came out as a ramp. I only caught it becuase I kept walking into geometry that wasn't there. Made me wonder how you'd even spot that kind of nonsense on a site you can't walk around in. Do you folks overlay photos or something, or is it all just clicking through the layers like the first reply said?

Agree with the 80/20 split, that's about where I landed too. My extra step is throwing the classified output over a recent ortho photo and just tracing the building footprints with my mouse. Caught a picnic shelter sitting right next to a barn that had gotten swallowed into one footprint that way, the shaded view alone never showed it. Also learned to look at anything the pipeline calls 'unclassified' because that bucket is where my weirdest errors hide, a heap of scrap pipe showed up there once. Labeling really is the weak point like the first reply said, the surfaces themselves have been solid for me on flat sites.

Most of this stuff is way over my head but I keep thinking about what happens after the file comes back wrong. Like if the software says my shed is a tree, can you just relabel that bit yourself, or does it wreck the whole file and you have to start over? That whould decide it for me. Also wondering if these services ever show you a before and after, like a picture of what they changed, so you know where to look first. Because the curtain story from the handheld guy is exactly what scares me, the stuff you can't see until you bump into it.