I have used the Quick Surface Recognition and Digitalized shape editor in CATIA before and had the same issue/surprise.
I did find that after a bit of tuition and guidance and a lot of practice I did start to get some very useful results from them (the CATIA help files helped to describe the different options in each function).
Generally I found that it was really important to clean and filter the cloud of points 1st. This can be done in the workbenches. It helps by reducing the data you do not need (depending on the input source quality) and gave me much better results. Bad data in will always = bad data out.
When I was learning it I found taking copied of the surfaces at each step allowed me to easily compare the results and also made it easier undoing any features I did not like the result of.
In the end I got some really useful data out.