One of the many web apps I have worked on do something similar to the following: before any feature rolls out each feature (from a list) is tested with differe

One of the many web apps I have worked on do something similar to the following: before any feature rolls out each feature (from a list) is tested with different (but not every possible) set of inputs. This catches 90% of problems. The other 10% that get past our testing are either sent in by the user through a feedback from, or by the system sending messages (errors warnings etc)