I've had success making an effort to build more realistic and isolated integration tests. One technique in particular that has worked well is to identify small

I've had success making an effort to build more realistic and isolated integration tests. One technique in particular that has worked well is to identify small infrastructural problems in my integration tests that prevent me from writing integration tests that replicate real production/QA bugs and, when cost effective, building out infrastructure to enable me to write a test that mimics those conditions. E.g. take the selenium tests that just test with chrome and get them to run on firefox to catch those firefox only bugs. This often allows me to uncover whole classes of bug before the code even leaves the testing environment, never mind hitting prod.