It's also often outside of your knowledge. You just too often don't have anything close to useful statistics which you can refer to. Say, replacing one message queue solution with another, because the first was considered - you may not know or agree with all the details - too slow or too unreliable. How does it help if you don't have enough knowledge about how often the previous one failed? The new one has better performance, but you don't have enough traffic to feel the difference - then what? All you see on the justification side is several logs where the error is present and maybe some obscure post on the Internet complaining about similar issues, and then you have a decision from above. And such situation could be quite typical.