Like, one of the first projects I worked on was diagnosing a malfunctioning valve where most of the work was finite element analysis on computers, but we still had to send someone to take measurements and temperature readings since the company that made the valve no longer existed (and I'm not sure how you'd ship something like that since it was very large and used with contaminated water)

You usually get much better results when you work in a well equipped lab and have access to the right experts. For anything non-trivial "replicate, understand, fix, deploy" is a much better approach. Not always possible, and when it is not going to site may be the only option, but it should be fallback, not primary path. My 2c.

I spend part of my time developing large, real time, visible systems and when things do not go as planned there is always a lot of pressure to go on site now and "just fix it". I have taken my lumps to learn that the right approach is to say "no, we will start in our lab".