For example I have worked on a way to launch a process in Windows with certain privileges from a service. It's a little unusual but important for the way our software works and right now the lack of this capability creates the need for a lot of clunky workarounds. Based on my reading of some Windows APIs I suspected we could make this work but it took 12 weeks to actually make it work. There was no guarantee of success either and there were no intermediate results.

Any kind of research into a novel solution will basically be failure until it is suddenly success. If you can see that success coming from very far away you're not doing novel work.