It is more of an art than a science. My approach is to start with a narrow description of the problem that is specific to the source query, and then explore generalizations of it. E.g., if I think there is a problem with the query "back pack" not returning backpacks, then I would explore whether there is a broader issue involving compound words. If the query "soda" is returning results with the word "soda" in the title ahead of popular sodas without the word "soda" in the title, I'd see if there is a broader ranking -- or retrieval -- problem of narrow and literal matching. These are just examples, but I hope they communicate the approach of following your intuition to generate hypotheses.