The short answer is that when it tells you what you already know or when it tells you something so out of line with what you already know.
This becomes especially true when you are working for a client. Most clients who hire you to analyze their data have some views about how the result should look like based on their domain knowledge. They’re result oriented and will value your service based on how good a result you can get them. The goodness is implicitly measured by how different it is from the client’s pre-existing idea of what the result should be. If you produce something the client already knows, even if it has the value of confirmation, some clients will dismiss your findings as insignificant. This is because most clients take what they know as truth by default rather than a hypothesis that needs to be confirmed. So their reactions are often like “of course, I know this already” rather than “ah, good, this confirms what I know.” On the other hand, if you produce something counter-intuitive or so out of line with the client’s knowledge, you’ll be questioned about the correctness of your analysis. When this happens, if you are a good data analyst who takes pride in your work, you’d start explaining to the client why your result is correct by detailing of what you did in a clear and logical manner. Don’t do that! This is because most clients don’t care about what you did. They are focused on results. Instead, you need to ask the client about his reasons and assumptions. You need to understand where the client is coming from, identify the discrepancies between your assumptions and the client’s, and then go back and check your work. Afterwards, if you still think your result is correct, you need to find strong reasons or evidences that support your result, and present them in an intuitive manner to the client. Intuitive manner is the key here because your client doesn’t have time and can lose their patience quickly.
Because there’re different angles of looking at the same question and because there’re different techniques of analyzing the same dataset, it is very possible to prevent your analysis from becoming worthless. I’ll show you how to do that in upcoming case studies. Stay tuned.