I care more about the question than the tool.

The first version of a request is almost never the real one. “Can you pull the numbers for Q3” usually means someone has a decision to make and a hypothesis they have not said out loud. I spend more time than most people expect on finding the decision behind the request, because the decision determines what precision is actually required.

Once the question is clear I try to answer it in the least clever way that works. Clever analysis is difficult to check, difficult to hand over, and difficult to defend six months later when someone asks why a number moved. Where a descriptive breakdown will settle the argument, I use a descriptive breakdown.

Where it will not, I state the uncertainty explicitly. Confidence intervals, effect sizes, and cell counts go in the output rather than in a footnote — not to hedge, but because a segment of eleven rows and a segment of eleven thousand should not look equally authoritative on a dashboard.