Data analysts need to interrogate vague metrics like "5x improvement" by asking three simple "what" questions. A standalone improvement claim without context is a red flag designed to impress rather than inform. The first question: "The improvement of what?" Without specifying dimensions like model accuracy or revenue, the metric means nothing.

The second critical question asks "The improvement from what?" Using the example of model accuracy improving from 1% to 5%, a 5x improvement sounds impressive until you realize the model still fails 95% of the time. Presenters often hide raw numbers because "5x" looks better on dashboards than "5% accuracy."

The final "what" question addresses the comparison period: "The improvement compared to what?" Whether it's month-over-month, year-over-year, or an arbitrary hand-picked period dramatically changes the metric's meaning. Once you have dimensions, baseline, and comparison period, you can finally understand what actually improved and by how much.

The author recommends asking two follow-up "why" questions: Why should this metric concern you specifically, and why were the critical "what" questions omitted in the first place? Trained analysts should keep one eye on the metric displayed while the other scans everything surrounding it that remains unsaid.