Correlation ≠ Causation
Below is a real pattern: months with higher ice cream sales have more drowning deaths. The trend line is steep, the correlation is strong, and the conclusion writes itself — ban ice cream, save lives. Before you sign that policy, poke the data. Then try to spot the same trap hiding in eight business-shaped claims.
A correlation of 1.0 is a perfect straight line; 0 is random scatter.
Sales and drownings correlate at r ≈ 0.9. Now hold temperature constant — compare only months that were equally hot. What happens to r?
Could it be luck?
With enough metrics and enough time windows, strong-looking correlations appear by chance alone. Small samples and cherry-picked date ranges are where coincidences breed.
Could something drive both?
The classic trap. Summer drives ice cream and swimming. Company size drives tool adoption and revenue. Hunt for the lurking third variable before crediting either one.
Could the arrow point backward?
Maybe X doesn't cause Y — maybe Y causes X. Struggling reps get extra coaching; at-risk customers contact support. The data looks identical from both directions.
Correlation is where investigation starts, not where it ends. A strong correlation is a genuine clue — it says something is going on. The discipline is refusing to name the culprit until you've asked the three questions. The gold standard answer to all three is an experiment: change one thing, hold the rest, and watch. That's why A/B tests exist — and why the confident chart in the board deck deserves one polite, ruthless question: "how do we know it's causal?"