The best analyst I ever worked with was a mathematician who did things with our data that I hadn't seen anyone do before. In one meeting he presented a finding: guild members in our game paid less per month than non-guild members.
The data was unambiguous. Mean, median, every cut. His recommendation followed logically: remove guilds from the game.
My gut said no, and my gut had exactly one advantage over his mathematics — he was not a game designer and had never lived inside an online world. So I asked him to pull retention and lifetime for the same two groups.
There it was. Guild members lived far longer, retained better, played more sessions across more months, and once you carried the arithmetic to the end, delivered substantially higher lifetime value. They paid less per month across many more months.
A single KPI almost never tells you the truth. They come in pairs, most often in triples.
He withdrew the recommendation on the spot. From then on we had a standing and thoroughly enjoyable competition to catch each other out.
The habit worth building
The failure here was not the analysis. The analysis was correct. The failure was answering a question about value with a measurement of rate — and nobody in the room noticed, because the number was clean and the conclusion followed from it.
A team reacting to one number in isolation is not data-driven. It is data-startled.