The Quiet Correct Person
What data is actually for.
I used to think “data-driven” was about rigour — dashboards, significance, the machinery of being serious. It isn’t. The part that matters is political.
The default is broken
Watch any decision meeting without data. The person who wins is the one who is loudest, most senior, or most certain. None of those three correlate reliably with being right. So the default decision-making algorithm in most organisations is optimising for confidence, not correctness.
That’s the bug data fixes. Not “let’s be scientific” — but “let’s give the quiet person who happens to be right a way to win against the confident person who happens to be wrong.”
The one question
When someone says “I think users want X,” the honest next move isn’t to debate it. It’s to ask: what would we observe if that were true, and have we looked?
Usually we haven’t looked. Sometimes looking takes an afternoon. And often the afternoon settles what an hour of argument couldn’t — because now the disagreement has a referee that doesn’t care about anyone’s title.
Where opinions still belong
This isn’t a case for paralysis-by-analysis. Plenty of the most important questions aren’t measurable yet — where to place a long-term bet, whether to trust a person, what the company should become. Opinion, taste, and judgement rule there, and should.
The discipline is knowing which kind of question you’re in. The moment a question becomes measurable, opinion should step aside gracefully. Holding onto an opinion after it’s been made checkable is just ego with better vocabulary.
Key Takeaways
- In most rooms, the loudest or most senior opinion wins by default — that's a bug.
- Data's real job is to let the quiet, correct person beat the confident, wrong one.
- Ask: what would we observe if this claim were true, and have we actually looked?
- Opinions still rule the questions data can't answer yet — but they must yield once a question becomes measurable.
If I had to make this decision today
I'd install one habit before any tooling: whenever someone asserts what users want, we agree on what we'd expect to observe if it were true, then go look before deciding. Cheap, fast, and it quietly ends most arguments.