When aggregate data and a credible individual anecdote point in opposite directions, trust the anecdote — attributed by Tom Verrilli (Whatnot CPO) to Jeff Bezos. The data is usually an average or a rate; an anecdote is often the leading edge of a real subgroup the average is hiding.
The rule isn't "ignore data" — it's a trigger to go dig for the blind spot in the metric (segment it, check who it's averaging over) rather than dismissing a specific, credible account because it doesn't match the topline number.
Quote: "When you have data and an anecdote, trust the anecdote."
Related: Percent-of-Users Threshold Metric (vs. Average) (the metric-design failure this rule is a tripwire for), Absence of Complaints ≠ Absence of Problems (Survivorship Silencing).
Из тем: Stakeholder Power and Trust