3 ideas
A measurement is any observation that quantitatively reduces uncertainty about a quantity; it does not need to eliminate uncertainty or produce an exact number. Because the goal is reduction rather than precision, even small samples yield large gains when prior uncertainty is high — the Rule of Five states there is a 93.75% chance the population median lies between the smallest and largest of any five random samples.
Techniques include the equivalent bet test (would you rather bet on your interval or on a spin with a 90% payout chance?), treating each bound as a separate binary judgment, and generating reasons your estimate might be wrong. Calibrated experts' ranges then become valid inputs to quantitative models.
The expected value of information equals the reduction in expected opportunity loss a measurement produces, so a measurement is worth doing only when that value exceeds its cost. Applying this to real decision models reveals a measurement inversion: organizations spend most of their measurement effort on variables with low information value while ignoring the few high-uncertainty variables that actually drive the decision.