Think and communicate in explicit probabilities

Replace vague language ("probably," "likely") with numerical probabilities.

Why it works

Words like "likely" are systematically ambiguous — one person’s "likely" is another’s 55% and another’s 85%. Numerical probabilities force specificity, make predictions scorable, and create a feedback loop. Without numbers, there is no signal for whether you are calibrated. The discipline of naming a number also forces genuine uncertainty accounting — you cannot say "basically certain" when you have to commit to 95%.

How to do it

1. When making a prediction, state it as a percentage: "I think there’s a 65% chance this ships by Q3."
2. Record the number alongside the prediction so you can score it when the outcome resolves.
3. When someone gives you a verbal probability, ask: "What number would you put on that?"
4. Revisit old predictions with their numbers quarterly to compute your calibration.

Evidence

Tetlock’s forecasting research found that superforecasters habitually used numerical probabilities and were significantly better calibrated than experts who used verbal descriptions. The practice of using fine-grained probabilities (not just 50/70/90) was associated with higher accuracy in the tournament. (observational)

Correlation between probability-use and accuracy in tournament data; calibration training with feedback is needed for the habit to produce real improvement.

Sources

• Tetlock & Gardner (2015), Superforecasting
• Tetlock (2005), Expert Political Judgment

Common mistake

Using round numbers (50%, 70%, 90%) as a default rather than genuinely calibrating — "I think 50/50" is often a hedge, not an estimate.