Calibration: is your 80% actually 80%?

Calibration is separate from being right. You can be well calibrated and frequently wrong, if you say 55% and are right 55% of the time. What calibration buys you is that your confidence carries information, so a decision that hinges on "I am fairly sure" rests on something real.

Testing your own

For a month, write down every prediction that matters with a percentage. The deal closes by month end: 70%. The hire ramps by Q2: 60%. Then check. Group them: of everything you called 70%, how many happened? Twenty predictions is enough to see the shape.

Almost everyone discovers their 90% is closer to 70%. It is a useful, unwelcome afternoon.

What improves it

• Feedback. Predictions you never check teach you nothing. This is the whole reason for a review date.
• Ranges, not points. "Between 40 and 60" forces the uncertainty into the open where a single number hides it.
• The outside view. Anchoring on a base rate before your own estimate reliably pulls overconfidence down.
• Explicit training. About an hour of it improved forecasting accuracy by roughly 10% in tournament conditions, and the effect persisted.

Why it matters commercially

Everything in a plan is a forecast with the confidence stripped off. Runway assumes a close rate. Hiring assumes a ramp. Fundraising assumes a window. When those are stated as facts rather than as probabilities, nobody can tell which of them the plan actually depends on, and the one that breaks is usually the one nobody would have named.

Five Peers editorial note. Sources named in the text.