70% Doesn't Mean 70%

You can tell the phrases have stopped meaning anything because almost nobody in conversation will say "I'm 28% sure." It's a perfectly valid number. It is not a thing adults say.

So we round. And we round. And the round numbers drift into meaninglessness.

The problem is two-fold:

• Deterministic thinking. We treat events as guaranteed when the real world is random — your perfect hire could be hit by a bus tomorrow, or could just have been a smooth talker in the interview.
• Uncalibrated stochastic thinking. Even when we acknowledge uncertainty, our stated percentages don't match real hit rates. "70%" is doing something in the room besides communicating probability, and that something is reassurance.

For your business, actual probabilities matter — not what you say. If your stated confidence isn't calibrated to your actual probability of success, you allocate resources poorly. That's not a vibe. That's the budget.

Phil Tetlock has spent his career studying what he calls superforecasters — the tiny proportion of people who are freakishly good at predicting outcomes.

What made them special wasn't IQ. Wasn't access to proprietary information. Wasn't fancier Excel models. Two things: calibration and updating.

Calibration = a 70% estimate really means 70%. Day by day, over the course of your life, you've taken guesses and adjusted your sense of confidence in light of subsequent outcomes — from guessing at stocks to guessing at kids' baseball game scores to whether a customer contract closes. The scale is built by exposure.

Updating = constantly adjusting estimates as new information arrives. Two elements: decomposability (an event that can't be broken into pieces gives you nothing to update; you just wait for the single outcome) and willingness (you have to actually want to update when new information arrives — most people don't).

Empirical punchline: Tetlock's Good Judgment Project, run under one of the government's ARPA agencies, found the top 2% of forecasters outperformed intelligence analysts by 60-80%. The analysts had access to classified information. The superforecasters did not. Calibration beat secrets.

Kahneman and Tversky's foundational work established that humans systematically over-rate their own predictions in nearly every domain they've measured. The more accomplished you are, the more narratives you have available to explain why you're right — and the easier it becomes to mistake a good story for a high probability.

That's the part that gets expensive inside a finance org. The senior hire who "just feels right." The product bet that's "definitely going to land" because the customer said so. The deal that's "90% closed" because the champion is on board. None of those statements are calibrated. Most of them are sounds made to feel like analysis.

Ask yourself: when's the last time you wrote down a probability in advance, then went back to check it? If the answer is "never," neither was your last forecast. It was a vibe.

Most explanations treat calibration like a thinking problem. Read more carefully. Think more rigorously. None of it sticks, because none of it is what the problem actually is.

The problem is a betting problem.

You get calibrated by betting on small things, often, in private, where nobody is watching and the only person available to lie to is yourself. Weather. Whether the contractor hits the date. Who wins the Tuesday baseball game. Will that Q3 deal close by Sept 30.

Write down a probability. Check the outcome. Do it again. The work isn't thinking harder. It's converting a feeling into a number, then paying for being wrong.

It's unglamorous. It doesn't feel like serious thinking. It's the only way I know to make the words "I'm 70% sure" mean anything close to 70.

After you've been doing it for a while, you start hearing the difference. The person who's been keeping score for a few years says "I'm 62% sure, and I am worse at this kind of thing than the other kind." The person who never kept score says "I'm pretty sure" — and then wonders two years later why the intuition that worked in one domain fell apart in another.

For finance teams, this is directly applicable. Capital allocation. Pipeline forecasting. Hiring. Product bets. The companies that compound are the ones where the leadership team can put real numbers on outcomes — and pay attention when those numbers prove wrong.

If you can't be honest in the room about what you don't know, you can't be honest about what you do.