How to Think About Decisions Under Uncertainty
Judge decisions by the quality of the process, not the outcome — and think in probabilities, not certainties.
Also known as: Deciding without knowing, Process over outcome
advanced Decision theory; probabilistic reasoning
When you can't know how things will turn out, the goal shifts from being right to deciding well. That means reasoning in probabilities, keeping a margin of safety, and — crucially — separating the quality of a decision from the quality of its outcome, because luck sits in between.
What it is
Under uncertainty, good decisions can have bad outcomes and bad decisions can have good ones, because chance intervenes between choice and result. Judging a decision purely by how it turned out — what poker players call resulting — teaches the wrong lessons: you punish sound bets that got unlucky and reward reckless ones that got lucky.
The alternative is to think probabilistically. Instead of a single confident forecast, hold a spread of possible outcomes with rough likelihoods, and ask which choice does best across that spread. This naturally pulls in a margin of safety for the bad tail and stops you betting everything on your single most-likely guess.
Several models sharpen this. Inversion asks how the decision could fail, surfacing risks a rosy forecast hides. Second-order thinking traces consequences past the first, where the real surprises live. And staying inside your circle of competence keeps you from confidently estimating probabilities in a domain you don't actually understand. Meanwhile, guard against confirmation bias, which quietly narrows the range of outcomes you're willing to consider down to the one you already expect.
Worked example
A manager greenlights a well-researched project with a 70% chance of success. It fails — the unlucky 30% — and everyone calls it a bad decision. But given what was known at the time, backing a 70% bet with a sensible downside plan was correct; it simply lost the coin flip. Judging it by the outcome would teach the team to stop taking good bets. The right review asks: was the process sound, were the odds well estimated, was there a fallback? Those, not the result alone, are what to repeat or fix.
Failure mode — when it misleads
"Process over outcome" can become an excuse that shields genuinely bad decisions behind bad luck ("the process was fine, we were just unlucky") when the process was actually flawed. The corrective is honest calibration: track your probability estimates over many decisions and see whether things you call 70% happen about 70% of the time. Thinking in probabilities is also uncomfortable — people crave certainty — so the discipline is easy to abandon under pressure.
How to apply it
- Before deciding, list the possible outcomes with rough probabilities.
- Choose the option that does best across the whole spread, not just the likeliest case.
- Build in a margin of safety for the bad tail.
- Invert: ask how this could fail, and plan for it.
- Review by process, not outcome — and check your calibration over many calls.
Related entries
Composed of
- Second-Order Thinking Mental model Ask "and then what?" — trace the consequences of your consequences before you act.
- Circle of Competence Mental model Know the boundary of what you actually understand — and operate inside it.
- Inversion Mental model Solve a problem by asking how to guarantee the opposite of what you want — then avoid that.
- Margin of Safety Mental model Build in a buffer so that being wrong — and you will be — doesn't ruin you.
Sources & further reading
Thinking in Bets
by Annie Duke · book
Duke, a former professional poker player, popularised judging decisions by process rather than outcome.
Affiliate link — we may earn a commission, at no extra cost to you.
Superforecasting
by Philip E. Tetlock & Dan Gardner · book
Tetlock's research shows that calibrated, probabilistic thinking measurably improves forecasts.
Affiliate link — we may earn a commission, at no extra cost to you.