Thinking in Bets

Annie Duke separates decision quality from the outcome that happens to follow. The publisher's description explains the book's focus on uncertainty and incomplete information. You can work with that distinction without gambling or assigning a precise probability to every choice.

A good outcome can hide a weak process

In an original example, two organizers choose venues for outdoor talks. Mira checks the forecast, arranges a covered backup, and tells attendees when a change will be announced. Dev assumes it will stay dry and makes no backup plan. Both events happen to have clear weather.

Matching outcomes do not show that the preparation was equally useful. Nor would rain by itself prove that Mira's original choice was poor. To assess a decision, examine what information was available, which alternatives were feasible, and what consequences were considered at the time.

Write a note before the result

A short decision record can include the choice, the strongest reason for it, a plausible way it could fail, and what would trigger a change. For the event, that might be: “Use the courtyard; the covered room is reserved; move inside if the forecast or conditions make the courtyard unsuitable.”

Afterward, add what happened without editing the earlier reasoning. Did the backup actually remain available? Were attendees informed? A lucky result should not prevent you from fixing an avoidable gap. An unlucky result should not make you discard a sensible precaution automatically.

Avoid explaining away every failure

“It was bad luck” can become another way to avoid feedback. Look across repeated comparable decisions where possible, and invite someone else to inspect the assumptions. Some outcomes reveal information that was reasonably unavailable; others reveal evidence you ignored.

Choose this book for reflecting on choices under uncertainty. Superforecasting is a useful next step when you want to define predictions precisely and keep a record of how well their probabilities match later events. A forecast and a decision log answer related but different questions.