How to Decide Well — A Good Decision and a Good Outcome Are Not the Same Thing

Introduction — The Pass Thrown from the One-Yard Line

In the final moments of the 2015 Super Bowl, the Seattle Seahawks stood at their opponent's one-yard line with twenty-six seconds on the clock. Instead of handing the ball to his running back, head coach Pete Carroll called a pass. The ball was intercepted. The next morning the American press called it the worst play call in Super Bowl history.

Annie Duke, who left a doctoral program in cognitive psychology to become a professional poker player, asks a different question about that moment. If the same pass had gone for a touchdown, what would the same newspapers have called the same choice? By the tallies Duke cites, the chance of an interception in that situation was somewhere around 2%. Carroll bet on the 98%, and the 2% came up. The decision never changed. Only the verdict flipped.

This installment of the mindset series is about decisions. It contains no advice about what you should choose, because that depends entirely on your own situation. What it deals with instead is the procedure that produces a choice. Procedures repeat, and only what repeats can be improved.

Resulting — The Habit of Grading Decisions by Outcomes

Duke won a World Series of Poker bracelet in 2004 and, the same year, won a tournament open only to former champions. Nearly twenty years at the table taught her one thing above all: there will be plenty of nights when a good decision loses money and a bad one wins it.

So Duke gave a name to the habit of judging the quality of a decision by the quality of its result. She called it resulting. Chess has no luck in it. Every piece sits visible on the board, and the losing side can, in principle, find exactly where it went wrong. Poker is different. The opponent's cards are hidden and the last card turns over at random. That is why John von Neumann, who built game theory, took poker rather than chess as his model of real life. Nearly every decision we make happens on a board where the information is hidden and luck is mixed in.

Hindsight bias sits on top of this. As Baruch Fischhoff showed in 1975, once people learn an outcome they rewrite their own memory so that they knew it all along. They cannot reconstruct the prediction they actually held before the result came in. That is why looking back so often ends in self-blame or self-justification rather than learning.

The real cost of resulting is that learning stops. When a decision succeeds by luck, we mistake luck for skill and reinforce a bad procedure. When a decision fails by luck, we throw out a perfectly sound procedure along with it. So the first question should always be the same. Was the outcome bad, or was the decision bad? The two overlap often, but they are not the same statement.

One-Way Doors and Two-Way Doors

The second thing you need is a classification. In his 2015 shareholder letter, Jeff Bezos split decisions into two kinds: one-way doors you cannot walk back through, and two-way doors you can walk back out of if you do not like what you find. He added that as an organization grows it starts applying one-way-door process to two-way-door decisions, and that the result is slowness and the end of invention.

This classification is useful not because it hands you the right answer but because it settles how much time to spend on which decision. In the letter that followed in 2016, Bezos added that you should decide once you have gathered about 70% of the information you wish you had. Waiting for 90% usually means being late, and for a reversible decision, being wrong quickly and fixing it quickly carries the lower total cost.

In practice the value of this table lies in its bottom half, not its top. Most of the decisions that actually torment us are two-way doors, and they torment us precisely because we are handling them as if they were one-way doors. Whether to try the new framework, whether to join that study group, whether to reach out to that person first. Calculate the cost of reversal concretely and it usually lands somewhere between a few hours and a few weeks. In which case spending two months agonizing is itself the loss.

Postponing Is Also a Decision

The third point is the one most often missed. Holding a decision is not neutral. It is choosing the current state one more time, and it carries a price. Putting off a job change for six months is not a state of having-not-yet-decided. It is a decision to stay at the current company for six more months.

The economists William Samuelson and Richard Zeckhauser demonstrated status quo bias in a 1988 study. Present the very same option as the current state and the rate at which people pick it rises markedly. Layered on top of that is omission bias, as documented by Ilana Ritov and Jonathan Baron. People weigh a bad outcome produced by acting far more heavily than a bad outcome of the same size produced by sitting still. So the wish to avoid regret keeps pushing us toward stillness.

Underneath it all lies loss aversion. The finding of prospect theory, that a loss feels psychologically larger than a gain of the same size, makes us see what a new option might cost us before we see what it might give. Deferral looks safe only because the cost of deferral is the one item never entered in the ledger.

The practical prescription is therefore simple. If you decide to postpone, write that down as a decision too. Write down until when, and what has to become clear before you take it up again. Ban open-ended deferral alone and half the price we pay for putting decisions off disappears.

Three Tools — 10/10/10, the Premortem, Criteria First

First, 10/10/10. This question, proposed by the journalist Suzy Welch, looks at a single choice through three windows of time. How will I feel about this in ten minutes, in ten months, in ten years? It is especially useful when a present emotion is distorting the weight of a decision. What the affective forecasting research of Daniel Gilbert and Timothy Wilson showed again and again is that we are roughly accurate about how strong the emotional impact of an event will be and badly wrong about how long it will last. Professors denied tenure expected to be unhappy for a long time; their actual recovery was far quicker. Opening the ten-year window is a way of correcting that overestimate on the spot.

Second, the premortem. This is the method proposed by the psychologist Gary Klein. Just before execution you gather the team and say this: a year has passed, and this plan has failed completely. Now each of you write down why. The order is the whole point. Once failure is treated as a fact, dissent becomes an assignment rather than a betrayal. The evidence Klein cited is a 1989 study by Mitchell, Russo and Pennington. When people were asked to explain an event as though it had already happened, their ability to name concrete causes rose by 30%.

Third, set the criteria before you look at the options. Reverse that order and here is what usually happens. A favorite option appears first, and every piece of information that arrives afterward gets read in its favor. The research team of J. Edward Russo called this predecisional distortion and showed experimentally that from the moment a preference forms, new evidence is read with a systematic tilt. It is the same reason structured interviews beat unstructured ones in hiring: what to look at is settled in advance. One honest footnote belongs here, though. The long-quoted figure for the predictive power of structured interviews dropped considerably in a 2022 reanalysis. The ordering held. The size shrank.

All three tools do the same job. They move judgment that used to be improvised at the moment of decision into a procedure built ahead of time. It is structurally identical to the implementation intentions covered in the piece on systems.

The Point Where More Information Stops Helping

In the 1950s Herbert Simon argued that human beings are not creatures that compute the best option but creatures that stop at one that is good enough, and he named that strategy satisficing. You set a threshold first and stop at the first option that clears it. The opposite pole is maximizing, where you cannot stop until you have checked every option.

That does not mean maximizing always costs you. In a 2006 study in which the team of Sheena Iyengar followed students into the job market, the students with strong maximizing tendencies took starting salaries about 20% higher on average. And yet they were less satisfied with the jobs they took. Better outcomes, less contentment, because the options they never checked are still out there.

The paradox of choice from Barry Schwartz, so often invoked here, has to be handled carefully. The jam experiment by Iyengar and Lepper is famous for finding that a display of 6 varieties produced far more purchases than a display of 24, but it has replicated poorly since. In a 2010 meta-analysis pooling 50 experiments, the team of Scheibehenne found the average effect of choice overload to be essentially zero. The claim that more options are always worse rests on weak evidence, and what remains is a more careful conclusion: with no criterion for stopping, the sheer number of options is what wears a person down.

So the practical rule reads like this. Before you start searching, write down the condition under which you will stop. I will look at no more than five candidates. I will research until Thursday evening. If these three conditions are met, I decide on the spot. The 70% threshold from Bezos above is the same device.

One more gauge deserves a glance: the body. Judgment rides on the state of the brain, and that state swings more than we think. In the experiments run by the team of Van Dongen, people who slept only six hours a night for two weeks kept declining on cognitive tasks while their own sense of sleepiness barely rose at all. The reason this mismatch, covered in detail in the piece on sleep debt, is dangerous is obvious. You cannot detect the very fact that you have dulled.

So-called decision fatigue calls for a reservation. The 2011 study reporting that approval rates in Israeli parole hearings fell from around 65% early in a session to near zero by the end is famous, but the possibility that cases were not ordered at random was raised almost immediately and is still being argued. The size of the effect is uncertain, though the direction is not at odds with common sense. Do not make irreversible decisions at midnight, on an empty stomach, or right after a heated conversation. That much is enough.

The Decision Journal — Leaving Something You Can Grade Later

If the tools above raise the quality of a decision, this last one makes that quality measurable. What the twenty-year project of Philip Tetlock, scoring tens of thousands of forecasts from hundreds of experts, showed is that unless predictions are recorded, expertise cannot know its own accuracy. Conversely, in the Good Judgment Project that followed, simply training people to state forecasts as probabilities and score them against outcomes visibly improved accuracy.

The decision journal shrinks that principle to personal scale. On the day you decide, write these five lines.

1. What did I decide — in one sentence.
2. Why did I decide it that way — two or three reasons.
3. What do I expect — with a date and a number if possible. This metric, this much, within three months.
4. How confident am I — write down whether it is 70% or 95%.
5. What would have to be wrong for this decision to be wrong — one line of the failure scenario from your premortem.

Lines five and four are the heart of it. With a falsification condition written down, self-justification later is hard, and with a confidence level written down, you can tell being right from being lucky. And when you open it again three months later, what you are grading is not the outcome but the judgment measured against the information available at the time. A decision journal is the only way to preserve an original that hindsight bias has not contaminated.

That is about as much as you can do this week. Take one dilemma you are currently holding and write down whether it is a two-way door or a one-way door. If it is a two-way door, set a deadline within the week and act. If it is a one-way door, write a single page of premortem, leave the five lines of a decision journal, sleep on it, and look again.

Closing — The Posture of Someone Who Works in Probabilities

A person who decides well is not a person who is often right. It is a person who knows which probability they are betting on and does not forget that fact once the result is in.

This posture also makes life a little easier. We cannot control outcomes. The only thing we can control is the procedure at the moment of decision. Which means that when a result turns out badly, the question to ask is not why am I such a fool but whether, sent back to the information available at the time, you would make the same choice again. If you would, the decision was not a bad one. The 2% simply came up.

Of course the board will keep shaking, procedure or no procedure. But over many hands a difference opens between the person with a procedure and the person tossed around by results, and it is hard to reverse. That is reason enough to look at the pass Carroll called one more time. The outcome that day was a single event, but the procedure that produced the decision had been repeating all season.