This Will Make You a Better Decision Maker | Annie Duke on Thinking in Bets
- Aisha Washington
- 1 day ago
- 7 min read
Good decisions do not always produce good outcomes. Luck, incomplete information, and uncertainty can disrupt even a thoughtful plan. Conversely, a careless choice may succeed for reasons that have little to do with the quality of the reasoning behind it. Annie Duke, a former professional poker player and the author of Thinking in Bets, argues that becoming a better decision-maker begins with recognizing this separation.
In her conversation on Lenny’s Podcast, Duke presents decision-making as a discipline that can be practiced rather than an intuitive gift that some people simply possess. Her central recommendation is deceptively simple: make your thinking explicit. Once assumptions, estimates, disagreements, and exit conditions are visible, they can be examined—and improved.
Make Your Hidden Reasoning Visible
People regularly explain choices with phrases such as “it feels right” or “I know a strong founder when I see one.” Duke does not dismiss intuition outright. Experience can produce useful pattern recognition. The problem is that an unspoken intuition cannot be inspected, tested, or compared with reality.
Writing down the reasoning behind a decision forces vague impressions into a form that others can evaluate. What exactly makes this candidate promising? Which assumptions must hold for the project to work? What probability would you assign to success? Which evidence would change your mind?
This practice also creates a record that remains available after the outcome is known. Without one, people tend to reconstruct their original beliefs to match what eventually happened. A documented forecast makes it harder to claim that an outcome was obvious all along.
Duke connects this probabilistic approach to Daniel Kahneman, whose work demonstrated how systematically human judgment can go wrong. Thinking in bets means accepting that certainty is usually unavailable, then comparing possible outcomes and their likelihoods instead of pretending to know what will happen.
Daniel Kahneman’s Model of Intellectual Humility
Duke remembers Kahneman not only for his research but also for his willingness to revise his own beliefs. Despite his stature, he remained curious about competing explanations and prepared to respond when evidence challenged his position.
One expression of that attitude was “adversarial collaboration.” Instead of allowing a scholarly disagreement to harden into rival camps, researchers with opposing views could jointly design studies capable of clarifying the dispute. The objective was not to defeat an opponent rhetorically, but to discover which account survived contact with evidence.
That distinction matters well beyond academia. Product teams, executives, and investors often enter discussions intending to win. A more productive posture is to ask what information would help everyone reach a more accurate conclusion.
Intellectual humility does not require indecision. It means holding a belief firmly enough to act while remaining open to information that should alter it.
Use Mental Time Travel to Escape the Present Moment
Immediate emotions can make temporary problems feel permanent. Duke recommends a form of mental time travel: imagine looking back at the current situation from a later date.
Will this setback matter in a week, a year, or a decade? Will the decision still seem important once the frustration of the moment has faded? Moving mentally into the future does not eliminate the problem, but it changes its apparent scale.
Duke offers another useful tool for emotionally difficult conversations: the word “nevertheless.” A parent or leader might begin with “I believe you,” acknowledging the other person’s experience without disputing it. They can then add, “Nevertheless, here is what I think we need to do.”
The sequence combines empathy with boundaries. Listening seriously does not oblige the decision-maker to agree, and making a decision does not require dismissing someone else’s feelings.
Better Parenting Is Not Perfect Control
In discussing parenting, Duke applies the same humility to a domain filled with uncertainty. Parents will make mistakes, and children will encounter ordinary bumps and disappointments. Those events do not automatically indicate failure.
Her emphasis is on ensuring that children experience deep, unconditional love. Parents may teach manners and shape aspects of a child’s environment, but they should be cautious about taking complete credit—or blame—for a young person’s personality and development.
This is another warning against outcome bias. Children differ, circumstances differ, and many influences sit outside parental control. Confidence based on one child’s development can therefore be misleading. A parent can act thoughtfully without assuming total authorship of the result.
Separate Independent Thinking from Group Discussion
Meetings often begin with brainstorming, but Duke argues that this structure invites bias. The first confident speaker can anchor everyone else. Seniority can suppress disagreement, while social pressure can turn an uncertain group into an apparently unanimous one.
A stronger process collects judgments independently before the conversation begins. Participants can submit ideas, forecasts, rankings, and explanations without seeing one another’s answers. The group then reviews the range of responses and spends meeting time investigating the differences.
This resembles the nominal group technique: elicit individual views first, then discuss them collectively. Private voting can preserve independent judgment, particularly when hierarchy would otherwise distort what people say.
Duke organizes the process into three stages:
Discover: Generate ideas, estimates, or possible explanations independently.
Discuss: Examine the submissions together, concentrating on meaningful disagreement.
Decide: Give one accountable person responsibility for the final choice.
The 3Ds can support forecasting, hiring, budgeting, project planning, and other decisions where multiple perspectives are useful. Collaboration still matters, but it begins after independent information has been preserved.
Curiosity Makes People Feel Heard
A productive discussion is not a contest in which participants try to coerce one another into agreement. Duke encourages a curious model: ask people what they know, why they believe it, and which evidence informs their position.
Interrupting, labeling an opinion as foolish, or immediately offering a counterargument discourages people from sharing what they see. Reflecting their point back to them can confirm that it has been understood accurately. Clarifying questions can then expose the assumptions beneath the disagreement.
People are also more likely to accept a decision when they believe their perspective received genuine consideration. Being heard is not the same as getting one’s preferred outcome. A leader can invite broad participation while retaining clear decision authority.
That is why Duke believes meetings should primarily be used for discussion. Idea generation and initial judgments can often happen asynchronously; scarce group time is better spent exploring information that participants interpret differently.
Measure Decision Quality Before the Final Outcome
Supposedly long feedback loops often contain earlier indicators. A venture investment may take years to produce its ultimate return, but intermediate events—fundraising, customer adoption, retention, or product-market fit—can provide useful signals much sooner.
Duke argues that teams should identify measures correlated with the result they ultimately care about. Objectives and key results can serve this purpose when they represent meaningful progress rather than convenient activity.
Poker illustrates the difficulty. Although every hand ends quickly, the feedback is noisy. A player may win after making a poor choice or lose after making the statistically stronger one. Speed alone does not make feedback informative.
The practical task is to define what a high-quality decision looked like before learning the result. Did the process use relevant information? Were probabilities estimated honestly? Were alternatives considered? Was the decision consistent with the organization’s goals?
Duke cites Kahneman’s work on structured hiring as evidence that process changes can matter substantially. In the example discussed, a more systematic method improved hiring accuracy from roughly 50 percent to 65 percent. The gain did not require perfect prediction—only a meaningful improvement over intuition alone.
Turn Subjective Impressions into Mediating Judgments
First Round Capital offers an organizational example of explicit reasoning. According to Duke, the venture firm began recording detailed assessments of factors such as market attractiveness, team quality, and founder potential.
Rather than relying on one global impression, the partners developed “mediating judgments”: defined components that contribute to the overall investment decision. Shared definitions help ensure that two people using the same label are evaluating something reasonably similar.
Historical outcomes can then be compared with the judgments recorded at the time. The firm can investigate which assessments correlate with company performance and whether individual partners show recurring patterns. One investor might evaluate markets well but systematically overestimate a particular founder trait.
The goal is not to remove human judgment. It is to make judgment legible enough to generate personalized feedback.
Premortems Need Kill Criteria
A premortem asks a team to imagine that a project has failed and then work backward to identify plausible causes. This exercise can uncover risks that enthusiasm or group loyalty would otherwise hide.
Duke warns, however, that identifying failure modes is not sufficient. Teams also need pre-commitments: agreements made in advance about what they will do when warning signs appear.
Kill criteria translate those agreements into observable conditions. A sales team, for example, might reconsider an opportunity if the request appears written for a competitor, the prospect will discuss only price, or no genuine decision-maker can be brought into the process.
Predefined criteria protect decisions from sunk-cost bias. Once time, money, and identity have been invested, people become skilled at inventing reasons to continue. It is easier to respond rationally when the response was chosen before the bad news arrived.
Quitting Usually Happens Later Than It Should
Duke argues that by the time people seriously wonder whether they should quit, the ideal exit point may already have passed. Uncertainty encourages delay, while sunk costs and attachment make an existing project feel more valuable simply because it belongs to us.
She points to Stewart Butterfield and the shutdown of the online game Glitch. Despite strong critical reception and committed users, the economics of acquiring enough customers did not support the hoped-for venture-scale business. Ending the project freed Butterfield and his team to pursue the opportunity that became Slack.
The lesson is not that persistence is misguided. It is that persistence has an opportunity cost. Continuing one effort consumes resources that cannot be directed toward a more promising alternative.
Better quitting begins before the crisis: define the signals that would justify stopping, agree on the response, and remember that walking away can be an active allocation decision rather than an admission of defeat.
