How to Make Better Decisions

Annie Duke spent two decades as a professional poker player before becoming one of the most cited voices in decision science. Her core insight is deceptively simple: the quality of a decision and the quality of its outcome are different things. A surgeon who follows every protocol perfectly can still lose a patient. A reckless driver who runs red lights can still arrive safely. If you judge decisions solely by their outcomes — what Duke calls 'resulting' — you systematically learn the wrong lessons.

Richard Zeckhauser, the Harvard economist, puts it differently: decisions are bets on the future, made with incomplete information. The best you can do is maximize your expected value across many decisions. Any individual outcome tells you almost nothing about whether the process was sound.

The practical implication is uncomfortable but liberating. You must separate your review process into two distinct evaluations: Was the decision process sound given what was knowable at the time? And separately, what happened? Organizations that conflate these — promoting the manager whose risky bet paid off, punishing the one whose careful analysis met bad luck — create incentive structures that reward gambling over thinking. The discipline of evaluating process independently from outcome is the foundation every other decision improvement builds on.

Jeff Bezos's most underrated contribution to management thinking isn't Amazon's leadership principles — it's his distinction between Type 1 and Type 2 decisions. Type 1 decisions are irreversible: once you walk through the door, you can't come back. Type 2 decisions are reversible: if you don't like what's on the other side, you can step back through. Bezos's observation, outlined in his 1997 shareholder letter and repeated many times since, is that most decisions are Type 2 — but most organizations treat them as Type 1.

The cost of this miscategorization is enormous. When every decision gets the full committee-review-and-approval treatment, organizations become slow, risk-averse, and exhausted. Decision fatigue sets in not because people face too many important choices, but because they apply heavyweight processes to lightweight decisions.

Dwight Eisenhower used a similar framework — his famous matrix sorted decisions by urgency and importance. The key insight from both frameworks is the same: before you decide how to decide, classify the decision. A hiring decision for a key role (irreversible, high-impact) demands a different process than choosing which analytics vendor to pilot (reversible, moderate-impact). Speed is a feature for Type 2 decisions. Deliberation is a feature for Type 1. Applying the wrong speed to the wrong type is where most decision failures begin.

Charlie Munger has spent sixty years arguing that the biggest thinking errors come from people who know only one discipline. His remedy — a 'latticework of mental models' drawn from multiple fields — remains the most practical framework for improving decision quality. Each model is a lens that reveals something the others miss.

Inversion, borrowed from the mathematician Carl Jacobi, flips the question: instead of asking how to succeed, ask what would guarantee failure, then avoid those things. When the engineering team at SpaceX evaluates a new rocket design, they don't just model the best case — they invert, asking which single failure points would be catastrophic, and design those out first.

Second-order thinking, which Howard Marks of Oaktree Capital has written about extensively, forces you past the obvious. First-order thinking says 'this company has great earnings, buy the stock.' Second-order thinking says 'everyone sees the great earnings, so the stock is probably already expensive.' The pre-mortem, developed by psychologist Gary Klein, asks the team to imagine the project has failed spectacularly, then work backward to identify the causes. Klein's research found that pre-mortems increase the ability to identify reasons for future outcomes by 30%. The point is not to pick a single model but to rotate through several, the way a doctor runs multiple diagnostic tests rather than relying on one.

Daniel Kahneman won a Nobel Prize for demonstrating that human judgment is riddled with systematic errors. These aren't random — they're predictable, which means they can be counteracted. The challenge is that you cannot simply will yourself to be unbiased. As Kahneman himself has admitted, decades of studying biases haven't made him immune to them.

Confirmation bias — seeking evidence that supports what you already believe — is the most destructive for decision-making. Philip Tetlock's research on superforecasters, documented in his book 'Superforecasting,' found that the best predictors actively hunt for disconfirming evidence. They treat their beliefs as hypotheses to be tested, not positions to be defended.

Anchoring, another well-documented bias, causes you to over-weight the first piece of information you encounter. A hiring manager who sees a candidate's salary history before evaluating their skills will anchor to that number. The fix is structural: remove the anchor from the process entirely, the way orchestras use blind auditions to eliminate gender bias.

Base rate neglect — ignoring the statistical frequency of an event in favor of a vivid anecdote — can be counteracted by asking a simple question before every forecast: 'How often does this actually happen?' Kahneman calls this 'taking the outside view,' and it consistently outperforms gut-feel estimates based on the specific case in front of you.

Michael Mauboussin, the investment strategist and author of 'Think Twice,' calls decision journals the most underused tool in business. The concept is straightforward: before making an important decision, write down what you're deciding, your reasoning, what you expect to happen, your confidence level, and what evidence would change your mind. Then revisit the entry once the outcome is known.

The power of the journal isn't in any single entry — it's in the pattern recognition that emerges across dozens of entries. You might discover that you consistently overestimate your ability to predict timelines, or that you're poorly calibrated on technical decisions but well-calibrated on people decisions. These patterns are invisible to introspection because memory is reconstructive — you unconsciously edit your past reasoning to match what actually happened. The journal defeats this by freezing your thinking at the moment of decision.

Ray Dalio's Bridgewater Associates formalized a version of this process at the organizational level, tracking predictions alongside outcomes and using the results to build 'believability-weighted' decision-making — where people with better track records in a domain get more weight in decisions about that domain. You don't need Bridgewater's infrastructure. A notebook or spreadsheet, reviewed quarterly, will surface the same calibration insights. The key requirement is honesty at the moment of writing and discipline in the review.

The biggest obstacle to good decisions isn't information or analysis — it's identity. Once a person publicly commits to a position, their ego becomes fused with the decision, and defending the decision becomes indistinguishable from defending the self. Robert Cialdini's research on commitment and consistency, published in 'Influence,' demonstrates how powerfully this bias operates: people will escalate commitment to a failing course of action simply because reversing course would mean admitting they were wrong.

The best decision architectures build structural separation between the decision and the decider's identity. Gary Klein's pre-mortem works because it gives people permission to voice concerns before commitment locks in. Red teaming — assigning someone to argue the strongest possible case against the proposed direction — works because it makes dissent a role rather than an act of insubordination. Intel used a practice called 'constructive confrontation,' where vigorous debate was expected and rewarded, separating the critique of an idea from criticism of the person who proposed it.

Bezos's 'disagree and commit' principle addresses the aftermath: once a decision is made, even dissenters align fully behind it. This only works because the disagree phase was genuine — people had real space to challenge the decision. Without that, 'disagree and commit' degrades into 'comply and resent.' The structure matters more than the intention.

John Boyd's OODA loop — Observe, Orient, Decide, Act — was developed to explain why American F-86 pilots dominated superior Soviet MiG-15s in the Korean War. The F-86 had a hydraulic flight control system and a bubble canopy that gave pilots faster orientation. Boyd's insight was that the pilots who cycled through the decision loop faster won, even when their aircraft was technically inferior.

The same principle applies far beyond aerial combat. Patrick Collison, co-founder of Stripe, keeps a list of ambitious projects completed on astonishingly fast timelines — the Empire State Building in 410 days, the first iPhone in 30 months — to counter the assumption that important work requires slow deliberation. Amazon's bias for action, codified as a leadership principle, captures the same idea: most decisions should be made with about 70% of the information you wish you had.

The risk of moving slowly is real but rarely measured. Every week spent deliberating on a reversible decision is a week of learning you didn't get from actually trying. Reid Hoffman's 'if you're not embarrassed by the first version, you launched too late' is an application of this to product decisions. The goal isn't recklessness — it's matching decision speed to decision stakes, and erring on the side of action when the stakes are low and the learning is high.