How to decide

How to decide

Author

Annie Duke

Year
2020
image

Review

Annie Duke's books on decision making deserve attention from Product Managers. Several techniques she discusses directly apply to product development, such as pre-mortems and backcasting. This book also offers valuable indirect lessons on making bets, approaching discovery, managing stakeholders, and improving decision-making skills …you just need to read between the lines.

You Might Also Like:

image

Key Takeaways

The 20% that gave me 80% of the value.

The fundamental problem with human decision-making isn't that we make bad choices but that we can't distinguish good decisions from bad ones. We judge decisions by their outcomes, a mental shortcut called "resulting" that corrupts our ability to learn. When a risky bet succeeds, we declare ourselves brilliant; when a careful plan fails, we question our judgement. This backwards reasoning leads us to repeat weak processes that happened to succeed while abandoning strong ones that got unlucky.

Outcomes cast shadows over the decisions that preceded them. A startup founder who bets everything on an untested product becomes either visionary or reckless depending solely on customer response. Yet the decision quality was identical in both cases. Only luck differed. This creates a learning disability: we cannot improve what we cannot accurately evaluate.

The problem compounds through hindsight bias, our tendency to see outcomes as inevitable after they occur. Memory creep happens when knowledge gained after the outcome seeps into our recollection of what we knew at decision time. Our minds don't time-stamp beliefs, so we unconsciously rewrite history, making outcomes feel obvious. The cure requires discipline: write decision memos before outcomes arrive, documenting context, options, expected ranges, reasons, and uncertainties. After the outcome, revisit the memo to separate process quality from result quality.

Before any decision, multiple futures branch out like a tree, with thick branches representing likely outcomes and thin twigs representing long shots. After the outcome arrives, we mentally chainsaw away all other branches, leaving only what happened standing as if it were inevitable. This retrospective narrowing makes single experiences terrible teachers. To combat this, reconstruct the tree after outcomes arrive. Write down the decision and actual outcome, then add other reasonable outcomes that were possible, noting which were better or worse. The reconstructed tree should look identical whether you succeeded or failed.

Effective decision-making requires mastering three elements: preferences, payoffs, and probabilities. Preferences reflect your personal goals and values. Payoffs measure how far outcomes move you toward or away from those goals, accounting for magnitude rather than just counting pros and cons. A single catastrophic downside can outweigh numerous small benefits. Probabilities express how likely different outcomes are. Without them, you'll over-credit lucky wins and over-blame unlucky losses.

Quality choices follow six disciplined steps organised around the Three Ps: Preferences, Payoffs, Probabilities.

  1. Map the plausible outcome tree for each option (not just best/worst).
  2. Order outcomes by preferences grounded in your values and goals
  3. Attach probabilities (even rough ones).
  4. Quantify payoffs: how far each outcome moves you toward or away from goals (magnitude and direction)
  5. Repeat for competing options on the same payoff dimension to enable apples-to-apples comparison.
  6. Choose the option with the most favourable likelihood-weighted payoff profile.

Natural language terms like "likely" or "rarely" are dangerously ambiguous. One person's "fair chance" might be 30% while another's is 70%. This fuzziness hides disagreement and blocks useful feedback. Instead, put numbers on beliefs. Add ranges to show uncertainty with a bull's-eye estimate plus reasonable bounds. Treat all estimates as educated guesses where you score points for getting close, not just for perfect accuracy.

The outside view provides powerful perspective by seeing your situation as others would rather than from your own position. People excel at solving others' problems while fumbling their own because the inside view corrupts judgement through confirmation bias and overconfidence. Base rates offer the quickest path to the outside view. Newlyweds estimate their divorce risk at nearly zero while accurately assigning 40-50% to strangers. If most restaurants fail within three years, your "this time is different" estimate should still orbit that statistic unless you have credible reasons to deviate.

Not all decisions deserve equal attention. Analysis paralysis wastes time on trivial choices while rushed thinking damages high-stakes calls. The time-accuracy trade-off governs every decision: increasing accuracy costs time, while deciding quickly costs accuracy. Reserve careful work for options with large potential losses. If a decision won't meaningfully affect your happiness in a week, month, or year, go fast.

Two-way-door decisions that are reversible with low quit costs invite faster choices and experimentation. One-way-door decisions that are irreversible with high quit costs merit slower work. Before committing, explicitly ask what it would cost to quit. Improve high-stakes calls by preceding them with low-impact trials like renting before buying.

Positive thinking helps set destinations but fails at route planning. The behaviour gap between intentions and execution closes through negative thinking that imagines obstacles and failure modes. Pre-mortems generate failure reasons split between factors you control and luck you don't. Pre-commitment contracts translate foresight into behaviour by creating friction for failure behaviours and removing friction for success behaviours.

Quality feedback requires quality inputs. Provide evaluators with relevant goals, constraints, and uncertainties rather than spinning narratives. Groups amplify biases through contagion while suppressing unique data. Combat groupthink through independent idea generation before discussion. Decision records preserve your state of knowledge at choice time, creating a laboratory for improving judgement.

The ultimate goal isn't perfection but portfolio improvement. Luck and incomplete information guarantee some bad outcomes even from excellent processes. A good decision wins more often across many iterations, not one that guarantees a single win. By separating process from outcome, documenting decisions before results arrive, considering alternative outcomes, and balancing inside and outside views, we can escape the trap of resulting and actually learn from experience.

image

Deep Summary

Longer form notes, typically condensed, reworded and de-duplicated.

Introduction

Decisions deserve a process that both improves quality and triages importance, because you face thousands of choices daily (some consequential, most trivial).

Life outcomes are driven by two forces: luck and decision quality, and you control only the latter, so building a high-quality decision process is the most reliable lever you have.

Most people lack a shared language or method for good decisions; education underemphasises decision-making, and even leaders default to gut instinct or pros-and-cons lists.

Good decision tools are repeatable, teachable, and auditable; your "gut" fails all three tests, and pros-and-cons lists often amplify cognitive bias rather than reduce it.

Because any decision is a prediction about multiple possible futures, the unattainable ideal is a crystal ball; the practical substitute is a structured process that improves belief accuracy, option comparison, and outcome forecasting.

This book supplies that structure: first, it shows how to learn from experience without being misled by luck, resulting, or hindsight bias.

Next, it builds a full framework for new choices (clarifying goals, enumerating options, estimating payoffs and probabilities) and later teaches when to streamline via the time-accuracy trade-off.

Finally, it helps you anticipate obstacles and leverage others' knowledge through decision hygiene and group practices that surface independent, uninfected feedback and avoid groupthink.

Chapter 1: Resulting

Resulting is a mental shortcut: judging a decision by its outcome. Because outcomes are vivid and easy to see while our decision making process is comparatively opaque, we let results stand in for decision making process quality. Decision quality and outcome quality are correlated but only loosely in the short run. Outcomes drive how we interpret the previous decisions. A good result makes the process look smarter; a bad result makes the process look sloppy. This "shadow" leads us to overfit process judgements to endings and to learn the wrong lessons.

Every decision creates a range of possible paths. We only observe one path after the fact. Forgetting the unseen alternatives makes us overconfident about judging process from a single outcome.

Luck is the element outside your control that determines which of the possible outcomes you actually get. Because luck intervenes, good outcomes can follow bad processes and vice versa. Here's a helpful classification:

  • Earned Reward: good process → good outcome.
  • Dumb Luck: bad process → good outcome.
  • Bad Luck: good process → bad outcome.
  • Just Deserts: bad process → bad outcome.

We tend to notice Bad Luck (it preserves our self-image when things go wrong) and overlook Dumb Luck (it would force us to surrender credit when things go right). That asymmetry stalls learning.

Our ability to learn from our decisions requires us to separate process from result: If we overfit outcomes, we can..(1) repeat weak processes that happened to win, (2) abandon strong processes that got unlucky, and (3) skip examining wins and losses that still contain teachable details.

To separate outcome quality from decision quality: Ask, "What were the realistic alternatives and their probabilities?" Record the information, assumptions, and advice you used before outcomes arrive. Identify factors outside your control, including other people's actions and timing. Compare the single observed outcome with other plausible ways things could have turned out. Evaluate decisions as part of a portfolio (your slate), not one-offs. Examine wins for flaws and losses for strengths with equal rigour. Practice compassion: judge processes fairly in yourself and others; don't equate self-worth with outcomes.

One outcome is thin evidence. You can't tell much about process quality from a single outcome. Choosing well is to pick the option with the best distribution of outcomes, not to guarantee a win.

Resulting disrupts learning by both repeating low-quality processes and abandoning high-quality ones. Keep examining good/good and bad/bad pairings; they also hold lessons. Resulting can reduce compassion toward others and yourself; resist that pull.

Resulting Checklist:

  • Is the outcome biasing your view of the decision quality?
  • For bad outcomes: Can you identify good decisions that were made?
  • What aspects of the decision process were effective?
  • For good outcomes: How could the decision have been better?
  • How could the decision process be improved?
  • What factors were outside the decision-maker's control?
  • What alternative outcomes were possible?

Chapter 2: Hindsight is not 20/20

Hindsight bias is the tendency to see outcomes as predictable or inevitable after they occur. It is dangerous because it distorts what you remember knowing at decision time, inflates perceived predictability, and poisons learning. You end up repeating weak processes that previously "worked," abandoning strong processes that got unlucky, and judging yourself and others without compassion.

Memory creep is when post‑outcome knowledge seeps into your memory of pre‑decision knowledge. Because our minds don't time‑stamp beliefs, we unconsciously rewrite the past, making the outcome feel obvious. This corrupts feedback loops and leads to false lessons.

Hindsight bias narrows perceived uncertainty, masks the role of luck, and overfits conclusions to a single realised path. It encourages harsh "I told you so" or "I should have known" narratives, which undermine honest reviews and future judgement under uncertainty

Listen for: "I knew it," "I told you so," "I should have known," "How could you not see that coming?" Treat these as alarms to pause and reconstruct what was actually knowable then.

Write brief decision memos before outcomes arrive: date, context, options, expected ranges, reasons, what you don't know, and what would change your mind. After the outcome, revisit the same memo with the Knowledge Tracker to disentangle process quality from result quality and recalibrate beliefs without rewriting history.

"I knew it" narratives often only surface after results are known. Treat confident after‑the‑fact stories with scepticism.

Checklist to identify and address hindsight bias:

  1. What information was revealed only after the fact?
  2. Was that information reasonably knowable at the time? (Consult your journal.)
  3. Is your conclusion about predictability relying on unknowable ex‑ante facts?
  4. Reassess how predictable the outcome truly was, and then evaluate the process, not the result.

Apply empathy to yourself and others by anchoring judgements in what was reasonable to know then. This preserves motivation, encourages honest review, and improves future decisions under uncertainty.

Chapter 3: The Decision Multiverse

The paradox of experience. Experience is necessary for learning, yet single experiences often don't teach us much about the quality of our decision making process, they can even mislead us. We process outcomes sequentially and overfit judgements to one realised path.

Before deciding, the future looks like a tree of branching possibilities. Thick branches represent likelier outcomes; thin twigs, long shots. After the outcome arrives, we tend to mentally "chainsaw" the tree, leaving only the branch that happened. The past then feels inevitable, even when it wasn't.

Reassemble the tree after the fact. Write the decision and the actual outcome, then add other reasonable outcomes that were possible at the time. Note which were better, which were worse, and which were adjacent. This shrinks the single outcome to its proper size, restores context, and improves learning. The tree you reconstruct should look the same whether you happened to succeed or fail, because the decision was made under uncertainty and laid out set the menu of futures.

A counterfactual is a "what-if": any plausible outcome that did not occur, or an imagined state of the world under the same decision. Counterfactual thinking clarifies luck's role, loosens the feeling of inevitability, allows fair comparison between the observed result and alternatives, and refines what to repeat or change next time.

We more readily contextualise bad outcomes to relieve self-blame, yet resist contextualising good outcomes to preserve credit. This asymmetry hides lucky wins, misses chances to improve strong results, and sustains overconfidence. Aim for symmetry: analyse successes and failures with the same rigour.

When evaluating whether the outcome provides a lesson about decision quality, create a simplified decision tree, starting with the following:

  • Identify the decision.
  • Identify the actual outcome.
  • Along with the actual outcome, create a tree with other reasonable outcomes that were possible at the time of the decision.
  • Explore the other possible outcomes to understand better what is to be learned from the actual outcome you got.

Chapter 4: The Three Ps: Preferences, Payoffs and Probabilities

Six Steps to Better Decisions:

  1. Map the decision tree of plausible outcomes (not just best/worst).
  2. For each outcome, given your values and goals, identify your preferences (gains, losses).
  3. Assign a probability (words or %) to each outcome
  4. Weigh upside vs downside by likelihood.
  5. Repeat for other options. Build comparable trees (same payoff dimension) for each alternative.
  6. Compare options. Choose the option with the most favourable mix of payoffs and probabilities.

Preferences are personal. Your goals and values help you determine what outcomes are more desirable (money, time, health, relationships, self-esteem, others' well-being). Make them explicit when seeking advice or comparing options; two people can rationally prefer different outcomes.

Take into account the payoff size. Pros/cons lists are flat: they ignore magnitude. Payoffs measure how far an outcome moves you toward/away from goals. Big downside can dominate many small upsides; many small losses can outweigh a rare windfall. Always ask, "How big is the gain or loss?"

  • Upside: the potential gains you value.
  • Downside: the potential costs you wish to avoid.
  • Risk: your exposure to the downside.

A good decision trades for upside only when its likelihood-weighted benefits justify the downside risk.

Your choice is always an estimate about how likely different outcomes are. Without likelihoods, you'll over-credit lucky wins and over-blame unlucky losses. Put a % likelihood against outcomes.

The Archer's Mindset (guessing and showing your work). There isn't only right/wrong; like archery, you score for getting close. All guesses are educated to some degree. Show your work: what you already know, what you infer, what remains unknown. Small accuracy gains compound across many decisions. Ask:

  1. What do I already know that narrows the range?
  2. What can I find out (quick research, expert input, historical rates, small tests) to narrow it further?

Deliberately move items from "don't know" to "know": clarify assumptions, collect data, run a probe, or pre-mortem/post-mortem your tree. Each bit reduces error and strengthens your decision's foundation.

Two forms of uncertainty:

  • Imperfect information (before the decision): you can reduce it by learning and testing.
  • Luck (after the decision, before the outcome): you can't control it on a single try; you can only choose options that win more often across many tries.

If one payoff dimension dominates (e.g., retention, health adherence, money), frame branches on that axis to enable clean, apples-to-apples comparisons in Steps 5–6.

Remember the three P's:

  • Preferences: driven by your goals/values; order outcomes accordingly.
  • Payoffs: quantify movement toward/away from goals; upside vs downside; risk is exposure to the downside; magnitude matters.
  • Probabilities: express likelihoods at a %; assess relative likelihood of liked vs disliked outcomes; compare options on the same payoff dimension.

Chapter 5: The power of precision

Natural‑language terms for probability (likely, rarely, always) are blunt and ambiguous, misleading decision makers when people map the same word to very different percentages. This fuzziness hides disagreement, blocks useful feedback, and lowers decision quality. High‑stakes failures can stem from this gap when a decision maker hears "fair chance" and assumes a much higher probability than the speaker intended.

Precision (putting numbers on beliefs) reveals disagreement and invites correction. Precision increases accountability, motivates you to refine estimates, and helps others give targeted, corrective information that improves your beliefs.

Apply % likelihood to your decision trees so options can be compared cleanly. When branches are mutually exclusive, ensure assigned probabilities do not exceed 100%; totals may be <100% because you're listing reasonable (not exhaustive) outcomes.

Add ranges to show uncertainty by giving a bull's‑eye estimate and a reasonable lower/upper bound. The range communicates how uncertain you are and signals where help can narrow it. Wide ranges are fine when knowledge is thin; they're honest and useful.

Use the Shock Test by choosing the narrowest range such that you'd be pretty shocked if the truth fell outside it. Aim for 90% of true values landing within your ranges over time. Most of us are overconfident; missing often is a calibration cue, not a failure.

Fight overconfidence with disciplined scepticism by asking:

  • If I'm wrong, why would that be?
  • What information would change my mind? Then go look for it, or stay alert to it.

Treat estimates as provisional and improvable.

Chapter 6: The Outside view

People are better at solving other people's problems than their own because they see from the outside view while the sufferer is stuck in the inside view. When you're the friend hearing endless dating disasters, it's easy to suspect causes beyond "bad luck"; when you're the protagonist, self‑protection blinds you to patterns you could change.

The inside view is judgment from your own perspective and it corrupts inputs to decisions. It fuels confirmation bias, disconfirmation bias, overconfidence, availability and recency biases, and the illusion of control making single outcomes loom large and mis-teach.

Pros‑and‑cons lists amplify the inside view by letting motivation steer what makes the page and how it's weighted. If you want an option, you pad the "pros"; if you don't, you pad the "cons" a biased tool masquerading as analysis.

The outside view is what's true independent of you, and it disciplines distorted intuition. Even with the same facts, other people often reach different conclusions; deliberately inviting those perspectives reduces junk in your decision process.

Base rates expose how inside‑view optimism collides with population reality. Newlyweds say divorce risk is ~0% for them while assigning ~40–50% to strangers; that asymmetry mirrors the better‑than‑average effect, where most people rate themselves above average and underprepare (e.g., few prenups despite high divorce rates).

Accuracy lives at the intersection of outside and inside views, starting with the outside anchor. Begin with what's true in general, then adjust for particulars; this keeps forecasts realistic without ignoring context.

Being smart often worsens motivated reasoning and the bias blind spot.

Base rates are the quickest path to the outside view and should act as a centre of gravity for forecasts. If gyms retain few new members or most restaurants fail early, your "this time is different" estimate should still orbit those statistics unless you can specify credible, causal reasons to deviate.

Actively seeking disagreement is key. Make dissent safe; precise, candid feedback prevents echo chambers that merely repackage the inside view as "objective."

Perspective Tracking turns outside‑in thinking into habit by journalling both views. Write a brief outside‑view pass (base rates + external feedback) and a separate inside‑view pass (your context, constraints, goals), then revisit after outcomes to improve learning symmetry crediting luck in wins and skill‑gaps in losses.

Embracing the discomfort of the outside view trades short‑term ego protection for long‑term decision quality.

The one page outside‑in method: Describe your situation entirely from the outside view (relevant base rates plus other people's perspectives stated in percentages where possible); then, describe it entirely from the inside view (your goals, constraints, edge cases); third, reconcile the two into an integrated forecast and plan, replacing pros‑and‑cons with a calibrated decision tree and clear assumptions; finally, invite explicit disagreement and record changes to beliefs for future calibration.

Chapter 7: Breaking free from analysis paralysis

Analysis paralysis wastes scarce time on low-impact choices. Don't spend too long on trivialities.

The time-accuracy trade-off governs every decision: increasing accuracy costs time, and deciding quickly costs accuracy. Aim to get the balance right, by spending more time only when the penalty for being less accurate is high.

Map possibilities, payoffs, and probabilities then reserve slow, careful work for options with large potential losses. If a decision's outcome won't meaningfully affect your happiness in a week, a month, or a year, you can go fast because the penalty for being "less right" is tiny.

Quick decision making is justified if you'll get another go soon; frequent, low-risk choices are ideal for cheap experiments.

Ask 'What's the worst that can happen?' to determine if it's a freeroll, seize the opportunity quickly, then take your time on execution. Freeroll traps come from repetition and accumulation: a "free donut" or weekly lottery ticket looks low-cost once, but repeated small losses compound into meaningful downside.

If options are genuinely similar in upside and downside, you can't be very wrong either way, when a decision is hard, that means it's easy.

Allocate decision making effort where it pays: spend time sorting (defining acceptable options by your goals/constraints), then save time picking among the acceptable set, because extra picking time rarely improves accuracy much.

Two-way-door decisions (reversible, low quit cost) invite faster choices and deliberate experimentation, while one-way-door decisions (irreversible, high quit cost) merit slower work; before committing, ask explicitly, "What would it cost to quit?"

Improve high-stakes calls by preceding them with low-impact, easy-to-quit trials (e.g. rent before buying, date widely before committing), turning unknowns into knowns at low cost.

Choosing options in parallel (when feasible) accelerates learning and shares risk, just mind resource limits and execution quality.

A clear stopping rule prevents endless analysis: once you have a "good enough" option, ask "Is there information that would change my mind, and can I get it at reasonable cost?"; if yes, fetch it, and if no (or too costly), decide and stop.

Satisficing beats maximising under uncertainty: "good enough" choices grounded in impact, reversibility, and opportunity cost free time for higher-leverage sorting and learning

Chapter 8: The Power of Negative Thinking

Negative thinking closes the behaviour gap: the gulf between what we intend to do and what we actually execute, by planning for obstacles, not just outcomes.

Positive visualisation is useful for setting destinations, but route planning requires imagining failure modes. Adopt a "Waze-style" planning for roadblocks makes success more reliable.

Techniques:

  • Mental contrasting: stating the goal and then listing barriers → improves execution across domains because it forces you to anticipate where skill or luck can block progress.
  • Mental time travel combats status quo bias by moving you to a future vantage point; from the "summit," you can see alternate routes and changing conditions that are invisible from the base.
  • Prospective hindsight: imagining it's after the event and you failed, then look back and explain why.
  • Premortem: generate reasons for failure split into skill (in your control) and luck (outside your control)
  • Backcasting: imagine you succeeded and work backward to the skillful actions and lucky breaks that produced the win.
  • Precommitment (Ulysses) contracts: translate foresight into behaviour: physically prevent bad choices, raise barriers (increase friction) to likely failure behaviours, and lower barriers (reduce friction) for success behaviours; accountability to others amplifies the effect.
  • Dr. Evil game: list small, easily justified choices that are harmless in isolation but guarantee failure in aggregate; once spotted, treat them as category decisions ("I don't do X") or elevate them for explicit deliberation.

Dealing with tilt: the hot emotional state after surprising outcomes that degrades judgement. Take a tilt inventory of your cues plus time-travel prompts ("Will I endorse this in a week?") and make prewritten rules for quitting or pausing reduce damage.

Chapter 9: Decision Hygiene

Divergence is where learning lives: mapping where your beliefs and someone else's part ways yields corrective information. The truth may lie between.

Ask "What should I do?" and iterate past events stepwise without revealing results to block resulting and hindsight bias. Neutral framing prevents signalling; avoid leading questions.

Groups amplify contagion and suppress unique data. Independent idea generation is essential to counter groupthink and herding.

Anonymising the first pass blunts status and the halo effect, giving low-status or contrarian perspectives space.

"But why?" questioning is a low-tech truth serum: repeated, genuine requests for explanation expose knowledge gaps, force clarification, and transfer expertise without intimidation.

Input quality governs feedback quality, so stop spinning narratives; use the outside view to prebuild a checklist of relevant goals, constraints, and situational facts for recurring decisions and provide exactly what evaluators need: no more, no less. If required facts are missing, refuse to give advice; doing so trains attention to critical variables and prevents confident but content-free guidance.

Decision records preserve your state of knowledge at choice time, making later feedback de-biased and comparable to what you actually knew then.

The right objective is portfolio improvement, not perfect hits: luck and incomplete information guarantee some bad outcomes, so abandon the defensive crouch of self-protection and practice real self-compassion by seeking divergence, scrubbing bias, and iterating toward higher decision quality.