Annie Duke’s Thinking in Bets: Making Smarter Decisions When You Don’t Have All the Facts (Portfolio, 2018) starts from a distinction internal auditors deal with constantly and rarely name: the quality of a decision and the quality of its outcome are different things. Because the world contains luck and hidden information, a sound decision can turn out badly and a reckless one can turn out well, and people who judge decisions by their results, which poker players call “resulting”, learn the wrong lessons from both. Duke, a former professional poker player with training in cognitive psychology, argues that the remedy is to treat every decision as a bet on an uncertain future: to say how confident you are rather than whether you are right, to separate the part of an outcome that skill explains from the part that luck explains, to recruit other people who will hold you to accuracy rather than agree with you, and to use techniques that bring your future self into today’s choice. For internal auditors the book is uncomfortable in a useful way. Much of audit work is judging decisions after the fact: root cause analysis, post-implementation reviews, incident investigations and the evaluation of management’s risk acceptances all take place when the outcome is already known, which is when resulting and hindsight bias have most room to operate. This guide summarises the book without reproducing it, sets out its structure from the printed contents, explains its main ideas and vocabulary, checks its best-known example against the record, records what reviewers have criticised, maps the ideas to audit practice and ends with a worked example in which a food manufacturer’s one-person audit function reviews how the company judges its own decisions.
In this guide
- The book in one page
- Who Annie Duke is, and how the book came about
- The structure: an introduction and six chapters
- Resulting: a bad outcome is not a bad decision
- Beliefs as bets, and learning from outcomes
- Truthseeking groups and the norms of dissent
- Mental time travel: the future self as a check
- The nine ideas worth keeping
- The vocabulary: thirteen terms defined for practitioners
- What it means for internal audit: judging decisions after the fact
- The criticisms, and what the book leaves out
- Editions, companions and how to read it
- A one-hour discussion guide for an audit team
- Worked example: Brightwater Foods reviews its decisions, not its outcomes
- Frequently asked questions
- Related guides
The book in one page
| Question | The answer, in our words |
|---|---|
| What is the thesis? | That a decision is a bet on an uncertain future, that its quality must be judged by the information, options and reasoning available when it was made rather than by how it turned out, and that people and organisations who judge by outcomes learn badly. |
| What is resulting? | The poker players’ term for equating the quality of a decision with the quality of its outcome. A good result is taken as proof of a good decision and a bad result as proof of a bad one, although luck and hidden information break the link. |
| Why poker rather than chess? | Chess has no hidden information and little luck, so a loss usually means a mistake. Poker, like most business and life decisions, involves both, so a loss may mean nothing about the quality of play. Duke argues that life is the second kind of game. |
| What does thinking in bets mean in practice? | Asking how sure you are and expressing it as a probability or a range; asking what outcomes are possible; and asking, after the fact, how much of the result came from skill and how much from luck. |
| How do people improve? | Not alone. Duke recommends a small group committed to accuracy, accountability and openness to other views, run on norms she draws from the sociologist Robert K. Merton’s account of how science works, so that dissent is expected rather than punished. |
| What about emotion and time? | Decisions made in the moment favour the present self. Techniques such as asking how a choice will look in ten minutes, ten months and ten years, imagining success and working backwards, imagining failure and asking why, and committing in advance to rules, bring the future into the present decision. |
| Who wrote it? | A former professional poker player who won a World Series of Poker bracelet in 2004, retired from the game in 2012 and now works on decision-making; she studied cognitive psychology at the University of Pennsylvania and completed her doctorate there in 2023. |
| Why read it as an auditor? | Because audit judges decisions with the outcome in hand, because the function’s own ratings and conclusions are bets made with incomplete information, and because the book’s group norms read like a description of a good quality review. |
Who Annie Duke is, and how the book came about
Duke received a National Science Foundation fellowship to study cognitive psychology at the University of Pennsylvania before she turned to professional poker, according to her publisher, and she earned a master’s degree there. She won her first World Series of Poker bracelet in 2004 and, in the same year, the World Series of Poker Tournament of Champions, which her Wharton executive education biography describes as a $2 million invitation-only event; in 2010 she won the NBC National Heads-Up Poker Championship. Her publisher records more than $4 million of tournament winnings before she retired from the game in 2012. Since then she has worked as a consultant, speaker and author on decision-making. She co-founded the Alliance for Decision Education, a nonprofit that promotes the teaching of decision skills in schools, in 2014; she returned to Penn as a visiting scholar in 2021 and completed her PhD in cognitive psychology in 2023, and she is a special partner for decision science at the venture firm First Round Capital.
Thinking in Bets was published by Portfolio, an imprint of Penguin Random House, on 6 February 2018, and the publisher’s description presents it as the work of a poker champion turned business consultant, using examples from business, sports, politics and poker. When it came out, Duke discussed it on podcasts, including an episode of Capital Allocators, an investment podcast, whose show notes list Merton’s norms of science among the topics. Two later books extend it. How to Decide: Simple Tools for Making Better Choices (October 2020) grew, she told Knowledge at Wharton, out of readers of the first book asking how to put it into practice, and offers exercises and checklists; Quit: The Power of Knowing When to Walk Away followed in October 2022. The paperback of Thinking in Bets appeared in May 2019, and we found no revised edition.
The structure: an introduction and six chapters
The printed contents list an introduction and six chapters, followed by notes and an index. The table follows that list; the descriptions are ours, drawn from a reader’s chapter-by-chapter notes, and we place a named example in a chapter only where a source does.
| Chapter | What it covers | What it contributes |
|---|---|---|
| Introduction: Why This Isn’t a Poker Book | Decisions as bets on the future; the two things that determine how a life turns out, the quality of decisions and luck | The frame, and the claim that the lessons are not about cards |
| 1 Life Is Poker, Not Chess | Resulting; why chess is the wrong model for real decisions, which involve hidden information and luck; the value of admitting uncertainty | The central distinction between decision quality and outcome quality |
| 2 Wanna Bet? | How beliefs are formed easily and doubted rarely; why intelligence can make it easier to justify a belief rather than test it; treating beliefs as bets and stating confidence as a percentage or a range | Calibration of belief as a habit |
| 3 Bet to Learn: Fielding the Unfolding Future | Why learning from outcomes is hard; deciding whether a result came from skill or luck; changing what we feel rewarded for, such as crediting luck and finding mistakes in good outcomes; learning from other people’s results | A method for learning from experience without being fooled by it |
| 4 The Buddy System | A small group committed to accuracy, accountability and openness to diverse views; why people are more candid when they are betting on being right than when they are trying to get along | The case for a deliberate truthseeking group |
| 5 Dissent to Win | Looking for reasons you might be wrong; sharing all the data, including the unflattering; judging an idea apart from its source; reviewing decisions before the outcome is known; communicating disagreement constructively | The group norms, framed by Merton’s norms of science |
| 6 Adventures in Mental Time Travel | Favouring the present self; asking how a choice looks over ten minutes, ten months and ten years; mapping possible futures with probabilities; backcasting and premortems; seeing life as a long game | Techniques for bringing the future self into the decision |
Resulting: a bad outcome is not a bad decision
The example several reviewers single out is a football play. In Super Bowl XLIX, on 1 February 2015 in Glendale, Arizona, the Seattle Seahawks trailed the New England Patriots 28 to 24 with 26 seconds left and the ball on second down at the Patriots’ one-yard line. Instead of handing the ball to their running back, the Seahawks’ coaches called a pass; the Patriots’ rookie cornerback Malcolm Butler intercepted it and New England won. A public radio headline the next day quoted the verdict that it was the worst call in Super Bowl history, and the coach, Pete Carroll, told his players that the fault was entirely his while explaining that the team had time left and had been playing for third and fourth down. Duke uses the episode early in the book to show resulting at work: the call was condemned because of how it ended. The contemporary record shows why the case is contested. Statistical analysts argued at the time that the pass was a defensible call, among them a FiveThirtyEight analysis reported on a University of Puget Sound economics blog. Whether the call was right can be argued either way; the point the book takes from it is that the verdict followed the result.
The chess comparison makes the same point in general form. Duke retells, from Jacob Bronowski’s The Ascent of Man, the story of John von Neumann saying that chess is not a game in his sense, because real life consists of bluffing and deception, which is why his game theory took poker as its model. In chess, with all information visible and little luck, a loss usually reveals a mistake. In poker, and in most decisions that matter, a player can make the right decision and lose, or the wrong one and win, and the result alone cannot tell which happened. The table is our framing of the distinction in audit terms.
| Decision quality | Outcome | What resulting concludes | What a decision review asks | Where an auditor meets it |
|---|---|---|---|---|
| Sound | Good | Skill, correctly, but for the wrong reason | Was the success expected, or did luck help more than the plan? | A project delivered on time that relied on a supplier nobody had assessed |
| Sound | Bad | A mistake; someone to blame | Given what was known, were the options, probabilities and limits reasonable? | A hedge that lost money on its own while doing what the policy required |
| Unsound | Good | Good judgment; perhaps a bonus | What would have happened if the bet had gone the other way? | A limit breach that made money and was praised |
| Unsound | Bad | A mistake, correctly, but often with the wrong cause | Which part of the process failed, and was it the person or the system? | An incident blamed on an individual whose procedure could not be followed |
The table’s second and third rows are where the cost lies. An organisation that punishes sound decisions with bad outcomes teaches its people to avoid reasonable risks, and one that rewards unsound decisions with good outcomes teaches them to take unreasonable ones. Neither lesson shows up in a single result; both accumulate. For an auditor the first practical consequence is about evidence: to evaluate a decision, the function needs a record of what was known and expected when it was made, and many organisations do not keep one. The root cause guide and the cause statement guide describe how to reach the process cause of a failure; the book adds the warning that the outcome itself is weak evidence of where the cause lies.
Beliefs as bets, and learning from outcomes
The second chapter turns from decisions to the beliefs they rest on. Its argument, as readers’ notes record it, is that people adopt beliefs readily and examine them rarely, that being clever helps people build justifications for what they already believe more than it helps them test it, and that the question in the chapter’s title is a useful corrective: a belief that its holder would not bet on at the confidence they express is held too strongly. Duke’s practical recommendation is to state beliefs with a probability or a range rather than as true or false, which both tells others how much weight to give them and invites people with better information to say so. One reviewer connected this to the calibration work of Philip Tetlock, which the Superforecasting summary covers.
The third chapter, on learning, introduces the idea of fielding outcomes: deciding, for each result, how much was due to the quality of the decision and how much to luck. The difficulty is that people field outcomes in their own favour, crediting skill for good results and blaming luck for bad ones, the self-serving bias, while doing the reverse for other people. Duke’s examples, as one reviewer lists them, include the professional poker player Phil Ivey reviewing his own play after a tournament and the gambler known as Nick the Greek, who ignored the feedback his losses offered. Her remedy is to change what one feels rewarded for: to take satisfaction in an accurate account of one’s own errors, in crediting luck where it helped and in finding the mistakes inside good outcomes. The table sets out the biases the book treats and their audit forms.
| Bias or habit | What it does | How it appears in audit work | The book’s counter, in our words |
|---|---|---|---|
| Resulting | Judges the decision by the outcome | Findings written only where something went wrong; post-implementation reviews only of failed projects | Review decisions on their merits, including those with good outcomes |
| Hindsight bias | Makes the outcome seem predictable once known | “Management should have foreseen” in a finding condition | Reconstruct what was known at the time before writing the condition |
| Self-serving bias | Credits skill for good results and luck for bad | Business units attributing good results to strategy and bad ones to markets | Field one’s own outcomes as one would field a rival’s |
| Motivated reasoning | Interprets evidence to protect an existing belief | An auditee or an auditor defending a first view against new evidence | Ask what evidence would change the view, before collecting it |
| Overconfidence in belief | States beliefs as certain | Ratings and conclusions expressed without any sense of how sure the function is | State a confidence level; ask whether you would bet on it |
Truthseeking groups and the norms of dissent
The fourth and fifth chapters argue that individuals cannot do this alone, because the biases are hardest to see from the inside. Duke’s answer is a small group, which she describes as a truthseeking group or pod, whose members agree to value accuracy over comfort, to hold one another accountable for the reasoning behind their decisions, and to seek out views different from their own. Readers’ notes record her observation that people speak more candidly when they are betting on being right than when they are trying to get along. Two reviewers note that the book holds up Heterodox Academy, and its call for social scientists to engage with conservative viewpoints, as an example of seeking out disagreement deliberately.
The fifth chapter supplies the group’s operating rules, and Duke frames them with the norms of science that the sociologist Robert K. Merton set out in 1942, often abbreviated as CUDOS. The table gives Merton’s four norms as commonly stated, how reviewers describe Duke’s use of each, and what each looks like in an audit function. The chapter also offers advice on communicating dissent without provoking defence, as a reader’s notes summarise it: express uncertainty, lead with the points of agreement, prefer “and” to “but”, and ask whether the other person wants to vent or wants advice. The Fearless Organization summary covers the organisational conditions that make such candour possible.
| Merton’s norm | Its meaning in science | Duke’s use, as reviewers describe it | The audit equivalent |
|---|---|---|---|
| Communism (shared data) | Findings belong to the community and are shared | Share all the relevant information, including what is unflattering | Working papers and exceptions shared in full with the reviewer, not curated |
| Universalism | Claims are judged on their merits, not by who makes them | Judge an idea apart from its messenger; do not dismiss it because of its source | An auditee’s or a junior auditor’s challenge evaluated on the evidence |
| Disinterestedness | Results are not bent to serve personal interest | Recognise that interests and conflicts shape beliefs; review decisions without knowing their outcomes | Reviewers kept independent of the conclusion; outcome-blind review where possible |
| Organised scepticism | Claims are tested before they are accepted | Ask why something might not be true; build in disagreement | A reviewer whose job is to find the weakest link in a conclusion before it is reported |
For an audit function the parallel is close. The Global Internal Audit Standards, effective from 9 January 2025, require internal auditors to maintain professional skepticism, critically assessing the reliability of information (Standard 4.3), and to make balanced assessments while identifying and managing potential biases (Standard 2.1). Duke’s group norms are a practical way of meeting both requirements at the level of a team rather than an individual. The professional skepticism guide describes the individual discipline, and the review notes guide describes the review conversation where the norms are most needed.
Mental time travel: the future self as a check
The last chapter deals with time and emotion. Decisions made in the moment favour the present self over the future one, and Duke’s techniques are ways of bringing the future into the room. Readers and reviewers list them consistently: Suzy Welch’s 10-10-10 rule, which asks how a choice will look in ten minutes, ten months and ten years; mapping the possible futures of a decision with probabilities attached; backcasting, which imagines that a goal has been achieved and works out how; premortems, which imagine that it has failed and ask why; and precommitment devices, among them the Ulysses contract, which binds the future self in advance, and what she calls a decision swear jar. Reviewers also note the book’s discussion of going on tilt, decisions driven by in-the-moment emotion, and of ticker-watching, which the Chicago Tribune’s reviewer said the book illustrates by setting one day’s movement in Berkshire Hathaway’s shares in January 2017 against its long-run record.
| Technique | What it does | An audit use |
|---|---|---|
| 10-10-10 | Weighs a choice over three time horizons | Testing whether a risk acceptance has considered the longer-term consequence as well as the immediate saving |
| Scenario mapping with probabilities | Lists the plausible outcomes of a decision and assigns each a probability | Asking management to state the scenarios behind a business case, not just the expected one |
| Backcasting | Imagines success and works back to what had to go right | Identifying the conditions a project or strategy depends on, so that each can be tested |
| Premortem | Imagines failure and asks what caused it | A planning exercise for a high-risk project or a new engagement; Gary Klein described it in the Harvard Business Review in 2007 |
| Ulysses contract | Commits the future self to a rule while thinking clearly | Limits, escalation triggers and stop-loss rules agreed before the pressure arrives |
| Tilt | Names the state in which emotion is driving decisions | Pausing an investigation, a difficult closing meeting or a disputed finding until the heat has gone |
The techniques are not new: Gary Klein described the premortem in 2007, and the book attributes the 10-10-10 rule to Suzy Welch. Their value in the book is the way they connect to its central argument. A premortem done before a decision creates exactly the record of what was expected and feared that a later review needs in order to judge the decision rather than the outcome, and a precommitment written in calm conditions is the standard against which a decision made in a crisis can fairly be measured. The strategic risk assessment guide shows how premortem-style questions fit into the assessment of the largest risks, and the risk acceptance guide covers the process by which management’s bets are formally recorded.
The nine ideas worth keeping
- Decision quality and outcome quality are different. Luck and hidden information separate them; a decision must be judged by what was known and reasonable when it was made.
- Resulting corrupts learning. Judging by outcomes punishes sound decisions that turned out badly and rewards unsound ones that turned out well, and both lessons accumulate.
- Life is poker, not chess. Most consequential decisions are made with incomplete information and some luck, so a single result says little about the quality of play.
- Beliefs are bets. State them with a probability or range, and ask whether you would wager on them at the confidence you express.
- Field outcomes accurately. Decide how much of a result came from skill and how much from luck, and apply the same standard to your own results as to other people’s.
- Look for the mistakes inside good outcomes. Successes are rarely reviewed, and a success can hide a lucky escape.
- Build a truthseeking group. Accuracy, accountability and openness to other views do not happen by default; they need people who have agreed to them.
- Use Merton’s norms. Share the data, judge ideas apart from their source, recognise interests, and test claims before accepting them.
- Bring the future self into the decision. Premortems, backcasting, 10-10-10 and precommitments are cheap, and they create the record that later review needs.
The vocabulary: thirteen terms defined for practitioners
| Term | What Duke means | What it looks like in an audit |
|---|---|---|
| Resulting | Judging a decision by its outcome | A finding that exists only because the outcome was bad |
| Thinking in bets | Treating decisions and beliefs as wagers on an uncertain future | Stating management’s and the function’s judgments with a probability or a range |
| Hidden information | What a decision-maker could not know when deciding | The facts that emerged only after the event, to be set aside when judging the decision |
| Wanna bet? | The test of whether a belief is held at the confidence its holder expresses | A challenge to a confident assertion in fieldwork or review |
| Calibration of belief | Matching stated confidence to the strength of the evidence | Risk ratings that carry an explicit level of confidence |
| Fielding outcomes | Sorting a result into the part due to skill and the part due to luck | The cause analysis behind a finding or a post-implementation review |
| Self-serving bias | Crediting skill for good outcomes and luck for bad ones | Asymmetric explanations in management’s performance reporting |
| Truthseeking group | A group committed to accuracy, accountability and openness to diverse views | A quality review or peer challenge built to find what is wrong |
| CUDOS | Merton’s norms: shared data, universalism, disinterestedness, organised scepticism | The working rules of supervision and review |
| Backcasting | Imagining success and working back to its causes | Mapping the conditions a strategy or project depends on |
| Premortem | Imagining failure and working back to its causes | A planning exercise that adds risks the register has missed |
| Ulysses contract | A precommitment binding the future self | Limits, triggers and escalation rules set in advance |
| Tilt | Decision-making driven by in-the-moment emotion | A reason to pause a heated meeting or investigation |
What it means for internal audit: judging decisions after the fact
Internal audit arrives after decisions have been made and, often, after their outcomes are known. That is its value, because it brings an independent view to what happened, and its hazard, because the outcome can colour the auditor’s judgment. The Global Internal Audit Standards require each finding to be evaluated for significance, with root causes identified where possible and the likelihood and impact of the risk considered (Standard 14.3); the book’s contribution is a reminder that the outcome that prompted the work is only one draw from the possible outcomes, and that likelihood must be judged as it stood when the decision was taken. The table maps the book to the main activities; the life of a finding guide and the finding severity guide carry the surrounding procedures.
| Audit activity | What the book implies | Procedure it changes |
|---|---|---|
| Root cause analysis | The outcome is weak evidence of the cause; hindsight makes failures look predictable | Reconstruct what was known, expected and decided at the time before writing the cause; see the root cause guide |
| Post-implementation and project reviews | Reviewing only failures teaches the wrong lessons | Select projects for review by size and uncertainty, not by outcome, and review some successes; the capital projects guide covers the programme |
| Incident and loss reviews | Blame follows outcomes; luck hides near misses | Ask what made the bad outcome possible and how often the same conditions passed without incident; the human error summary develops the method |
| Evaluating management’s risk acceptances | An acceptance is a bet and should be recorded as one | Test whether accepted risks carry a stated probability, a loss range and a review date; Standard 11.5 addresses communicating the acceptance of risks |
| Incentives and performance measures | Paying for outcomes rewards luck and can reward breaches that happened to pay | Check whether bonus measures can reward a policy breach with a favourable result; the conduct risk guide covers the wider risk |
| Audit planning and engagement planning | Premortems and backcasting surface risks that registers miss | Run a premortem for the largest planned engagement and for high-risk projects in the audit universe |
| Supervision and quality review | A truthseeking group needs norms, not goodwill | Give the reviewer the explicit task of finding the weakest point in each conclusion; record challenges and their resolution |
| Reporting conclusions | Beliefs held without a stated confidence invite overconfidence | Where a conclusion rests on judgment, say how confident the function is and what evidence would change its view |
Two cautions keep the map honest. The first is that bad luck must not become a universal excuse. The book’s argument cuts both ways: a bad outcome does not prove a bad decision, but it is still evidence, and over many decisions outcomes do reveal the quality of a process. An auditor who accepts “it was a sound decision, just unlucky” without seeing the record of what was decided and why has replaced one bias with another. The second is that judging decisions requires a standard of what a reasonable decision looked like at the time, and the auditor must take it from the organisation’s own policies, limits and appetite rather than invent it afterwards. The risk appetite guide describes the standard most decisions should be measured against.
The criticisms, and what the book leaves out
The reviews we found were mostly favourable, and the substantive criticism has come from specialists. Eric-Jan Wagenmakers, a professor in the Psychological Methods Group at the University of Amsterdam, gave the book four stars out of five in a two-part review in September 2019 and found it fun and informative, but made four objections. It never uses the word Bayesian and does not acknowledge Bayesian predecessors such as Laplace, De Morgan, Ramsey and Lindley, the last of whom had described life as a gamble. Its account of p-values and confidence intervals as expressions of uncertainty is, in his view, naive. He objected to its praise of Heterodox Academy’s effort to bring conservatives into social science, arguing that learning from opponents helps only when the other side has a minimum level of knowledge and argues in good faith. And he argued that the analogy the book uses near its end to explain hindsight bias, a tree and a chainsaw, is mostly wrong, because hindsight bias is a failure to remember how ignorant we were rather than a matter of futures that could have happened. Cedric Chin, writing in 2020, agreed with much of the book but argued that it follows the mainstream view of heuristics as sources of bias, when for experts, poker players included, heuristics are how expertise works. David Foulke, reviewing the book for Alpha Architect in the week it appeared, called his reservations quibbles but found that the book draws on so many sources that its core message is at times diffused.
Three further limits matter for an auditor, and they are our reading. The first is that the book’s advice is addressed to individuals and small groups, and an auditor has to carry it into the organisational machinery, such as incentive schemes, approval limits and governance, that shapes how decisions are made and judged in a company. The second is that the Super Bowl example shows the difficulty as clearly as the principle: the call can be argued either way, which means that judging decision quality requires a model of the probabilities, and in most business decisions no one has written one down. The third is accountability. A framework that separates decisions from outcomes protects sound decision-makers from unfair blame, but it can also shield poor ones, and the defence against that is a record made at the time of the decision, which many organisations do not keep. For the psychology of judgment behind the book’s biases, the Thinking, Fast and Slow summary is the starting point, and the Noise summary covers the variability that persists even when bias is controlled.
Editions, companions and how to read it
| What | Details | Best for |
|---|---|---|
| Hardback and ebook | Portfolio, 6 February 2018; the library record gives ix, 276 pages with notes and index | The first edition |
| Paperback | Portfolio, 7 May 2019 | The edition most readers will find |
| Audiobook | Read by the author; the publisher lists a running time of 6 hours 51 minutes | Listening |
| Later books by the author | How to Decide (2020), a book of tools and exercises that grew from readers’ requests; Quit (2022), on knowing when to stop | Putting the ideas into practice |
| Reading plan | The introduction and the first chapter for the central idea; the third chapter for learning from outcomes; the fifth for the group norms; the sixth for the techniques | An auditor with one evening |
| Companion reading | The site’s summaries of Thinking, Fast and Slow for the psychology, Superforecasting for calibration and Noise for consistency of judgment | Placing the argument |
Read it with the function’s last three root cause analyses or post-implementation reviews to hand. The question to ask of each is whether it judged the decision or the outcome, and whether it could have done otherwise with the records the organisation kept. Most functions will find that the answer to the second question is the more important finding.
A one-hour discussion guide for an audit team
- Ten minutes: resulting in our own reports. Take the last three findings that followed a bad outcome. For each, ask whether the same decision with a good outcome would have produced a finding.
- Fifteen minutes: the good outcomes. Name one decision in the organisation that turned out well and was never reviewed. List what would have happened if luck had gone the other way, and decide whether it deserves a look.
- Ten minutes: wanna bet? Each person states, as a percentage, how confident they are in the function’s most contested current conclusion. Discuss the spread and what evidence would move it.
- Fifteen minutes: a premortem. Assume the largest engagement in next year’s plan has failed to find a significant issue that later emerged. Each person writes the reasons independently; compare and add the new ones to the planning memo.
- Ten minutes: our norms. Agree one change to supervision that applies Merton’s norms, such as sharing all exceptions with the reviewer or asking the reviewer to argue the opposite conclusion, and name an owner.
Worked example: Brightwater Foods reviews its decisions, not its outcomes
Brightwater Foods is a fictional US food manufacturer with $180 million of revenue, three plants and an internal audit function of one auditor supported by a co-source firm. In 2026 three events prompted the audit committee to ask how the company evaluates its decisions. The treasurer was criticised in the monthly finance pack for a $1.4 million loss on wheat futures after prices fell, although the company’s hedging policy required 60 to 80 percent of the next six months’ wheat and sugar needs to be hedged and the hedge sat at 70 percent. The purchasing director was praised, and received a larger bonus, after leaving two quarters of sugar purchases unhedged in breach of the same policy, because prices fell and the purchase price variance came in $0.9 million favourable. And a product recall for an undeclared allergen, costing $2.3 million, was attributed by the operations team’s review to an operator on a night shift. In the 2027 plan the internal auditor scheduled a review of decision evaluation, with the co-source firm providing analytics and facilitation, and used the book’s distinctions to frame it.
| Test | What the review found | Result |
|---|---|---|
| Are decisions recorded when they are made? | Of 14 decisions above $1 million in 2025 and 2026, three had a written record of the alternatives considered and none recorded the expected outcome, the main uncertainties or what would change the decision | Finding: a one-page decision record for decisions above $1 million, stating options, key assumptions as ranges and the conditions for revisiting |
| Are reviews triggered by outcomes? | Five of the 14 decisions had been reviewed after the event, all of them after bad outcomes; none of the decisions that turned out well had been reviewed | Finding: reviews selected by size and uncertainty, with at least one good-outcome decision reviewed each year |
| Was the hedge loss a bad decision? | The futures loss of $1.4 million was matched by lower costs on the physical wheat purchased, which the finance pack had not shown alongside it; the hedge complied with policy | Observation: hedge results to be reported net of the hedged purchases, so that the committee sees the economic result |
| Was the sugar gain a good decision? | The unhedged position breached policy for two quarters without approval; had prices risen 20 percent, the exposure on the $8 million of sugar purchased would have been about $1.6 million; the bonus plan paid on purchase price variance with no adjustment for policy compliance | Finding: the breach reported to the audit committee; the compensation committee asked to exclude results achieved through policy breaches from bonus measures |
| Did the recall review field the outcome? | The co-source firm sampled 40 allergen changeovers across the three plants and found the cleaning verification step skipped on 11, mostly where a changeover fell at the end of a shift; the night-shift operator had followed the same practice as colleagues | Finding: root cause restated as a changeover schedule that left no time for verification; schedule and sign-off redesigned |
| Would a premortem add anything? | The co-source firm ran a premortem with the project board of the planned $6 million system replacement; of 23 reasons for failure the board generated, four were not in the project risk register, including the migration of allergen data | Observation: the four risks added to the register with owners |
The report carried four findings and two observations, and the audit committee’s discussion turned on the two commodity decisions. The committee chair observed that the committee had criticised the person who followed the policy and applauded the person who broke it, on the strength of a price movement neither of them controlled, and asked for the finance pack to show hedging results net of the hedged purchases from the next quarter. The purchasing director’s view, recorded in the management response, was that he had read the market correctly; the auditor’s report did not dispute that he might have, but noted that the policy existed because no one reads the market correctly every time, and that a breach which pays is still a breach. The decision on the bonus plan was the compensation committee’s, and the report said so.
The auditor applied the same discipline to the function’s own work. Each risk rating in the 2027 plan now carries a stated confidence, and the co-source firm’s engagement partner reviews the two engagements with the least confident ratings with the explicit task of arguing the opposite conclusion before the report is issued. The one-person audit shop manual and the sourcing model guide describe how a function of one can build that kind of challenge into its year, and the treasury audit guide covers the hedging oversight that the review touched.
Frequently asked questions
What is resulting?
The poker players’ term, which Duke adopts, for judging the quality of a decision by the quality of its outcome. It is a mistake because luck and hidden information mean that good decisions can turn out badly and bad ones well; a single outcome says little about the decision behind it.
Is Thinking in Bets a book about poker?
No. The introduction is titled to say so. Poker supplies the model, a game of incomplete information and luck, and some of the examples, but the argument is about decisions in business, sport, politics and ordinary life, and no knowledge of poker is needed.
How does it relate to Thinking, Fast and Slow, Superforecasting and Noise?
It is a practitioner’s account of decision-making under uncertainty that sits alongside them: the Thinking, Fast and Slow summary covers the psychology of bias, the Superforecasting summary covers calibrated prediction, and the Noise summary covers inconsistency in judgment. In our view, Duke’s distinctive contribution is the separation of decision quality from outcome quality.
Does the book make bad luck an excuse?
It can be misread that way. Its argument is that an outcome is weak evidence about a single decision, not that outcomes are irrelevant; over many decisions they do reveal the quality of a process. For an auditor, the defence against the excuse is a record of what was decided and why, made at the time.
What is a truthseeking group, and does an audit function need one?
A small group whose members agree to value accuracy over agreement, hold each other accountable for their reasoning and seek out different views. An audit function already has the structure, in supervision and quality review; the book’s contribution is the norms that make the review challenge conclusions rather than confirm them.
Where does the book sit in an auditor’s reading?
With the books on judgment and decision-making; the internal audit reading list organises the site’s recommendations by theme and career stage.
Related guides
- Thinking, Fast and Slow by Daniel Kahneman: Summary for Auditors — the psychology of judgment and bias
- Superforecasting by Tetlock and Gardner: Book Summary and Review — calibrated belief and measured prediction
- Noise by Kahneman, Sibony and Sunstein: Summary and Lessons for Auditors — consistency of judgment across people
- Root Cause Analysis for Audit Findings: 5 Whys Worked Examples — reaching the cause rather than the outcome
- Risk Acceptance by Management: The Right Process for Letting Issues Go — recording management’s bets
- The Field Guide to Understanding ‘Human Error’ by Sidney Dekker: Summary for Auditors — investigating failure without hindsight
- The Best Books for Internal Auditors — where this book sits among the others
- All book summaries — the full set of summaries for auditors
- How to Measure Anything by Douglas Hubbard: Summary and Lessons for Auditors — calibrated estimates and the value of information.
Leave a Reply