Sign inGet started
Skills / Decision-making

Thinking in Bets

What makes a decision great is not that it has a great outcome. A great decision is the result of a good process, and that process must include an attempt to accurately represent our own state of knowledge.

Thinking in Bets is Annie Duke's decision-making framework drawn from professional poker and cognitive psychology. It treats every decision as a bet on an uncertain future and trains you to separate decision quality from outcome quality — so you stop getting fooled by luck, stop repeating bad-process decisions that happened to work out, and stop punishing yourself for sound decisions that suffered bad variance. Built on two published books, including a structured workbook with named exercises and templates, it produces concrete written artifacts at every stage.

By Annie Duke · Free
Specimen 01 · Live diagnosisThinking in Bets
Input

“I turned down a job offer at a startup six months ago and they just got acquired for $200M. I feel like I made a terrible mistake.”

Diagnosis
Before we dig in, one question: are you working through a decision you're about to make, evaluating a choice you've already made, or trying to understand why an outcome went the way it did?
Full transcript ↓
Calibrated referenceagent-skills.ai
The gap

Stop judging decisions by their outcomes — judge the process that made them

The Thinking in Bets framework treats every decision as a probabilistic wager on an uncertain future and demands that decision quality be evaluated by process rigor — not outcomes. The central concept, 'resulting,' names the universal cognitive error of reverse-engineering the quality of a choice from how it turned out. The framework operationalizes six core tools: (1) the Resulting Audit, which strips outcome information to evaluate the decision as it existed at the time it was made; (2) the Wanna Bet? calibration test, which assigns explicit percentage probabilities to key assumptions and exposes the gap between stated and actual confidence; (3) the Decision Matrix, a 2×2 grid mapping process quality against outcome range into four quadrants (Deserved Success, Dumb Luck, Bad Luck, Preventable Failure); (4) Premortem analysis, which imagines catastrophic failure from a future vantage point to surface hidden risks that forward-looking analysis misses; (5) Backcasting, which imagines success and maps the path backward to identify necessary conditions; and (6) Outside View / Reference Class Forecasting, which counters personal narratives with base rates from comparable decisions. Every step is designed to produce written artifacts — probability logs, premortem documents, backcasting plans — making this decision engineering, not reflection.

The problem

Most people judge the quality of their decisions by outcomes — if it worked out, it was a good call; if it didn't, it was a mistake. This is 'resulting,' and it's a systematic cognitive trap that corrupts learning and ruins future decision-making. Because outcomes contain so much noise, a terrible decision can produce a great outcome through luck, and a rigorous decision can produce a bad outcome through variance. Without a framework to separate process from result, you can't learn from either wins or losses — you can only react to them.

The solution

Stop letting outcomes rewrite your memory of why you made a decision. With Thinking in Bets, you evaluate choices by the rigor of the process that produced them — not the variance that followed — so both wins and losses teach you something real.

You bring
  • A current decision, recent outcome, or belief you want to pressure-test — any domain works
  • Your honest account of what you knew at the time the decision was made (not what you know now)
  • Willingness to assign explicit probability percentages rather than vague labels like 'probably' or 'likely'
  • Context on stakes, timeline, and what a good versus bad outcome actually looks like for this situation
You get
  • A Resulting Audit separating what you knew at decision time from what you learned after — eliminating hindsight contamination
  • A Probability Log recording key assumptions with explicit confidence levels, flagged for future calibration review
  • A Premortem Failure Document mapping the most likely failure modes before they occur, with early warning signals
  • A Backcasting Success Plan identifying the necessary conditions and critical milestones between now and a successful outcome
  • A Decision Matrix classifying your decision as Deserved Success, Bad Luck, Dumb Luck, or Preventable Failure — with calibrated learning extracted from each
In action

Watch the methodology work.

Three specimens from a single real session: the same situation, unaided and calibrated, the full transcript, and the skill answering live in the channel where the work happens.

Fig.01The same situation, two ways
Specimens A · B
Specimen AVague
Unaided

“You received a job rejection and spent the evening convinced you had bombed the interview, replaying every answer and concluding you're not cut out for that level of role — even though you'd prepared thoroughly, your interviewers gave positive signals mid-interview, and the company had quietly announced a hiring freeze two weeks later. You're now about to turn down a similar opportunity, telling yourself you've 'learned your lesson,' without examining whether you've actually identified a real pattern or are just reacting to the sting of the outcome.”

×Vague — nothing concrete to act on.
Specimen BCalibrated
With Thinking in Bets

“You run a Resulting Audit and separate what you knew at interview time from what you learned after. You run a Luck vs. Skill Decomposition and find the rejection was 70% attributable to the hiring freeze (variance) and 30% to one genuinely underprepared answer on cross-functional leadership (skill gap). You update that specific skill and move forward — without falsely concluding you're unqualified for senior roles. For the new opportunity, you run a Premortem and Backcasting before deciding, and log your key assumptions with probability estimates so you can review them against reality in six months.”

Specific, grounded, and ends with one clear next step.
Fig.02A real session, unedited
In-app · assistant
Thinking in BetsClaude
Thinking in Bets
Before we dig in, one question: are you working through a decision you're about to make, evaluating a choice you've already made, or trying to understand why an outcome went the way it did? The answer changes our approach — because the biggest trap in decision review is letting the outcome rewrite your memory of the process.
I turned down a job offer at a startup six months ago and they just got acquired for $200M. I feel like I made a terrible mistake.
That stings — and it's also a textbook Resulting trap. You're judging the quality of a six-month-old decision by a recent outcome, but those are two entirely separate questions. Let's run a Resulting Audit first. Set aside the acquisition entirely. At the moment you made that decision, what did you actually know about that startup? And what were the realistic base rates — the odds of a $200M exit versus the odds of the company failing entirely, which is the most common outcome at Series A?
It was a Series A company. Their product was good but unproven. The team was solid but I didn't know them well. I'd say maybe a 15-20% chance of meaningful success at the time. But I still feel like I blew it.
Good — that 15-20% estimate is doing important work here. You've just applied the Outside View: the base rate for Series A exits of that magnitude is indeed in that range. If you made this same decision 100 times with identical information, you'd turn it down 80% of the time and be right to do so. The one time in five where it paid off massively doesn't retroactively make the other decisions wrong. Now let's look at the other side: why did you actually decline? What was your process — not your outcome — and was it sound given what you knew at the time?
I declined because the salary cut was significant, I had a new mortgage, and the equity terms seemed unfavorable compared to the market. I talked to two advisors who both flagged the valuation as aggressive. At the time it felt like a disciplined choice.
That's a good process. You had a real financial constraint with concrete consequences, you sought outside views from people with relevant expertise, and your advisors identified a specific valuation concern — which is exactly the Outside View step. Here's the Resulting Audit verdict: you made a reasonable decision under uncertainty using solid process, and variance broke against you. The acquisition tells you nothing about whether your reasoning was flawed. What would it actually take to conclude your decision-making has a genuine gap? Not 'the outcome was bad' — but what specific process failure would need to be true?
I guess... if I consistently undervalue upside risk? Or if I'm letting short-term financial anxiety override longer-term thinking?
Now we're doing real decision analysis. Those are specific, testable hypotheses — not just 'I made a mistake.' Let's run a Luck vs. Skill Decomposition on this outcome and find out which it is. And if you want, we can run a 10-10-10 check on whether the mortgage constraint was a genuine binding constraint or a present-bias trigger. The goal isn't to feel better about the outcome — it's to extract only the learning that's actually there, not the learning the bad feeling is inventing.
Reply to Thinking in Bets
What the skill is doing
01Resulting Audit
02Wanna Bet? Calibration
03Decision Matrix Mapping
Fig.03The same skill, in the channel you already use
WhatsApp · live
9:41
TIThinking in Betstyping…
Messages are end-to-end encrypted. No one outside of this chat can read them.
Today
I turned down a job offer at a startup six months ago and they just got acquired for $200M. I feel like I made a terrible mistake.9:14✓✓
The answer changes our approach — because the biggest trap in decision review is letting the outcome rewrite your memory of the pr…9:14
Message

The same skill, where the work happens.

No new app to learn. The methodology runs over the WhatsApp Business API, so the answer lands as a reply in the thread you’re already in — same rigour, zero context-switch.

Reads the situation, names the pattern, returns one concrete next move.
Delivered in seconds, inside a conversation that already exists.
Specimen · WhatsApp Business API · live
Capabilities

What it does, specifically.

Each capability is a distinct move drawn straight from the source methodology — not a generic assistant guessing.

CapabilityC-01

Resulting Audit

Before evaluating any decision, this capability strips outcome information and reconstructs the decision as it existed at the moment it was made. This prevents hindsight bias from distorting the assessment and forces evaluation of what you actually knew and did — not what you now know happened. The output is a clean, outcome-independent judgment of process quality.

Based on Duke's core insight that 'resulting' — the reflex to judge decision quality by outcome quality — is the foundational error in decision learning, described in Chapter 1-2 of 'Thinking in Bets' (2018) as the primary target of the entire methodology.
CapabilityC-02

Wanna Bet? Calibration

This capability challenges every stated belief or prediction with a structured betting question: 'Would you put real money on that at those odds?' It forces conversion of vague confidence ('I think this will work') into explicit probability estimates ('I'm 65% confident'), exposing the gap between what you claim to believe and what you actually believe. Probability assignments are logged for future calibration review.

Directly implements Duke's 'Wanna Bet?' technique, developed from her professional poker career and formalized in 'Thinking in Bets' as a tool for calibrating actual vs. stated probability — the most direct way to surface overconfidence.
CapabilityC-03

Decision Matrix Mapping

The skill produces a 2×2 Decision Matrix mapping your process quality against the probable outcome range. This classifies the decision into one of four named quadrants — Deserved Success, Dumb Luck, Bad Luck, or Preventable Failure — and extracts only the learning that's actually warranted by each quadrant. Not all bad outcomes justify self-criticism; not all good outcomes justify repeating the process.

Based on the Decision Matrix framework from Duke's 'How to Decide' (2020), a structured workbook that prescribes this 2×2 grid as the core tool for post-decision evaluation and learning extraction.
CapabilityC-04

Premortem Facilitation

Before committing to a decision, this capability guides a structured failure scenario: assume it is 18 months from now and the decision failed spectacularly. Working backward from that assumed failure, you surface risks, blind spots, and single points of failure that forward-looking optimism systematically hides. The output is a Premortem Failure Document with a ranked risk inventory and specific warning signals to monitor.

Implements the Premortem technique, which Duke draws from Gary Klein's prospective hindsight research and applies as a core tool in both 'Thinking in Bets' and the 'How to Decide' workbook — always paired with Backcasting to bracket the full outcome space.
CapabilityC-05

Backcasting & Success Path Mapping

As the counterpart to Premortem, this capability imagines a successful outcome and maps the path that led there. This identifies the necessary conditions, critical actions, and decision points that must go right — converting an abstract goal into a concrete, testable plan with checkpoints. Prevents the trap of optimizing toward a goal without specifying the route.

Implements Duke's Backcasting framework from 'How to Decide,' which complements Premortem as a paired prospective analysis: Premortem surfaces failure risks, Backcasting maps success requirements — both run before committing.
CapabilityC-06

Outside View Base Rate Analysis

This capability prompts you to step outside your specific situation and ask: what typically happens to people who make this type of decision? By surfacing base rates and reference class data, it counteracts the inside-view bias that leads most people to believe their situation is uniquely likely to succeed — the same bias that drives the planning fallacy and overconfident forecasting.

Applies Kahneman and Tversky's reference class forecasting, which Duke explicitly incorporates in 'Thinking in Bets' Chapter 5 as one of the highest-leverage calibration moves available — especially for decisions where you lack personal experience in the relevant domain.
CapabilityC-07

10-10-10 Temporal Perspective

For decisions colored by strong emotion or immediate urgency, this capability walks through how you'll feel about the choice in 10 minutes, 10 months, and 10 years. This temporal stretching counteracts present-bias — the tendency to over-weight immediate discomfort or excitement — and reveals which decisions actually matter at longer time scales, so deliberation effort is proportionate to genuine stakes.

Based on the 10-10-10 technique Duke incorporates from Suzy Welch's framework within the Thinking in Bets methodology as a specific tool for emotionally charged decisions where present-moment feelings distort the evaluation of long-term consequences.
CapabilityC-08

Luck vs. Skill Decomposition

After an outcome has occurred, this capability facilitates a structured retrospective that explicitly separates what portion of the result was attributable to your decisions and skill versus external variance and luck. This prevents over-learning from lucky successes and over-punishing yourself for unlucky failures — the two most common forms of distorted learning from outcomes.

Implements Duke's Luck/Skill Decomposition reflection prompt, a core post-outcome tool from 'How to Decide' designed to produce accurate — rather than emotionally distorted — learning by forcing explicit attribution of outcomes to controllable vs. uncontrollable factors.
Tested

Graded before it shipped.

Every skill is scored against independent scenarios for methodology fidelity before it goes live — not vibes, a rubric.

What it produces
OutputD-01

Resulting Audit Report

A structured document that reconstructs a decision as it existed at the moment it was made — before outcome information contaminated the evaluation. Documents what you knew, what you assumed, the probability you assigned, and what process you followed, enabling clean learning regardless of what happened next.

OutputD-02

Probability Assumption Log

A running log of explicit probability assignments to the key assumptions underlying a decision. Each assumption is stated, given a numerical confidence level via the Wanna Bet? test, and flagged for calibration review. Over time this log reveals systematic overconfidence patterns across domains.

OutputD-03

Decision Quality Matrix

A 2×2 grid mapping process quality against outcome range for a specific decision, classifying it into one of four named quadrants: Deserved Success, Dumb Luck, Bad Luck, or Preventable Failure. Each quadrant prescribes what learning, if any, should be extracted — not all bad outcomes warrant self-criticism.

OutputD-04

Premortem Failure Document

A prospective failure analysis written from an imagined future vantage point where the decision has gone badly. Names the top failure modes, the assumptions that would need to be wrong for each, and the early warning signals to monitor — all produced before the decision is made.

OutputD-05

Backcasting Success Plan

A backward-mapped success narrative identifying the necessary conditions, critical milestones, and key decision points between now and a successful outcome. Converts a goal into a testable, condition-specific roadmap with explicit checkpoints and dependencies.

The source

Grounded in the original work.

Every answer traces back to a real source and the practitioner who wrote it — not a secondhand summary. Here is the source of record.

Source authorA-01

Annie Duke

Annie Duke is a former World Series of Poker champion with over $4 million in tournament earnings and a World Series bracelet. She holds a cognitive psychology graduate degree from the University of Pennsylvania and left professional poker to become a corporate decision-making consultant and author. Her book 'Thinking in Bets' (2018, Portfolio/Penguin) became a Wall Street Journal bestseller and is widely used in business schools and investment firms. Her follow-up 'How to Decide' (2020) operationalizes the framework into structured workbook exercises. She is co-founder of the Alliance for Decision Education, a nonprofit working to bring decision science into K-12 schools.

Status · Inspired by Annie Duke’s work — not yet claimed. Are you Annie Duke?
Primary sourceS-01

Thinking in Bets: Making Smarter Decisions When You Don't Have All the Facts (2018); How to Decide (2020)

by Annie Duke

World Series of Poker champion; $4M+ tournament earnings; cognitive psychology graduate, UPenn; WSJ bestselling author; corporate decision consultant; co-founder, Alliance for Decision Education.

Read the original ↗
Citationannieduke.com
In the build queue

Be first to run it.

Thinking in Bets is being built right now. Leave your email and we’ll tell you the moment it goes live.

Notify meEmail
At launchI need to run a Resulting Audit on a decision I made that had a bad outcome — I keep second-guessing myself but I'm not sure if I actually made a mistake or just got unlucky. Can you help me separate those?