Cited from real sources 6 min read Updated September 2026

A decision framework by Annie Duke

Annie Duke's Thinking in Bets: How to Judge a Decision Without Its Outcome

Thinking in Bets is Annie Duke's reframe. Every decision is a bet placed under uncertainty. Its quality lives in the process, not the result. Duke played poker for a living. She is now a special partner at First Round Capital. She argues that judging a call by how it turned out teaches you little. Luck and hidden facts sit between the choice and the outcome. So write the bet down before you know.

The framework

A good result is not evidence of a good decision

Two things determine how anything turns out: the quality of your decisions and luck. You control one of them. Duke's framework starts by refusing to let the second one grade the first. She calls the error resulting, and it is the habit of reading backward from an outcome to a verdict on the choice that produced it.

It survives because resulting feels like evidence. You shipped. The number moved. So the call was right. But two things sit between the decision and the result that you could not see at the time. One is luck. The other is information that only arrives after you have committed.

Her sharpest illustration is an investor who was early into Uber. Everyone now treats that as proof of judgment. Duke points out that nobody knows which it was. Not even the investor. It could be a read on a real pain point in a developing market. Or it could be a friend who happened to start the company.

we don't actually know what the decision quality was, right, like all we know is that you had a good result

Duke on the investor who was early into Uber Watch at 53:21

This is not academic. If you cannot separate luck from skill, every fortunate win hardens a process that will fail you. Every unlucky loss talks you out of one that works. Betting language fixes that, because a bet forces you to say how sure you are before the cards turn over.

How to apply it

How do you actually think in bets?

Seven moves, drawn from Duke's framework and from how she rebuilt decision records at a venture fund.

1

State the decision as a bet, not a conclusion.

You are wagering money, headcount or a quarter on one future out of several. Name what you are risking and what has to be true for it to pay off.
2

Express confidence as a percentage.

Not "I'm sure" but "I'm 72 percent". New evidence can move a number. You can only defend a certainty. That is why people defend it.
3

Record the forecast before the result exists.

A prediction reconstructed after the fact is worthless, because hindsight rewrites it. Write it where someone else can read it back to you.
4

Shorten the feedback loop on purpose.

Duke's position: the loop is as long as you choose it to be. Ask what correlates with the outcome you care about, then track it instead of waiting years for a verdict.
5

Field the outcome before you change anything.

Place each result somewhere on the luck-to-skill spectrum first. Most sit in the middle. Extracting a lesson from the luck portion is how teams end up with superstitions.
6

Run the "wanna bet" test on anything that feels certain.

If wagering on a belief makes you less sure, you were holding it at a confidence the evidence never earned.
7

Match the effort to the stakes.

Duke's happiness test: will the outcome affect you in a week, a month or a year? If not, decide fast and move on. Reserve the full process for bets that can hurt.

What this looks like at a venture fund

Duke has spent five years inside First Round Capital doing this. When she arrived, the partnership recorded the vote and little else. Who said yes. Who said no. She had them agree on shared definitions of what they were judging. She had them rate components like market and team on a scale of one to seven so the opinions carried spread. And she had them forecast one thing: the probability the company funds at Series A.

Hundreds of companies later, the fund can ask a question it could not before. Is a given partner better than random at that forecast? The answer goes back to the partner, which is the only mechanism by which calibration improves.

let's make it explicit because you're doing it implicitly anyway

Duke on why the forecast already exists Watch at 58:52

Note what changed. Not the judgment. The record of it.

Boundary conditions

When it works, when it fails

Works best when

The decision type repeats often enough that a base rate can form
You can name a leading signal correlated with the outcome you care about
You write the forecast down before the result, where someone can check it
The room rewards being accurate over sounding confident

Fails when

Nobody ever scores the forecasts, so the numbers are theater and calibration never moves
You attach the percentage to something unknowable, which buys false precision
You reconstruct the prediction after the outcome, where hindsight bias rewrites it
Being wrong in the room costs status, so people stop making the call explicit

The last one is the real constraint, and it is psychological rather than technical. A long feedback loop is comfortable. If you were early into a company that worked, people already credit you for judgment. Finding out whether you deserve it can only cost you. Tightening the loop means volunteering to be wrong sooner.

it's very very difficult for human beings to deal with feeling wrong in the moment even if it helps them in the long run

Duke on why nobody volunteers for a tighter loop Watch at 54:04

Where operators disagree

Duke's rule is to put a number on your confidence. Douglas Hubbard pushes it further: anything that matters is observable and therefore measurable, and calibrated probability estimation makes even intangibles quantifiable. Nassim Taleb draws the opposite line. We overestimate our ability to predict in complex domains. A precise-looking model there is more dangerous than admitted ignorance, because it manufactures false confidence.

The reconciliation both camps accept: calibrate where outcomes stay within a boundary and repeat often enough to score them. Where they do not, stop forecasting and build for robustness. Duke's own practice sits in the first case. She picked Series A funding to forecast because hundreds of instances exist to score it against.

Thinking in bets is the judging half of Duke's work. The acting half is her kill criteria. They convert the bet into a pre-committed exit condition, so you are not relying on clear judgment at the moment you have none. The decision-making framework guide puts both moves in sequence alongside operators who attack the other failure modes.

The sources

Where Duke discusses this

47:46

Timestamp

Lenny's Podcast · This will make you a better decision maker

Why poker is a noisier feedback loop than it looks

Duke answers the objection that poker gives instant feedback. Winning a hand tells her almost nothing, because she almost never sees an opponent's cards. She needs many hands to know if she has an edge.

53:21

Timestamp

Lenny's Podcast · This will make you a better decision maker

The investor who was early into Uber

Duke uses the celebrated early investor to show that a good result is not evidence of decision quality. A long feedback loop is often comfortable.

58:52

Timestamp

Lenny's Podcast · This will make you a better decision maker

Making the implicit forecast explicit at First Round

Every seed investor is already forecasting whether a company funds at Series A. Duke's change was to write the forecast down so the fund could measure accuracy and feed it back to the partner.

71:44

Timestamp

Lenny's Podcast · This will make you a better decision maker

Luck and hidden information sit inside every start

Duke names the two forces that make a decision a bet. One is luck, which is not in your control. The other is information you only learn after you have committed.

Useful? Send it to whoever on your team is about to face a grade on an outcome they did not control.

Want the full playbook?

Get 457 strategic thinking frameworks.

89 frameworks 125 rules 235 heuristics & principles 11 operators

From Annie Duke, Donella Meadows, Ray Dalio, and 8 more. Drop one .md into Claude, Cursor, or ChatGPT. Your AI cites practitioners, not guesses.

See the pack

Instant .md download · One-time purchase · No subscription

Related frameworks

Shipping

Kill Criteria

Annie Duke

Name the signals that mean stop before you launch, and attach an action to each one while you still have no stake in the answer.

Decision Making

The Four Flavors of Risk

Shaan Puri

Failure is the cheap risk. The thing that stays just okay for years is the one that takes your time.

Decision Making

The Flywheel Effect

Jim Collins

Breakthrough is cumulative, not decisive. Asking which push caused it is, in Collins's words, a nonsense question.

Decision Making

Believability-Weighted Decisions

Ray Dalio

Not the boss deciding, not one-person-one-vote. Opinions weighted by repeated track record plus explainable cause and effect.

A good result is not evidence of a good decision

How do you actually think in bets?

State the decision as a bet, not a conclusion.

Express confidence as a percentage.

Record the forecast before the result exists.

Shorten the feedback loop on purpose.

Field the outcome before you change anything.

Run the "wanna bet" test on anything that feels certain.

Match the effort to the stakes.

When it works, when it fails

Works best when

Fails when

Where Duke discusses this

Get 457 strategic thinking frameworks.

Know when the next expert lands.