Venture Step E015: Thinking in Bets and Decisions

What Thinking in Bets Changed About How I Review Decisions

When I recorded Venture Step E015 in May 2024, I had just finished Annie Duke's Thinking in Bets. I liked the stories, but the idea that stayed with me was simple: the result of a decision and the quality of the decision are not the same thing.

That sounds obvious until an outcome arrives. Once something works, we build a story about why it was smart. Once it fails, we see every warning as if it had been impossible to miss.

flowchart LR
    A["Decision-time evidence"] --> B["Choice under uncertainty"]
    B --> C["Outcome plus luck"]
    C --> D["Process review"]
    C --> E["Outcome review"]
    D --> F["Updated decision rule"]
    E --> F

The idea I took from the book

The publisher describes Thinking in Bets as a guide to making choices when facts are incomplete and luck affects results. Poker gives Duke a useful setting because a player must act before all hidden information is known.

That is also the setting for product work, career moves, hiring, investing, and many operating decisions. We make a choice with partial evidence, then the world supplies a result that contains both signal and chance.

The transcript shows me trying to use that frame on my own life. I was asking whether I could separate a choice I made from the discomfort or satisfaction that followed it.

The football example needed correction

The book opens with Seattle's late pass in Super Bowl XLIX. My transcript described the game state incorrectly. Seattle was not ahead with several minutes remaining.

The NFL's official account records Seattle trailing New England 28 to 24, facing second-and-goal at the one-yard line with 26 seconds left. Malcolm Butler intercepted the pass.

The corrected state does not settle whether the call was optimal. It shows why the facts must be preserved before the argument begins. A lesson about decision quality cannot rest on a rewritten decision context.

Outcome bias is more than a slogan

In their original outcome-bias studies, Jonathan Baron and John Hershey gave people decisions made under uncertainty and varied the outcomes. Participants judged the decision process and decision maker more favorably when the outcome was favorable, even when the information available at decision time was held constant.

That finding does not tell us to ignore results. It tells us to reconstruct the decision before the result changes how the evidence looks.

My current decision-review sequence

I now want the record made before commitment. What was the decision? Which alternatives were real? What evidence did I have? What did I not know? What probabilities did I assign? Who was affected? What authority and constraints applied? Which harms were foreseeable? What would make me stop?

After the outcome, I review the process first. Did I define the question, gather relevant evidence, consider alternatives, state uncertainty, seek dissent, and plan recovery?

Then I open the result. Was it inside the predicted range? Did luck dominate? Did the outcome reveal an assumption I should have tested? Is one result noise, or is a repeated pattern exposing a weak rule?

This order prevents two opposite mistakes. I do not want to condemn a sound process because one uncertain result was bad. I also do not want to protect a careless process by calling every failure bad luck.

What the transcript got right

The episode emphasized speaking in probabilities, finding people willing to challenge a story, and creating a contract for honest disagreement. Those ideas still belong together.

A number forces more precision than saying probably. A dissenting reviewer can expose evidence I ignored. A written record can show whether I changed my explanation after learning the outcome.

The transcript also used future perspective. Imagine how a decision looks later, or imagine it failed and explain why. Those prompts can surface effects that short-term excitement hides.

What I would add now

A sound process is not a defense against negligence. It does not excuse missing authority, conflicts, foreseeable harm, required controls, or evidence that was available but ignored.

The standard should rise with consequence and irreversibility. A reversible content experiment needs less documentation than a decision involving safety, employment, customer data, regulated activity, or large capital.

Probability is not a costume for confidence. If I write 70 percent, I should define the event, deadline, evidence, and resolution. Across many comparable forecasts, I should be willing to see whether my 70 percent events happen at anything close to that rate.

The practical result

E015 began as a book response. The useful output is now a sequence: journal the decision, calibrate uncertainty, run a premortem, make the choice, preserve the record, review the process without the result, then use the result as new evidence.

The book profile explains the source. [[Decision Quality Is Not Outcome Quality]] develops the principle. The journal, premortem, calibration, and review guides turn it into practice.

The goal is not to feel right after every result. It is to leave enough evidence that the next decision can be better.

AI assisted with research organization, structure, drafting, and validation. Dalton Anderson remains the attributed author and final editorial authority. The transcript and linked public sources control factual claims. Publication remains unauthorized.