Decision Frameworks
A good decision can still lose money, kill a project, or look stupid for 3 years. Decision frameworks do not remove luck; they separate process from outcome so you can improve the part you control. The useful ones force a person to name uncertainty, size risk, find failure paths, and choose speed on purpose.
The small toolkit
Six frameworks cover most of the ground.
Bayesian updating asks: what did I believe before, what evidence arrived, and how much should that evidence move me? The math is Bayes' theorem, but the practice is humbler: turn "I think this will work" into "I give this a 65% chance," then check the score later. Philip Tetlock's forecasting tournaments found that calibrated forecasters update in small increments, not heroic reversals.
OODA loop came from US Air Force colonel John Boyd: Observe, Orient, Decide, Act. The part people underprice is Orient. Raw data does not decide anything; models do. In a dogfight, a trading desk, or a live incident room, the actor who cycles through the loop faster can make the opponent respond to an older world.
Kelly criterion sizes bets when the odds and payoff are estimable. The simple form is f = (bp - q) / b, where f is bankroll fraction, b is net odds, p is win probability, and q is loss probability. Kelly maximizes long-run geometric growth, which makes it a cousin of concept compounding. The catch is brutal: if your edge estimate is wrong, full Kelly punishes you. That is why many practitioners use half-Kelly.
Pre-mortem comes from Gary Klein. Before committing, ask each person to privately write: "It is one year from now. This failed badly. Why?" The trick is psychological judo. People are often poor at forecasting, but good at explaining failure after the fact. The pre-mortem borrows that retrospective machinery before the damage exists.
Inversion is the old Jacobi move popularized by Charlie Munger: invert, always invert. Instead of asking how to build a great career, ask how to guarantee a bad one: stop learning, work with people you dislike, spend all income, avoid hard feedback, pick status over fit. The negative map is often clearer than the positive one.
Reversible vs irreversible decisions was popularized by Jeff Bezos as one-way and two-way doors. One-way doors deserve slow thought because exit is expensive. Two-way doors deserve speed because delay is often costlier than error. Many organizations treat a copy change and a factory purchase with the same meeting ritual. That is not caution; it is category error.
Where each one earns its keep
| Framework | Best use | Failure mode it attacks |
|---|---|---|
| Bayesian updating | Beliefs that change with evidence | Frozen priors, overreaction |
| OODA loop | Fast adversarial environments | Slow orientation |
| Kelly criterion | Repeated bets with measurable edge | Ruin from oversizing |
| Pre-mortem | Project and strategy risk | Social silence before failure |
| Inversion | Fuzzy life or product choices | Chasing vague success |
| One-way / two-way doors | Time allocation for decisions | Treating all choices as permanent |
The table matters because these tools are not interchangeable. Kelly is dangerous when the probability is fantasy. OODA is weak when feedback takes 5 years. A pre-mortem will not tell you the correct price for a bond. Inversion can prevent obvious self-sabotage, but it will not discover a new market.
The sharp line: use frameworks to reduce avoidable stupidity, not to manufacture certainty.
What's contested
The contested part is not whether structure helps. The contested part is how much structure survives contact with real judgment. Kahneman and Tversky showed systematic biases in human reasoning, but Gerd Gigerenzer argued that simple heuristics can outperform formal models in noisy environments with limited data.
Another fight sits inside probability itself. Bayesian decision-making wants explicit priors; many practitioners hide priors because naming them feels fake. The honest answer may be worse: unnamed priors still exist, only now they cannot be audited.
There is also a moral hazard in frameworks. A team can run a pre-mortem, assign probabilities, write a decision memo, and still use the ritual to bless a decision already made. Process theater is harder to spot than no process.
Why this has to do with other realms
Decision frameworks are philosophy with a stopwatch. They ask old questions from concept pragmatism in operational clothing: what belief should cash out in action, and what action should change the belief? That is why the same page belongs beside concept bayesian reasoning and concept artificial intelligence.
AI systems make the bridge sharper. A model can rank options, simulate scenarios, and update on new data, but it still needs an objective function and a loss function. Kelly without human risk preference is just arithmetic. OODA without values is speed in no agreed direction. The machine can help with the loop; it does not tell you which game deserves the loop.
An open question
If the best decision-makers are not the ones with the best framework, but the ones who know when to abandon one, what would a page on concept judgement have to measure?
Key Sources
- Thinking, Fast and Slow by Daniel Kahneman (2011) — the modern public entry point for biases and judgment under uncertainty.
- Superforecasting by Philip E. Tetlock and Dan Gardner (2015) — evidence from forecasting tournaments on calibration and updating.
- John Boyd, "The Essence of Winning and Losing" (1995 briefing) — compact source for OODA as Boyd framed it.
- John L. Kelly Jr., "A New Interpretation of Information Rate" (1956) — original Kelly criterion paper in Bell System Technical Journal.
- Gary Klein, "Performing a Project Premortem" (Harvard Business Review, 2007) — short canonical write-up of the pre-mortem exercise.
- Jeff Bezos, Amazon shareholder letter (2016) — one-way and two-way door decision framing.
Further Reading
- Sources of Power by Gary Klein — naturalistic decision-making from firefighters, soldiers, and operators under time pressure.
- The Psychology of Intelligence Analysis by Richards J. Heuer Jr. — why analysts get trapped by mental models.
- Poor Charlie's Almanack by Charlie Munger — inversion, incentives, and error avoidance in plain language.
- concept compounding — Kelly makes the most sense when growth is geometric, not linear.
- concept first principles — useful when the inherited decision menu is itself the problem.
Abhishek's take
I use this on a buy when a fabric call has to move before the final read arrives. The useful split is not “data versus instinct”; it is whether the decision is a QR tag I can change next week or a 100-day lead-time commitment that locks the range.
See Also
- concept bayesian reasoning
- concept compounding
- concept first principles
- concept power laws
- concept artificial intelligence
- concept pragmatism