How to Make Good Decisions Under Uncertainty

By Pritesh Yadav June 22, 2026 24 min read —

You make thousands of decisions a day. Most are tiny and automatic: which shoe to put on first, when to cross the street. A few are huge and slow: which job to take, whether to move cities, how to invest your savings, when to quit something that isn't working. This chapter is about those bigger decisions, and the surprising fact that there is a science to making them well.

The hard part of almost every important decision is that you don't know how it will turn out. You can't see the future. You're choosing in the fog. So the question this whole chapter answers is: how do you make a good decision when you can't know the outcome in advance?

We'll build the answer from the ground up. No formulas you can't follow. No jargon left undefined. By the end you'll have a clear mental toolkit and a few habits that, used consistently, will measurably improve your choices for the rest of your life.

28.1 What "Decision Science" actually is

Decision science is the study of how people decide, and how they should decide, when they don't have all the facts. It borrows from probability (the math of chance), economics (how people value things), psychology (how the brain really works), and operations research (planning under constraints).

There are three different questions hiding inside "how should I decide," and keeping them separate is the single most useful habit in this field. We'll use these three words throughout the chapter:

Normative: What a perfectly rational decision-maker should do. The ideal. The gold standard you aim at.
Descriptive: What real human beings actually do — including all our predictable mistakes.
Prescriptive: The practical tricks that close the gap between the two. The "what to do on Monday morning" tools.

Analogy: Think of running. Normative is perfect textbook form. Descriptive is how you actually run — slouching, breathing wrong. Prescriptive is the drills a coach gives you to slowly fix the gap. Decision science gives you all three: the ideal, an honest picture of your flaws, and the drills.

Key takeaway: A good decision is not the same thing as a good outcome. You decide before you know how things turn out, using whatever foresight you have. You judge outcomes after, with hindsight. Judging a decision purely by its result is a trap — and avoiding that trap is the foundation of everything that follows.

28.2 The most important idea: decision ≠ outcome

Imagine you're driving home from a party. You've had a couple of drinks. You decide to drive anyway, and you arrive home safely. Was that a good decision?

Of course not. The outcome was fine, but the decision was terrible — you took a huge, foolish risk and got lucky. Now flip it: a sober, careful driver gets hit by a truck running a red light. Bad outcome, but the decision to drive carefully was perfectly sound.

The poker champion and author Annie Duke calls the mistake of judging a decision by its result "resulting." It's everywhere. A company makes a reckless bet that happens to pay off, and everyone calls the CEO a genius. A careful plan fails because of bad luck, and everyone blames the planner.

Common mistake — Resulting: "We won, so it must have been a great call." "It failed, so the decision was stupid." Both confuse luck with skill. You can only control the quality of your decision. The outcome is decision quality plus luck. Separate them, always.

Why does this matter so much? Because if you judge decisions by outcomes, you learn the wrong lessons. You'll copy lucky-but-reckless choices and abandon sound-but-unlucky ones. The whole point of decision science is to get better at the part you control: the process. The process is the product.

28.3 Three flavors of not-knowing: certainty, risk, and uncertainty

Not all "I don't know" is the same. Pinning down which kind you're facing changes how you should think.

Certainty: You know exactly what will happen.
Risk: You don't know the outcome, but you do know the odds.
Uncertainty (also called Knightian uncertainty, after economist Frank Knight): You don't know the outcome and you don't even know the odds.

Analogy: Certainty is a vending machine — press B4, get the chips, every time. Risk is roulette — you can't predict the spin, but you know the exact odds of every number. Uncertainty is launching a brand-new product in a market nobody has ever measured — you're guessing the odds themselves.

Here's the catch most people miss: most of real life is uncertainty dressed up as risk. We invent confident-sounding numbers ("there's a 70% chance this product will succeed") when we have no real basis for them. That's fine as a rough guess — but pretending a guessed number is a measured one is dangerous false precision.

Common mistake — confusing risk with uncertainty: Treating a number you made up as if it were a number you measured. A made-up "73.4% likely" feels scientific but isn't. Be honest about which kind of not-knowing you're in.

28.4 Probability and base rates: starting from the right place

Probability is just how likely something is, on a scale from 0 (impossible) to 1 (certain), or 0% to 100%. When a forecaster says "70% chance of rain," it means: on 100 days that look like today, it rained on about 70 of them.

The most underused idea in all of decision-making is the base rate: the background frequency of something in general, before you look at your specific case. It's also called taking the "outside view."

Example: You're about to start a company and you feel certain it will succeed. Helpful fact: roughly 9 out of 10 startups fail. That 90% failure figure is the base rate — the outside view. Your gut optimism is the inside view ("but mine is different!"). The professionals who forecast best — the "superforecasters" studied by researcher Philip Tetlock — almost always start from the base rate first, then adjust.

Common mistake — base-rate neglect: Getting swept up in the vivid details of your specific case and ignoring the boring background numbers. The details feel more real, so they win your attention — but the base rate is usually the stronger signal.

Best practice: For any prediction, ask first: "How often does this kind of thing succeed in general?" Start there. Then nudge up or down for what's special about your case. Base rate first, then adjust — in that order.

28.5 Expected Value: how to weigh a gamble

Now we get a tool for putting numbers on choices. Expected Value (EV) is the average payoff you'd get if you could repeat a choice many, many times. You calculate it by multiplying each possible outcome by its probability, then adding them all up.

EV  =  (outcome 1 × its chance)
     + (outcome 2 × its chance)
     + ...

Example — the lottery ticket: A ticket costs €2. It gives a 1-in-10-million chance of winning €5,000,000.
EV of the prize = €5,000,000 × (1 / 10,000,000) = €0.50.
You are paying €2 for something worth, on average, fifty cents. That's a losing bet by €1.50 every time. This is exactly how lotteries, casinos, and insurance companies make money: they always sit on the positive-EV side of the table, and let the crowd take the negative-EV side.

A bet whose EV is positive — worth taking on average — is called +EV ("positive expected value"). A bet with negative EV is one you should usually skip. But notice the words "on average": a +EV bet can still lose this particular time, and a -EV bet (like the lottery) can occasionally win. EV tells you the smart long-run play, not what happens on any single try.

Key takeaway: Think in terms of "what happens if I made this exact choice 1,000 times?" That mental move — from a single outcome to the average over many — is the heart of thinking clearly about chance.

28.6 Decision trees: mapping out the choices

When a decision has several steps and several uncertain events, it helps to draw it. A decision tree is a branching map: squares are choices you control, circles are chance events you don't, and the numbers at the end are payoffs. You then "fold back" from the tips — calculating the EV of each branch — to see which opening move is best.

Analogy: It's a choose-your-own-adventure book where, at every fork, you've written the odds and the prize. Once it's all on paper, you trace backward to find the best first page to turn to.

                  /-- success (60%) --> +€50,000
   [Take new   ]-O
   [ job offer ]  \-- fail   (40%) --> -€10,000
        |
        |          EV = .60×50,000 + .40×(-10,000)
        |             = 30,000 - 4,000 = +€26,000
        |
   [Stay in    ]----> sure +€5,000
   [ current   ]
   [ job       ]

Laid out like this, the new job's average payoff (+€26,000) clearly beats staying (+€5,000) — though, as the next section shows, raw money isn't the whole story.

28.7 Utility: why money isn't the whole story

Pure EV has a blind spot. Consider two options:

A: a guaranteed €1,000,000.
B: a 50/50 coin flip for €2,000,000 or nothing.

The EV is identical: option B is 0.5 × €2M = €1M, the same as the sure million. So a pure EV calculator says "either is fine." But almost everyone sane takes the sure million. Why? Because the first million changes your life completely, while the second million adds far less. The jump from broke to a millionaire is enormous; the jump from one to two million is nice but smaller.

This is where utility comes in:

Utility: The personal value or satisfaction an outcome gives you — not its dollar amount.
Expected utility: The same averaging as EV, but using utilities instead of raw dollars.
Diminishing marginal utility: Each extra dollar matters a little less than the one before. The hundredth slice of pizza means less than the first.
Risk aversion: Preferring a sure thing over a gamble with the same EV — a direct result of diminishing marginal utility.

Analogy: €1,000 means everything to someone who's broke and almost nothing to a billionaire. Same dollars, wildly different utility. Money is the measuring stick; utility is what the money actually buys for this particular person.

This idea is old. Back in 1738, the mathematician Daniel Bernoulli introduced utility to solve a famous riddle called the St. Petersburg paradox — a coin-flip game whose EV in dollars is technically infinite, yet no sane person would pay more than a few coins to play it. The resolution: people care about utility, which grows slowly, not raw dollars, which the game grew explosively.

28.8 The rational ideal: Expected Utility Theory

In 1944, mathematicians John von Neumann and Oskar Morgenstern proved something elegant. If your preferences obey four simple consistency rules — they're complete (you can compare any two options), transitive (if you prefer A to B and B to C, you prefer A to C), and a couple of technical others — then you behave as if you were maximizing expected utility. This is Expected Utility Theory (EUT), the normative gold standard: the benchmark for what a perfectly rational chooser does.

Analogy: These rules are like the rules of arithmetic for choices. Break them, and you can be "money-pumped" — tricked into a loop of trades that lands you back where you started, only poorer. (If you'll trade your apple for my orange, then my orange plus €1 for your banana, then your banana for my apple... I can spin you in circles taking a euro each loop.)

Common mistake — treating the ideal as reality: EUT describes how a perfectly rational agent should behave (normative). It does not describe how real people actually behave (descriptive). Experiments — the famous Allais and Ellsberg paradoxes — show humans reliably break these axioms. Don't confuse the ideal with the reality. We spend the next sections on the gap.

28.9 Updating beliefs: Bayes' rule in plain words

New information arrives. How much should you change your mind? Too little (stubbornness) is bad; too much (overreaction) is also bad. Bayes' rule, named after Thomas Bayes, is the recipe for getting it just right. In plain words:

Start from the base rate (your prior belief), then move toward the new evidence — but don't forget where you started.

Prior: Your belief before seeing the new evidence.
Posterior: Your updated belief after seeing it.
Bayesian updating: Revising your belief in proportion to how well the evidence fits.

The classic worked example does more teaching than any formula, so let's walk through it slowly.

Example — the medical test that fools everyone:

A disease affects 1% of people (the base rate / prior).
The test catches 90% of people who truly have it.
But it also gives a false alarm 9% of the time for healthy people.

You test positive. How worried should you be? Most people, including many doctors, blurt out "90%." The real answer is about 9%.

Here's why, using 1,000 people:

10 people actually have the disease (1%). The test flags about 9 of them.
990 people are healthy. The test falsely flags 9% of them ≈ 89 people.
So about 9 + 89 = 98 people test positive, but only 9 are truly sick.
Your chance of being sick = 9 / 98 ≈ 9%.

The rare base rate dominates. A positive result on a rare disease usually still means you're fine — because there are so many more healthy people that even a small false-alarm rate produces a flood of false positives.

Common mistake — the prosecutor's fallacy: Confusing "the chance of a positive test if you're sick" (90%) with "the chance you're sick given a positive test" (9%). These are completely different numbers. Flipping them is one of the most expensive errors in medicine, law, and everyday reasoning.

Best practice: When numbers confuse you, switch from percentages to "out of how many real people?" Counting actual heads (1,000 people, 98 positives, 9 truly sick) makes Bayesian reasoning click instantly. Researchers call this using "natural frequencies."

28.10 Why smart people decide badly: heuristics and biases

So far we've covered how a rational agent should decide. Now the honest part: how real brains actually work. The psychologists Daniel Kahneman and Amos Tversky spent decades showing that human judgment runs on mental shortcuts that are usually helpful but sometimes systematically wrong.

Heuristic: A fast rule-of-thumb the brain uses to decide quickly without doing full calculations.
Cognitive bias: A predictable, systematic error that results from those shortcuts.

Analogy: Heuristics are like your phone's autocomplete — fast, right most of the time, occasionally hilariously wrong. Biases are like optical illusions for the mind: even when you know two lines are the same length, they still look different. Knowing about a bias does not switch it off.

Kahneman's bestselling book Thinking, Fast and Slow frames this as two mental systems:

System 1: Fast, automatic, intuitive, emotional. Always running.
System 2: Slow, effortful, logical. Lazy — it only grabs the controls when something feels off.

Analogy: System 1 is the autopilot doing the routine flying. System 2 is the pilot who only takes the controls when an alarm goes off — and who'd rather stay reading the newspaper.

Here are the biases worth knowing by name. Notice how each is a shortcut gone slightly wrong:

Bias / heuristic	Plain meaning	Everyday example
Availability	Judging how likely something is by how easily examples spring to mind.	Fearing plane crashes (vivid, on the news) more than car crashes (far deadlier, but routine).
Representativeness	Judging by how well something matches a stereotype, ignoring base rates.	Assuming the quiet bookish person is a librarian, though there are far more salespeople than librarians.
Anchoring	Over-relying on the first number you hear.	A "was €200, now €120" tag makes €120 feel cheap — even if €120 is the real value.
Confirmation bias	Seeking evidence that supports what you already believe.	Only reading reviews that agree with the car you already want.
Sunk cost fallacy	Continuing because you've already invested, not because it's still worth it.	Sitting through a terrible 3-hour movie "because I paid for the ticket."
Overconfidence	Being more certain than your accuracy justifies.	"I'm 99% sure" — and being wrong a third of the time.
Planning fallacy	Underestimating how long and how much things will take.	Every home renovation, ever.
Hindsight bias	"I knew it all along" — after you find out the answer.	Claiming the market crash was "obvious" only once it happened.

Common mistake — "biases only affect other people": Smart, educated, well-meaning people are not immune. Knowing a bias exists removes only a little of its pull. That's why good decisions rely on processes and tools (next section), not on willpower or being clever.

28.11 Framing and loss aversion: the wording changes the choice

One bias deserves its own spotlight because it's so easy to fall for. A framing effect means the same facts lead to different decisions depending on how they're worded.

Example: A surgeon tells one patient "this operation has a 90% survival rate" and another "this operation has a 10% death rate." Identical facts. Yet patients (and doctors) consistently feel better about the first and choose differently. This is Kahneman and Tversky's "Asian Disease Problem."

Why is the loss framing ("death rate") so much scarier? Because of loss aversion: losses hurt about twice as much as equal gains feel good. Researchers put the multiplier around 1.5 to 2.5, with about 2 as the textbook figure. Losing €50 stings more than finding €50 delights.

This is the core of Prospect Theory (Kahneman & Tversky, 1979), the leading descriptive model of how people really treat risk. Its key insights:

People judge outcomes as gains or losses relative to a reference point (usually where they currently are), not as final wealth.
Losses loom about twice as large as gains (loss aversion).
People are risk-averse for gains (take the sure win) but risk-seeking for losses (gamble to avoid a sure loss).
Probability weighting: we overweight tiny probabilities and underweight moderate ones — which is exactly why the same person buys both lottery tickets (overrating a tiny jackpot chance) and insurance (overrating a tiny disaster chance).

Best practice — widen the frame: Before any important choice, deliberately restate it both ways. Reframe every loss as a gain and every gain as a loss. "10% chance of failure" is the same as "90% chance of success" — look at both, and your gut reaction stops steering you.

28.12 When others are deciding too: a peek at game theory

Some decisions don't depend only on chance — they depend on what other people choose. Game theory is the math of these interdependent choices: my best move depends on your move, and yours on mine.

Analogy: Rock-paper-scissors, or merging in heavy traffic — you can't choose well without modeling the other player's likely move.

Two ideas are enough to get the gist:

Dominant strategy: A move that's best for you no matter what the other side does. (Like wearing a seatbelt — better whether or not you crash.)
Nash equilibrium (after John Nash): A stable state where no player can do better by changing only their own move.

The most famous setup is the Prisoner's Dilemma. Two players each do better individually by "defecting" (betraying), yet if both defect they end up worse off than if both had "cooperated."

                  Other shop:
                  Keep price   Cut price
   You: Keep      both happy   you lose
                  (good,good)  (bad,great)
   You: Cut       you win      price war
                  (great,bad)  (bad,bad)  <- Nash

Example: Two rival shops. Each can cut prices to grab customers. Whatever the other does, cutting looks tempting — so both cut, both end up with thin margins, and neither dares raise prices first. That mutual price war is the Nash equilibrium: stable, but worse for both than if they'd both held prices.

Common mistake: Thinking a Nash equilibrium is the best outcome. It's only stable — nobody can improve by acting alone. The Prisoner's Dilemma proves stable and optimal can be very different things.

The good news: when the game repeats, cooperation can emerge. In political scientist Robert Axelrod's famous 1980 tournaments, the winning strategy was the simplest one, Tit-for-Tat: cooperate first, then just copy whatever the other player did last. It's nice (cooperates first), retaliatory (punishes cheating), forgiving (returns to cooperation when they do), and clear (easy to predict). That's a remarkably good recipe for long-term relationships, in business and life.

28.13 The prescriptive toolkit: what to actually do

Now the payoff. We know the ideal (EV, utility, Bayes) and our flaws (biases, framing, loss aversion). Here are the practical tools that close the gap. You don't need to be smarter — you need a better process.

Second-order thinking — "and then what?"

Don't stop at the immediate effect of a choice. Ask what that sets in motion, and what that sets in motion.

Analogy: A chess player who grabs a free pawn (first-order: feels good) and gets checkmated three moves later (second-order: disaster) versus one who thinks ahead. Investors like Howard Marks and Ray Dalio swear by it: "First-order: feels good now. Second-order: what does this cause?"

Mental models — bring a full toolbox

A mental model is a reusable concept from any field (supply and demand, compounding, base rates) that helps you make sense of a situation. The investor Charlie Munger urges building a "latticework" of many models from many disciplines.

Analogy: A handyman with only a hammer treats everything as a nail. Carry a full toolbox and you pick the right tool for each problem.

Checklists — defeat memory under pressure

A checklist is a pre-written list of must-check items. Pilots and surgeons run them before takeoff and incision — not because they're forgetful or dumb, but because memory fails under stress. Surgeon Atul Gawande's Checklist Manifesto showed simple checklists cut surgical complications dramatically.

Premortem — imagine it already failed

Before you act, imagine your decision has already blown up spectacularly, and ask "why?" This is psychologist Gary Klein's premortem, and it works because it flips your optimism off — it's far easier to spot risks when you're explaining a failure than predicting one.

Example: A month before a wedding, the planner says: "Imagine the day was a total disaster — what went wrong?" Out comes "the caterer was never actually confirmed." You found the landmine while you could still defuse it.

One-way doors vs. two-way doors — match speed to reversibility

Amazon's Jeff Bezos splits decisions in two. A two-way door is reversible — if it's wrong you just walk back through. A one-way door can't be undone. Decide reversible things fast; reserve slow, careful analysis for the irreversible ones.

Calibration and forecasting — track whether you're actually right

Being calibrated means that when you say "70% sure," you turn out right about 70% of the time. It's measured with the Brier score (0 = perfect, 0.5 = random guessing — lower is better). The crucial, hopeful finding from Philip Tetlock's Good Judgment Project: ordinary trained people can become excellent forecasters, scoring Brier values around 0.20–0.25 and beating credentialed experts. Calibration is a learnable skill — but only if you write predictions down and check them.

Decision journal — the single highest-leverage habit

Keep a notebook (or a doc). For each important decision, write: what you decided, why, the odds you'd give it, and what you expect to happen. Later, go back and compare. This one habit does three jobs at once: it builds calibration, it kills hindsight bias ("I knew it all along" — no, here's what you actually wrote), and it forces you to separate decision quality from outcome.

Better group decisions

Teams have their own failure modes — groupthink (everyone agreeing to keep the peace), anchoring on whatever the boss says first, and information cascades (everyone echoing the first loud voice). The fixes are process tricks: have everyone write their estimate privately before any discussion (the Delphi method), formally assign a devil's advocate, and pre-commit your decision criteria before you see the options.

Key takeaway: You can't control luck, and you can't fully out-think your own biases. What you can control is your process. Good tools — checklists, premortems, decision journals, base-rate-first thinking — make a bias-resistant process your default, so you don't have to be a genius in the moment.

28.14 Common traps, gathered in one place

Common mistakes to watch for:

Resulting — judging the decision by the outcome.
Base-rate neglect — ignoring background frequencies for vivid details.
Sunk cost fallacy — "I've spent so much, I can't quit now." (Ask instead: "Knowing what I know today, would I start this from scratch?" If no, quitting is the +EV move.)
False precision — treating a made-up probability as if it were measured.
Single-number predictions — giving one figure instead of a range, and never checking it.
"More information = better decisions" — past a point, extra info mostly adds noise and false confidence, not accuracy.
"Rational means cold and selfish" — no. Rational just means consistent with your own goals and values, generosity included.

28.15 Why this matters in real life

These ideas aren't academic. They show up in every expensive decision you'll ever make:

Money: Diversifying, ignoring sunk costs, not panic-selling in a crash (loss aversion in the wild), and buying insurance for the disasters you can't recover from rather than the cheap things you can.
Career: A job offer or a relocation is a decision tree — lay out the branches, estimate the odds, and know when quitting (Annie Duke's later work, Quit) is the smart move, not the failure.
Health: Reading a test result correctly with Bayes, and weighing a treatment's risk against its benefit instead of reacting to scary framing.
Business: Pricing, market entry, and competitive moves are game theory; big launches deserve a premortem; hiring with structured checklists beats gut feel.
Everyday consumer life: Spotting the decoy price, the "limited time only!" scarcity framing, and the default option a company quietly pre-checked for you.

The meta-reason it all matters: most expensive life mistakes are failures of process, not of knowledge. People rarely fail because they didn't know enough facts. They fail because they judged by outcomes, ignored base rates, fell for a frame, or threw good money after bad. The skill of deciding well transfers across every single domain — which makes it one of the highest-return things you can ever learn.

28.16 Your starter operating system

You don't need all of this at once. Start with five habits, and you'll already decide better than most people:

Keep a decision journal. Write what you decided, why, and the odds you'd give it. Review later.
Think in probabilities and ranges, not certainties. Replace "will it work?" with "what odds would I give, and what's my range?"
Start from the base rate, then adjust. Outside view first, inside view second.
Run a premortem before anything big and irreversible. Imagine it failed; list why; fix what you can.
Match your speed to reversibility. Reversible (two-way door) → decide fast. Irreversible (one-way door) → slow down.

Key takeaway for the whole chapter: You cannot control outcomes — only inputs. So pour your energy into the quality of the decision: separate it from the result, think in probabilities, anchor on base rates, watch for the predictable biases, and lean on simple tools to keep your process honest. Do that consistently, and the lucky and unlucky breaks will average out — while your judgment, the part you actually own, keeps getting better.

Continue reading

🧭 Ten Disciplines

Seeing the Whole: What a System Is and Why It Behaves the Way It Does

🧭 Ten Disciplines

The Machinery of Systems: Stocks, Flows, Feedback Loops, and Delays

🧭 Ten Disciplines

Leverage Points, System Traps, and Thinking in Systems (Advanced & Real-World)