All the machinery so far exists to serve one moment: the choice. This chapter turns beliefs and preferences into a defensible decision — and tells you when to pay for more certainty first.
Probabilistic reasoning describes the world; it does not tell you what to do about it. For that you need to fold in preferences — what outcomes are worth to you — and combine them with your beliefs. Simple decisions are the one-shot version of that problem, and they contain the ideas that make risk analysis quantitative rather than merely descriptive.
1From coherent preferences to a utility
Suppose your preferences over uncertain outcomes obey a few reasonable conditions — you can compare any two options, your preferences don’t form contradictory loops, and mixing in a little of a better option never makes things worse. A foundational result says that any such coherent preferences can be captured by a single number, a utility, assigned to each outcome, such that the rational choice is the one with the highest expected utility — the probability-weighted average of the utilities it might yield.
This is the quiet engine under all of decision analysis: don’t chase the best possible outcome, and don’t merely avoid the worst — weigh each outcome by how likely it is and how much it’s worth, and pick the option whose weighted total is highest.
2Why a rational manager is risk-averse
Crucially, utility is not money. For almost everyone the utility of money is concave: the first \$100k of a loss hurts more than the tenth \$100k, and a windfall’s tenth dollar means less than its first. Concave utility produces risk aversion as a logical consequence, not a character flaw.
That single fact explains behaviour that “expected monetary value” alone calls irrational. You will accept a certainty equivalent — a guaranteed outcome worth less than a gamble’s expected value — simply to be rid of the variance. The gap between the two is the risk premium: what you’ll rationally pay to avoid exposure. It is why firms carry contingency, buy insurance, and prefer a reliable \$1M to a coin-flip between \$0 and \$2.2M, even though the flip “wins” on average.
3Laying the decision out: decision trees
To structure a real choice, extend the network with two new kinds of node: a decision node for the levers you control, and a utility node for what you value. The unrolled picture is the familiar decision tree — square decisions branching into round chance events branching into outcomes. You solve it by working backwards: average the utilities at each chance point by their probabilities, then at each decision point keep the branch with the highest expected utility. What’s left is the optimal policy and its value.
4What is more certainty worth? Value of information
The most valuable idea in this chapter for a risk manager is the value of information. Before commissioning a geotechnical survey, a prototype, or a market study, you can compute how much better your decision could become if you knew the result in advance. That expected improvement is a hard ceiling on what the information is worth paying for.
Sometimes it’s large — the survey could flip your decision, and paying for it is obviously right. Sometimes it’s zero: if no possible result would change what you do, the study is worthless no matter how interesting, and buying it is theatre. Value of information turns “should we investigate first?” from an instinct into a calculation, and it is one of the highest-leverage moves in the entire discipline.
Rational choice under uncertainty means maximising expected utility, not expected money. That one substitution makes risk aversion, contingency, insurance, and the price of information all fall out as consequences rather than exceptions.
Representation gave us a shared model of uncertainty; inference let us update it with evidence; parameter and structure learning let us build and calibrate it from data. Simple decisions closed the loop — turning all of it into a single, defensible choice. Everything in Part I, though, assumed the decision was a one-shot. Real projects unfold over time, where today’s choice reshapes tomorrow’s options.
Score your options by probability-weighted value to the business, not by best case or worst case alone — and make your risk appetite explicit as a utility curve rather than leaving it to whoever is loudest in the room. Lay real decisions out as trees and fold them back; the discipline routinely overturns the gut answer. And before you spend on studies, surveys, or pilots, ask what result could actually change your decision. If none could, you already have your answer — keep your money.