Decision Making

Choosing Under Constraint

Management's core act is choosing. The theory of how comes in three models, arranged by how much of the ideal survives contact with reality.

The rational model — the textbook eight-step machine: identify the problem, set criteria, weight them, generate alternatives, evaluate each against every criterion, choose the maximum, implement, evaluate. It assumes complete information, clear preferences, unlimited computing time, and a single goal: maximize.

Bounded rationality (Simon): real decision-makers face limited information, limited attention, and limited time — so they build simplified models and satisfice: search until the first alternative that is good enough, then stop. Not laziness; arithmetic. When search is costly, stopping early is often the rational strategy about being rational.

Intuition: pattern recognition compiled from experience — fast, holistic, and worth exactly as much as the experience base behind it. Expert intuition in stable, feedback-rich domains (chess, firefighting) is real; 'gut feel' in noisy novel domains is usually noise with confidence.

Where choices bend: systematic biases distort every model:

• Anchoring — the first number gravitationally pulls the estimate.
• Confirmation — hunting evidence for what we already believe.
• Sunk cost — throwing good money after bad because the bad is already spent.
• Overconfidence — the planning fallacy's engine.
• Availability — judging frequency by memorability.
• Escalation of commitment — sunk cost with a public audience.

Risk arithmetic: under risk, alternatives are compared by expected value — each outcome weighted by its probability:

$$EV = \sum p_i \cdot v_i$$

A 60% chance of 50 and a 40% chance of losing 20 is worth $0.6 \times 50 - 0.4 \times 20 = 22$: the number a rational chooser compares against alternatives, and the number sunk costs are forbidden to touch — only future consequences count.

Three models of choosing

Common pitfall: Treating satisficing as a character flaw. When search is costly, stopping at good-enough is the optimal policy — the error is pretending you maximized when you actually stopped early.