Beyond Heuristics: Context-Dependent Decision Rules

The most dangerous idea in behavioural science is that heuristics are shortcuts—mental rules that sacrifice accuracy for speed, universally applied across situations. This framing has calcified into orthodoxy since Kahneman and Tversky's foundational work, but it obscures something more interesting: the same cognitive mechanism produces wildly different outcomes depending on what decision environment it encounters.

We treat heuristics as if they are tools with fixed properties. A hammer drives nails poorly into water. But human decision rules are not hammers. They are adaptive systems that calibrate to context, and when we ignore that calibration, we misdiagnose both the problem and the solution.

Consider availability bias—the tendency to judge probability by how easily examples come to mind. In a stable environment where recent events predict future ones, availability is rational. A surgeon who recently encountered three cases of a rare complication is right to weight it more heavily in her next decision; the clustering signals something real about current conditions. But the same mechanism in a volatile or manipulated information environment becomes a liability. The availability of shark attack stories in summer news cycles has no bearing on actual risk, yet the heuristic fires identically. The mechanism is unchanged. The context is everything.

This is where most interventions fail. We design debiasing programs as if the goal is to suppress heuristics—to make people slower, more deliberate, more "rational." But the real problem is mismatch. A decision rule optimised for one environment becomes maladaptive when transplanted into another. The fix is not to eliminate the rule; it is to help people recognise which environment they are actually in.

Take anchoring. In negotiations, the first number mentioned disproportionately influences the final outcome. This is often presented as a cognitive flaw—people should ignore irrelevant anchors and reason from first principles. But anchoring serves a function in contexts where information is genuinely scarce. If you are buying a house in an unfamiliar market, the asking price is not irrelevant; it contains real information about what sellers believe the property is worth. The anchor is a signal. The problem emerges when anchors are deliberately planted by someone with asymmetric information and incentive to deceive. The mechanism does not change. The trustworthiness of the signal does.

This distinction matters operationally. If you believe anchoring is a flaw to be corrected, you train people to ignore numbers and think independently. If you believe it is context-dependent, you train people to ask: Is this anchor informative or manipulative? The first approach produces resistance and cognitive fatigue. The second produces discrimination.

The same applies to framing effects—the finding that logically equivalent choices produce different decisions depending on whether outcomes are described as gains or losses. Decades of research has treated this as evidence of irrationality. But in environments where losses and gains have asymmetric consequences—where losing £100 matters more than gaining £100—loss aversion is not a bias. It is calibration. The problem is not that people weight losses more heavily. It is that they weight them equally heavily across contexts where the stakes differ.

What changes when you see this clearly is the entire architecture of decision support. Instead of designing systems that override human judgment, you design systems that help people recognise the decision context they face and apply the appropriate rule. Instead of assuming one optimal decision process, you build flexibility. Instead of treating variation in choice as error, you treat it as signal.

This requires a harder kind of research than documenting biases. It requires mapping which heuristics perform well in which environments, and building tools that help people diagnose their situation before deciding. It requires accepting that the same person making the same choice in two superficially similar contexts might—and should—decide differently.

Kahneman's insight was that intuition is not infallible. The next insight is that intuition is not uniformly fallible either. Context is not noise in the decision process. It is the decision process.