Five well-documented cognitive biases shape underwriting decisions: status quo bias, loss aversion, anchoring, availability bias and the endowment effect. Each one pushes carriers toward keeping old models, blocking whole areas, trusting the first score, overreacting to the last catastrophe, or favoring the existing book. None can be removed; all can be designed around.
The five biases at a glance
| Bias | What it is | How it shows up in underwriting | What counters it |
|---|---|---|---|
| Status quo bias | An irrational preference for the current state of affairs (Samuelson and Zeckhauser, 1988). | Carriers keep wildfire models built on slope, fuel and access because they are familiar, not because they perform best. | Run the old and new model in parallel and compare results before switching. |
| Loss aversion | A loss hurts roughly twice as much as an equal gain pleases (Kahneman and Tversky, 1979). | After a major fire, whole ZIP codes are shut down, blocking insurable addresses with the risky ones. | Restrict the specific addresses above the risk threshold instead of the whole area. |
| Anchoring | The first number seen becomes the reference point for every later judgment (Tversky and Kahneman). | A legacy or prior-year score colors each new assessment, so old errors persist year after year. | Blind scoring: form the new judgment before the previous score is shown. |
| Availability bias | Events that are easy to recall are judged more probable (Tversky and Kahneman, 1973). | Underwriting tightens after a bad season and loosens after a quiet one, although the underlying risk barely moved. | Anchor decisions to long-term models, updated on a schedule rather than in reaction. |
| Endowment effect | People value what they already own more highly than what they do not (Thaler, 1990). | A policy that would be declined as new business is renewed because it is already in the book. | Renewal re-underwriting: periodically score the whole book as if every policy were new. |
Why underwriting is especially exposed
Underwriting combines the conditions under which these biases are strongest: decisions under uncertainty, rare but vivid losses, feedback that arrives years later, and a long institutional memory of what was done before. A decision that feels safe, such as keeping the model the regulator already knows or closing a ZIP code after a fire, is often the expensive one.
The common remedy
None of the five is solved by asking people to be less biased. Each is addressed by changing what information is shown, in what order, and at what resolution. When risk is visible address by address, the impulse to shut everything down loses its grip; when the prior score is hidden until a judgment is formed, the anchor has nothing to hold.
The series index is at Behavioral economics and insurance.
Common questions
What is the most costly bias in insurance underwriting?
They compound, so they are hard to rank. Status quo bias keeps weaker models in use; loss aversion and availability bias then drive broad, reactive restrictions on top of them; anchoring and the endowment effect keep past errors in the book.
Can data eliminate cognitive bias?
No. Bias is part of human decision-making. Data and process can route around it: granular risk data replaces all-or-nothing choices, and procedures such as blind scoring remove the trigger.
What is blind scoring?
Presenting a new risk assessment without the previous score visible. The underwriter forms an independent judgment first and compares it with the prior assessment afterwards.