The biases and how to counter them

Four biases, four counters

Anchoring

The first number spoken determines all the rest. If the facilitator asks "would you say €10 million?", estimates will cluster around €10 million.

Counter: individual written estimation before any discussion. Participants write their figure, you collect them, and only then discuss the spread.

Availability

We overestimate what we remember. A recent or heavily covered incident crushes the estimate; a risk that has never materialised is underestimated.

Counter: bring external data — published sector losses, consortium databases, authority reports — before the estimation.

Overconfidence

Intervals supplied by experts are systematically too narrow. A "90% confidence" interval contains the true value in roughly 50% of cases.

Counter: explicitly ask for the extreme value — "what amount would be exceeded only once in a hundred times?" — then check the proposed interval is consistent with it.

Conformity

In a group, estimates converge on the most senior or most assertive participant's.

Counter: anonymous estimation, feedback of the spread without naming authors, and discussion of the reasons for divergence rather than of the people.

Separating frequency and severity

A frequent methodological error is asking directly "what annual loss do you expect?". The answer blends two judgements of different natures.

Estimate separately:

• Frequency: "how many times does this event occur in ten years, in an organisation like ours?";
• Severity: "if the event occurs, what is the median loss? the 95th percentile loss? the maximum plausible loss?".

Experts are far better at conditional severity than at frequency. It is therefore useful to feed frequency from external data and reserve expert judgement for severity.

Bow-tie as structure

Before quantifying, structure. The bow-tie links:

• on the left, the causes and preventive barriers;
• in the centre, the top event;
• on the right, the consequences and protective barriers.

This structure yields three benefits: it makes the scenario debatable in committee, it exposes missing barriers, and it allows separate estimation of the event's probability and of the distribution of its consequences.

It is also the representation a non-specialist executive grasps fastest — which, in a risk committee, is not a detail.