The override problem in claims and underwriting

Overriding the model is sometimes the best decision in the building and sometimes the most expensive.

5 min read

Two kinds of override

One comes from experience that the model does not have. The other comes from discomfort with being told what to do by software.

In the data they look identical, which is why override rates on their own are a poor management metric.

When you score judgment behaviorally you learn whether people can articulate a defensible reason for going their own way.

Teams that can are usually right. Teams that cannot are usually slow and expensive.

Publish a small set of worked examples of good overrides and poor ones, drawn from real files with the details changed.

It is the single cheapest intervention available and it moves the score.