Algorithmic Decision Making and Model Explainability Preferences in the Insurance Industry: A Delphi Study

There is growing attention to the transparency, fairness, and explainability of algorithmic decision-making systems as they permeate many aspects of our lives. Despite the awareness of the need for algorithmic transparency and the right-to-meaningful-explanation provided by GDPR, little is known regarding what makes such explanations meaningful and useful. This issue becomes especially challenging in certain situations in which high levels of transparency may conflict with the best interest of organizations. The insurance industry poses an interesting case as the business model of insurance providers depends on the discrimination of customer groups. In this paper, we present the results of a Delphi study with experts from the Dutch insurance industry informed by an initial survey. Our results include the preferred explanation elements towards customers, from the perspective of the insurer, for five commonly used algorithmic decision-making systems. They also show that there is not a one-size-fits-all explanation approach, and that it depends on the system itself and the context in which it is used. Finally, the results highlight risks and concerns of the insurance experts regarding the disclosure of sensitive information in the form of explanations.