PRA sets expectations for AI and machine learning use in insurance

The discussion paper notes that AI in the insurance sector can improve the efficiency of data processing and decision-making in underwriting and claims processing, reflecting growing operational deployment across the value chain. In life insurance, where products include an investment component, DP5/22 states that firms could use AI to support the investment choices of policyholders. In general insurance, the paper notes that AI could be used to automate claims management, extending machine learning into post-sale operations.

DP5/22 records that insurers can use AI to analyse new unstructured data sources, including telematics and data from wearable devices, to provide more tailored products and pricing. The supervisory authorities also highlight that these applications sit alongside existing legal and conduct requirements, which are cross-referenced in the annexes to the paper.

On risk, the discussion paper states that AI-related underwriting risks could lead to inappropriate pricing and marketing, raising concerns about product suitability and fairness. It gives the example of AI models trained solely on historical data that may not account for a breakthrough healthcare treatment, which could result in mispriced policies. The supervisory authorities link these issues directly to policyholder protection, one of the PRA’s primary objectives.

For claims, DP5/22 notes that risks such as concept drift and lack of explainability in AI claims management systems could affect policyholders’ ability to claim and their overall protection. The paper places these concerns within the existing expectations on firms to manage model risk and to ensure that decision-making processes remain strong over time.

The discussion paper also highlights risks from building AI models for cash-flow and capital reserve estimates, warning that inaccurate predictions and reserve levels could affect insurers’ ability to meet future liabilities. The supervisory authorities present these prudential risks alongside conduct and operational considerations, indicating that AI will be assessed across multiple regulatory dimensions.