Problem Statement

Insurance processes involve complex decision-making under uncertainty, with significant consequences for accuracy, fairness, and trust. The introduction of AI into this domain offers the potential to improve efficiency, decision quality, and responsiveness, but also raises important questions about reliability, transparency, and robustness to manipulation.

Project Approach

In particular, three areas present both high potential for Agentic AI:

• Claims handling – intricate, multi-step workflows that require consistent evaluation and clear communication.
• Customer engagement and sales – matching products to individual needs in a timely, comprehensible way.
• Fraud detection – identifying increasingly sophisticated and AI-enabled attempts at deception.

We investigate how task-specific AI agents can be designed, implemented, and evaluated in these contexts. Across all three projects, the focus is on:

• Determining which tasks are technically and operationally suitable for AI support.
• Assessing impacts on decision accuracy, efficiency, and user trust.
• Ensuring compliance with regulatory requirements and resilience to adversarial behaviour.

By working with insurance partners to access representative, labelled datasets, we aim to develop prototypes, test them in realistic scenarios, and derive evidence-based guidance on the safe and effective use of AI agents in insurance.

Topics