Behind the Agent: How Admiral Handles Insurance Calls with AI Agents

From a first policy question to a claim filed in the middle of a crisis, insurance customers call when it matters most. The quality of that conversation determines everything that comes after. Admiral handles millions of customer interactions a year across the UK, Italy, France, and Spain - and they're now leveraging AI agents to help with this. A multimodal, multi-channel & multilingual agent that authenticates callers, detects intent, handles sensitive conversations like bereavement calls, and hands off to a human without losing context. Dominika Kampa, Group Head of GenAI at Admiral, will walk through exactly how they built it - the architecture, the testing framework they developed to validate agents across languages before go-live, and what broke in production that nobody warned them about. What we'll cover - Admiral's North Star for AI transformation and the roadmap to get there - Why regulated industries are actually well-suited for voice AI - and what the compliance constraints force you to get right - How Admiral authenticates callers through a 5-point ID check without the conversation feeling clinical - The multi-agent handoff: how context passes from one agent to the next so customers never have to repeat themselves - Building "Deimos" - Admiral's custom testing framework for running agents across multiple languages at scale before deployment - Rolling out across markets: what changes country to country, and what doesn't - What failed in production - guardrail edge cases, regional language surprises, and the internal buy-in battles nobody tells you about