Duck Creek AI platform targets underwriting and claims
Duck Creek Technologies has launched an insurance-specific agentic AI platform designed to deploy, orchestrate, and govern AI agents across property and casualty workflows - targeting what Boston Consulting Group estimates could represent up to $80 billion in annual industry impact in the United States alone.
The Duck Creek Agentic AI Platform combines generative AI, machine learning, and neuro-symbolic reasoning - a hybrid approach pairing probabilistic AI with deterministic rules-based logic. The company argues this combination is essential for insurance workflows where decisions must satisfy regulatory requirements and be fully explainable.
The platform includes a proprietary insurance-domain Model Context Repository (MCR), an orchestration layer for designing and deploying agents, and a built-in governance module providing decision traceability, auditability, and compliance controls.
Two initial agentic applications accompany the platform launch. The Agentic Underwriting Workbench automates submission intake, triage, and enrichment, prioritising high-value opportunities and preparing decision-ready submissions to reduce quote turnaround times.
The Agentic First Notice of Loss (FNOL) application handles claims intake across digital, voice, and mobile channels, applying AI to policy verification, data validation, and early fraud detection. The FNOL application was developed in collaboration with Google Cloud using Gemini models.
The platform is structured in five layers: agentic intelligence (the insurance domain model); orchestration (agent design and deployment); AI Assurance (governance and auditability); AI Gateway (ecosystem integration including MCP and A2A protocols); and native integration with Duck Creek's core policy, billing, and claims systems.
For insurers already running Duck Creek core systems, the platform leverages existing data, configurations, and APIs rather than requiring data migration. The announcement states the system can also integrate with other core systems, though specifics were not provided.