Tokio Marine & Nichido Fire Insurance Co., Ltd.

In property and casualty insurance, claims professionals process large volumes of policyholder-submitted documents and images — repair estimates, medical records, photos of damaged property — to determine coverage and detect fraud. This review process has traditionally been manual, time-intensive, and prone to inconsistency, particularly when content spans both structured forms and unstructured sources like handwritten notes or photographs. For an enterprise insurer of Tokio Marine's scale operating across complex P&C lines, the cumulative burden on adjusters constrains throughput, introduces human error, and creates gaps that fraudulent actors can exploit. The cost of the status quo was twofold: operational inefficiency in routine claims intake and elevated exposure to fraudulent claims slipping through incomplete analysis.

Tokio Marine partnered with Shift Technology to integrate new generative AI capabilities directly into its claims fraud detection and intake workflows. The core enhancement is a Gen AI-powered visual intelligence feature that analyzes images and documents with greater precision than prior rule-based or purely ML-driven approaches. The implementation combines three layers: generative AI for contextual interpretation of unstructured content, machine learning for pattern recognition and anomaly scoring, and intelligent document processing for structured data extraction. This layered architecture allows the system to handle heterogeneous claim submissions — from scanned forms to photos of damaged property — within existing claims operations rather than requiring a separate adjudication step. Shift Technology's platform provided the underlying AI infrastructure, enabling Tokio Marine to extend fraud detection capabilities without rebuilding its core claims systems.

The Gen AI visual intelligence capability is now supporting Tokio Marine's claims operations with improved precision in data extraction and classification across both structured and unstructured document types. While specific quantitative benchmarks have not been publicly disclosed, the insurer has identified the technology as a meaningful tool in reducing manual workload for claims professionals and strengthening fraud identification accuracy. Qualitative outcomes include:

• Improved adjuster efficiency through automated extraction and categorization of claim documents and images
• Enhanced fraud signal detection by surfacing anomalies in visual and documentary evidence that manual review may miss
• Broader societal mandate alignment, with Tokio Marine framing the deployment as part of its commitment to protecting customers and society during times of need

• Combining generative AI with machine learning and intelligent document processing creates more robust fraud detection than any single technique alone — each layer compensates for the other's blind spots.
• Visual intelligence capabilities are increasingly critical in P&C claims, where image and document quality varies widely and unstructured content is the norm.
• Enterprise insurers should integrate AI into existing claims workflows rather than creating parallel systems, reducing friction for adjuster adoption.
• Fraud detection AI delivers dual value: operational efficiency gains and risk reduction — quantifying both separately strengthens the business case internally.