Artificial intelligence (AI) is moving rapidly from experimentation to execution in the property/casualty insurance industry. From AI-driven claims estimation to predictive underwriting and fraud detection, insurers are realizing tangible value in faster decisions, improved accuracy and stronger customer experiences. Yet, for every success story, there are stalled initiatives and missed expectations.
Executive Summary
Ajay Kelshiker, co-founder of InsurTech data and analytics software provider Percipience, believes the success of AI implementations in the P/C insurance space depends on having trusted, high-quality data. Here, he advocates for building a strong data foundation through modern architectures like Data Lakehouse with Medallion layers, Microsoft Fabric for unified analytics and AI, and Data Mesh for decentralized governance.He also explains some simple explanations of these terms in a related article: Demystifying the Data Landscape: Lake, Warehouse and Lakehouse Explained
The difference often comes down to one critical factor: data readiness.
Most P/C insurers operate within a patchwork of legacy core systems (policy, billing and claims), bolstered by spreadsheets and specialized departmental applications. These siloed environments create inconsistent definitions, incomplete records and limited visibility across the enterprise. When AI models are trained on fragmented or poor-quality data, they inevitably underperform—eroding trust and slowing adoption.
Data First, AI Second
Rather than rushing to build narrowly focused AI models, insurers and their partners must first strengthen their data foundation. Without ensuring data quality, lineage and accessibility, no AI initiative will achieve success across the enterprise.
