Artificial intelligence has moved from pilot projects to production in many property and casualty insurers. Claims estimation, predictive underwriting, and fraud detection are delivering measurable gains in speed, accuracy, and customer experience. Yet we see too many AI initiatives stalling or falling far short of expectations. The determining factor is rarely the quality of the algorithms, but mostly the readiness of the underlying data.
Most P&C carriers continue to rely on decades-old core systems for policy administration, billing, and claims. These platforms are supplemented by departmental applications and extensive use of spreadsheets. The result is fragmented data with inconsistent definitions, missing values, and limited enterprise visibility. When AI models are trained on this foundation, performance suffers and business users lose confidence.
Successful AI deployment requires a deliberate sequence: build a robust data foundation before scaling model development. Three architectural approaches are proving effective in the insurance industry.
Data Lakehouse with Medallion Architecture
A lakehouse combines the low-cost storage of a data lake with the governance and performance of a data warehouse. The Medallion layered approach...Bronze, Silver, and Gold, progressively refines raw data.
Bronze holds raw ingested records.
Silver applies cleaning, standardization, and validation rules.
Gold delivers curated, business-ready datasets with full lineage.
This structure gives data teams repeatable processes while maintaining auditability across claims, policy, and exposure data.
Microsoft Fabric for Unified Analytics and AI
Fabric provides a single platform that integrates data engineering, data science, real-time analytics, and business intelligence. For insurance leaders, this eliminates the need to maintain multiple tools and complex data pipelines. Actuarial, underwriting, and claims teams can work from the same governed data without duplicating extracts. The tight integration between lakehouse storage and AI services accelerates both model training and operational deployment.
Data Mesh for Decentralized Governance
Centralized data teams often become bottlenecks as insurance organizations grow more complex. Data Mesh shifts ownership to domain teams, underwriting, claims, actuarial, and distribution, while still enforcing enterprise standards for discoverability, security, and quality. Each domain treats its data as a product, complete with service-level agreements. This model improves data timeliness and accountability without sacrificing control.
Executives responsible for data and analytics should focus on four priorities:
In our experience, insurers that invest in these foundations report faster model deployment cycles, higher model accuracy, and greater adoption by frontline users. Underwriters receive more reliable risk scoring. Claims teams see improved triage and estimation. Actuaries gain better exposure data for catastrophe modeling and ratemaking.
The gap between AI experimentation and sustained value is not closing through better models alone, but through disciplined investment in modern data architecture. Data should be a strategic asset...not a byproduct of core platforms. You will be best positioned to use AI when the business treats data as that strategic asset.