An April 2026 AM Best survey of more than 150 rated insurers and MGAs found that 45% still cite data readiness as the top barrier to AI deployment, even among analytically mature organizations.
Insurance carriers have invested years building trusted analytics platforms. Data warehouses and governed metrics now deliver consistent loss ratios, premium trends and claims data that actuaries, underwriters and executives rely on. Yet when these same carriers launch AI initiatives, results often disappoint. Underwriting assistants hallucinate risk scores, fraud models produce unexplainable flags and catastrophe projections miss critical context buried in broker narratives and loss-run documents.
Your Analytics Layer Works - For What It Was Designed For
Analytics-ready data serves human decision-makers asking backward-looking questions. Actuaries calculating reserves, underwriters reviewing portfolios, compliance teams preparing filings and executives setting strategy all need consistent, summarized and explainable information. The central question is always the same: what happened?
The semantic layer defines business metrics, dimensions and access rules. It creates a single governed version of the truth so an actuary in New York and an underwriting director in London both get the same answer to "What was our Q3 combined ratio?" This layer excels at aggregation and stability. It compresses complex reality into clear, auditable numbers that humans interpret using their own judgment and domain knowledge.
AI Systems Are a Different Consumer Entirely
Large language models, reasoning engines and autonomous agents do not aggregate and summarize. They traverse, reason and act. They consume tokens, embeddings and context windows. They need properties that analytics infrastructure was never designed to provide:
- Full context and completeness behind every number.
- Provenance and trust: every fact carries its source, confidence level and resolution history for audit and regulatory defense.
- Semantic richness: named, typed relationships between policies, claims, insureds and risk factors so agents can reason without custom code.
- Conflict handling: the system preserves multiple sources, flags discrepancies and tracks resolutions rather than forcing a single version of the truth.
The enabling structure is an ontology - a knowledge layer that makes the domain understandable and navigable for machines.
Key Structural Differences
Analytics-ready data answers "what happened" by compressing thousands of transactions into governed numbers. AI-ready data answers "what should happen next" by preserving the full evidence trail behind every fact. A semantic layer translates business questions into consistent reports. An ontology maps the domain so machines can traverse it, weigh evidence and explain conclusions. Both are necessary. They serve different audiences through different infrastructure over the same foundational data.
The Right Architecture: Parallel Layers, Shared Foundation
Leading insurers build both capabilities over the same foundational event data (policies, claims, submissions). The semantic layer powers governed dashboards and regulatory reporting. The ontology gives AI agents the structure to reason, track provenance and produce explainable decisions.
The layers reinforce each other. Unified governance (lineage, confidence scoring, conflict rules and human review) serves both without compromise.
Why the Stakes Are Higher in Our Industry
Regulators demand explanations that dashboards cannot provide. The NAIC AI Model Bulletin and Colorado SB21-169 require clear, contemporaneous provenance for AI-influenced decisions. Most underwriting intelligence lives in unstructured documents that analytics systems were never designed to handle. Renewal workflows are change-detection exercises that benefit enormously when both layers work together.
Where to Start
Insurers with mature analytics platforms do not need to start over, but they must:
- Capture foundational events at the atomic level before aggregation.
- Protect your existing analytics layer. It works for actuaries and regulators.
- Build the ontology around one high-value use case, such as underwriting submission intake.
- Apply governance that serves both layers.
- Measure success separately: analytics by dashboard trust and reporting accuracy; AI by reduced hallucinations, explainability and business impact.
The Bilingual Insurer
The carriers that will lead the next decade are fluent in both data languages: one that helps humans trust and explain the past, the other that enables machines to reason about the future with transparency and confidence.
Your current analytics investment remains essential. It is also insufficient for AI. Building a parallel ontology capability is one of the highest-leverage moves you can make today. Closing the gap will unlock better risk selection, stronger regulatory resilience, lower fraud losses and genuinely intelligent operations.
