AI in Insurance: Why It Fails Without a Unified Data Layer

For most insurance companies, the BI stack is a maze of disconnected systems and conflicting definitions that usually looks like this: 

• Data scattered across legacy policy admin systems, claims platforms, and CRMs, each with its own schema and logic.
• Multiple BI tools like Power BI, Tableau, Qlik, Excel, and SAS operate as isolated islands, each rebuilding the same metrics differently.

• Critical metrics like “loss ratio” are calculated differently across teams, producing conflicting dashboards and eroding trust in the numbers.

• Governance held together by spreadsheets, SharePoint docs, and tribal knowledge, with no consistent source of truth.

This kind of environment breaks the one thing AI depends on: consistency. 

When definitions change across systems and metrics are rebuilt in every tool, AI models don’t just inherit bad data; they inherit conflicting logic. The result? They produce answers that can’t be trusted or scaled with confidence.

The real issue isn’t the BI stack or the AI model. It’s the absence of consistent logic between all these systems. 

AI doesn’t fail because of algorithms. It fails because the logic underneath the data is fragmented. When it’s consistent, AI can deliver accurate, reliable answers, regardless of a department’s workflow. 

But when core business logic for KPIs is trapped inside individual BI tools, SQL scripts, or departmental dashboards, the AI is effectively working across multiple versions of the truth. In practice, it’s forced to rely on: 

• Conflicting definitions: like Product and Finance using different formulas for “Net Earned Premium,” leading to different outputs for the same question.
• Fragmented lineage: where the same metric is calculated differently across various systems, with no way to trace which version is correct.
• Missing context: because no single source of truth exists to anchor how metrics are defined, calculated, and interpreted.

Without unified data logic, AI produces answers that teams can’t align or act on. 
This means two teams can ask the same question and get two different answers, both technically correct, but neither usable for decision-making. 

When business logic is embedded inside individual tools, it creates a chain reaction across the entire organization. 

An AI model might provide an underwriter with one Combined Ratio, while the actuarial dashboard shows a different number because it treats reinsurance recoverables differently. 

Both outputs are technically correct, but neither can be trusted as a reliable source of truth. 

This kind of inconsistency carries directly into pricing, reserving, and risk decisions, leading to mispriced policies, premium leakage, and delayed action. 

In a highly regulated environment, "close enough" isn’t acceptable. If leadership can’t trace how a number was calculated, they can’t rely on it, especially when AI is involved.  

As a result, many AI initiatives stall at the proof-of-concept stage, not because the models fail, but because the logic behind the data can’t be verified.  

To compensate, insurers fall back on what’s familiar: manual validation.  

They rebuild metrics in Excel, cross-check dashboards, and rely on offline calculations just to verify what should already be trusted. Validating the data takes more time than actually acting on it. 

This was exactly the challenge faced by Helsana, Switzerland’s leading health insurer. 
Their problem was twofold: the sheer volume and complexity of medical data, and its siloed nature across disconnected legacy systems. 

They were trapped in a cycle of fragmented governance, where different business units were using the same data but understanding it differently.   

Every report required manual reconciliation, slowing down decisions and creating delays across the organization. 

For Helsana, the shift came from unifying how business logic was defined and applied across their systems, instead of leaving it embedded inside individual tools.  

By introducing a universal semantic layer over their existing data stack, they were able to define key metrics once and apply them consistently across every department and use case.  

This transformed a fragmented data landscape into a governed, consistent foundation for analytics, where metrics remained the same whether accessed by a business user, analyst, or AI model. 

Discover the full story

In this environment, AI models no longer operate across conflicting definitions. Instead, they are grounded in a single layer of logic that powers every tool, app, and workflow with consistent insights. 

What Helsana built isn’t just a cleaner data environment.  
It’s a different architectural model. 

A universal semantic layer sits between your data and every consuming system. It creates a standardized map of your insurance KPIs, ensuring that every system works from the same definitions and logic.   

This is what makes consistent, scalable AI possible. Without it, AI in insurance doesn’t scale; it fragments. With it, every system operates from the same foundation. 

It allows AI to move from isolated use cases to something that can actually operate across the business. 

Strategy Mosaic is how this model is implemented in practice, giving insurers a governed, consistent foundation for both analytics and AI. 

It enables teams to modernize reporting without losing control over how data is defined, calculated, and used. 

Strategy Mosaic ensures that whether a query comes from a senior executive looking at Combined Ratios, a field agent checking policy endorsements, or an autonomous AI agent triaging a claim, the answer is always the same: accurate, governed, and trusted.