The Strangler-Fig Approach to Introducing Agentic AI in Insurance Platforms

Traditional core insurance systems were built to:

• Process policies
• Calculate premiums
• Manage claims
• Maintain regulatory records

They were not built to:

• Support real-time AI-driven decisioning
• Handle unstructured documentation
• Enable conversational interfaces
• Orchestrate autonomous workflows
• Integrate seamlessly across regions

As a result, insurers face two flawed options:

1. Full core replacement (high disruption)
2. Superficial AI overlays that lack governance

Neither option provides a sustainable path forward.

‍

Introducing Agentic AI into insurance platforms requires architectural discipline.

The recommended approach involves five progressive layers:

‍

Step 1: Wrap the Core With a Semantic Layer

Before introducing AI agents, insurers must:

• Standardize definitions (loss ratio, margin, risk score, TTM, etc.)
• Create an unambiguous data dictionary
• Document calculation logic and refresh cadence
• Define ownership and regulatory applicability

This semantic layer becomes the grounding context for AI.

Without it, agents will amplify ambiguity.

‍

Step 2: Introduce Transformation-Driven Gold Data Layers

Legacy insurance systems often store operational data in:

• Transactional schemas
• Region-specific databases
• Batch-driven reporting tables
• AI agents require decision-grade data.

Using a medallion architecture (Bronze → Silver → Gold):

• Raw operational data is transformed
• Business logic is encoded explicitly
• Cross-region inconsistencies are resolved
• Data becomes AI-ready and governed

This step is foundational.

Agents must operate on transformed, validated data — not raw transactional tables.

‍

Step 3: Extract Business Logic Into Middleware Services

In most legacy insurance platforms, business rules are:

• Hard-coded in stored procedures
• Embedded in underwriting engines
• Spread across rating modules

These rules must be abstracted into:

• Middleware services
• API-driven logic layers
• Version-controlled calculation modules

This enables:

• Transparency
• Auditability
• Gradual migration away from monolithic logic

Over time, decision intelligence shifts from the database to modular services.

‍

Step 4: Deploy Domain-Constrained AI Agents Outside the Core

Instead of embedding AI directly inside the core platform, deploy specialized agents around it.

Examples:

Claims Agent

• Validates documentation
• Detects anomalies
• Flags potential fraud
• Reconciles claim amounts against policy rules

Pricing Agent

• Evaluates risk thresholds
• Tests rate sensitivity
• Simulates underwriting scenarios

Compliance Agent

• Monitors regulatory triggers
• Tracks documentation completeness
• Generates audit alerts

These agents:

• Operate on governed gold-layer data
• Follow domain-specific SOPs (system prompts)
• Log all decisions
• Respect role-based access controls

They augment the core — without destabilizing it.

‍

Step 5: Gradually Shift Decision Intelligence Away From the Monolith

‍

Over time:

• Business logic moves into middleware services
• Decision-making moves into orchestrated AI layers
• Core database dependencies reduce
• Regional deployments become more flexible

The platform becomes:

• Database-agnostic
• API-first
• AI-ready
• Modular and scalable

This is modernization without disruption.

Applying the Strangler-Fig approach to Agentic AI enables:

Reduced Modernization Risk

No big-bang system replacement.

‍

Faster Time to Value

Agents deliver operational improvements early.

Improved Compliance

Continuous monitoring and structured audit trails.

Operational Efficiency

Automation of low-value, repetitive workflows.

Long-Term Platform Flexibility

Gradual evolution into modular, AI-ready architecture.