Embedded COBOL Logic And Premium-Parity Risk Were Blocking Safer Policy Modernization
For a leading U.S.-based property and casualty insurer, policy administration sat at the centre of underwriting, rating, policy issuance and servicing across multiple products and lines of business. Its portfolio spans personal insurance, commercial insurance and employee benefits, covering auto, home, specialty, workers’ compensation, commercial property, liability, commercial auto, cyber liability, professional liability, group life, disability insurance and paid family and medical leave programs.
Over decades, that legacy environment had become deeply embedded in the business, making modernization urgent but highly sensitive.
The legacy environment was built on IMS/COBOL and supported by JCLs, copybooks, IMS database structures, and FICO Blaze Advisor rules. It still supported older policies and underwriting processes, but it had become increasingly difficult to maintain, enhance, and scale. The client wanted to migrate to Duck Creek Policy Administration, yet the business rules and data logic required for that move were buried across thousands of legacy artifacts. Manual modernization was proving slow, risky, and cost-intensive.
The pressure showed up in five ways:
- Legacy complexity and embedded business logic: Underwriting logic, rating rules, validations, dependencies and product-specific behavior were scattered across COBOL, IMS/DB2 and rules engines.
- Talent and support constraints: IMS/COBOL expertise was getting harder to find, increasing dependency on a shrinking specialist pool.
- High maintenance cost and continuity risk: Older policies were still being underwritten and managed through the legacy platform, making failure costly in both operational and business terms.
- Modernization risk: Migrating to Duck Creek without complete visibility into legacy rules could create incorrect underwriting behavior, rating mismatches, premium discrepancies or compliance exposure.
- Limited agility and scalability: The platform slowed product changes, constrained rollout speed, raised costs, and limited modernization at scale.
The client had already attempted modernization through a traditional approach and that effort had not succeeded. Manual discovery and rule extraction were infeasible at the speed and confidence the business required. The challenge was no longer whether modernization should happen, but how to move a mission-critical legacy estate without losing pricing accuracy.
Building A Safer Modernization Path into Duck Creek
Persistent addressed the challenge by deploying SASVA™, its GenAI-powered software engineering platform, to accelerate legacy discovery, business rule extraction, Duck Creek manuscript generation, data migration, and validation. The work began with a focused modernization wave covering the identified insurance line of business, creating a controlled proving ground for accuracy, automation, and scalability before broader rollout. The success criterion was explicit: migrate the logic and policies to Duck Creek while staying within the client’s accepted premium variance of ±5%.
The approach combined AI-driven automation with the controls that mattered in insurance modernization:
- Focused modernization wave for business-critical validation using real policy logic and test data
- AI-driven legacy discovery across COBOL programs, JCLs, copybooks, IMS database structures and Blaze Advisor rules
- Automated rule extraction and translation into Duck Creek-ready business rule definitions
- Duck Creek manuscript generation for underwriting rules, rating logic, policy validations and intermediate artifacts
- End-to-end data migration with quality controls through extraction, transformation, validation, XML generation and a Data Quality Engine with more than 150 checks covering structural integrity, profiling rules, mandatory fields and migration accuracy
- Human-in-the-loop validation to ensure business correctness, compliance alignment and premium accuracy
McKinsey noted in April 2026 that agentic AI may finally help insurers modernize core technologies that have long resisted transformation. That is relevant here because the challenge was not simple code conversion; it was the extraction and preservation of policy logic that directly affected underwriting and premium outcomes.
The value was not just in accelerating technical discovery. It was in doing so with business control intact.
Where Legacy Discovery Turns into Migration-Ready Assets
At the core of the solution was SASVA™, an AI-powered modernization accelerator built to decode legacy insurance systems and translate embedded logic into Duck Creek-ready assets. Instead of treating migration as a one-time conversion event, the platform turned it into a structured workflow with traceability, validation and parity checks built in.
The workflow operated through six coordinated components:
- Discovery Agent: Scanned COBOL, JCLs, copybooks, IMS structures and rules-engine artifacts to identify in-scope components, dependencies and architecture flows.
- Logic Extraction Agent: Analyzed COBOL and Blaze Advisor logic to surface underwriting rules, rating conditions, validations, decision paths and embedded business behavior.
- Rule Translation Agent: Converted extracted legacy logic into Duck Creek-ready business rule definitions and manuscript inputs.
- Manuscript Generation Engine: Created Duck Creek-compatible manuscripts for underwriting, ratingand policy validations, reducing manual rewriting effort.
- Data Migration and DQ Engine: Supported extraction, transformation, validation, XML generationand 150+ data quality checks.
- Validation and Parity Engine: Compared legacy and Duck Creek outputs to confirm rule consistency and premium parity. In the engagement, six policies were migrated and validated across both environments, with premiums matching exactly.
Deloitte reports that as gen AI programs scale, data integrity, model accuracy, human oversight and trust become essential to production quality. That principle maps closely to this modernization effort, where confidence came not from automation alone but from parity validation, data-quality checks, masking controls and expert review.
Together, these capabilities created a repeatable AI-led modernization framework that could analyze legacy code, extract business rules, generate target-ready artifacts, migrate data and validate outcomes with human oversight. That is what turned modernization from a risky manual excavation into a more controlled path to migration.
In insurance modernization, speed matters. But certainty matters more.
Business Impact
From Legacy Discovery to Premium-Accurate Migration
The engagement delivered measurable modernization gains across discovery, rule extraction, migration and validation:
- 100% premium parity achieved against an allowed premium variance of ±5%
- 112,588 lines of legacy COBOL distilled into 12 validated Duck Creek-ready business rules in 2 days
- 68–70% overall automation across the modernization lifecycle
- 84% automation for COBOL rules
- 69% automation for Blaze rules
- 66% automation in data migration
- Manual effort compressed from six to nine months to days
- Reduced modernization risk through domain review, data quality checks, parity validation and governance controls
For the client, the visible result was acceleration. The strategic result was proof that legacy insurance logic could be migrated with both speed and premium accuracy.
Premium parity was not just a metric. It was the permission to scale.
Extending The Blueprint Beyond the First Wave
Following the success of the initial modernization wave, the client gained a validated blueprint for scaling AI-led modernization across additional states, lines of business and policy products. The engagement showed that mission-critical insurance logic could be extracted, translated, migrated and validated with speed and control.
The foundation now supports a broader agenda:
- Scaling Duck Creek migration beyond the initial auto insurance scope
- Accelerating mainframe decommissioning by reducing long-term dependency on aging IMS/COBOL systems
- Building a reusable modernization factory with playbooks, templates, accelerators and validation models
- Improving business and IT confidence through premium-parity validation and preserved business rules
- Creating a safer path for AI in core insurance systems with context, governance, human oversight and domain expertise
Why Persistent Was Chosen
Persistent was chosen for reasons that were both technical and business-specific:
- AI at enterprise scale: SASVA could understand code semantically, not just syntactically.
- Insurance and Duck Creek expertise: The team combined AI-led modernization with domain knowledge and Duck Creek implementation experience.
- Speed with governance: Automation was balanced with masking instructions, redacted test data, validation checks and human review.
- Outcome-driven execution: The engagement stayed anchored on the business outcome that mattered most: premium parity.
- Reusable modernization blueprint: The work was designed as a scalable model, not a one-off migration exercise.
This engagement demonstrated how AI-led engineering can redefine legacy modernization in insurance. More than a successful proof point, it created a foundation for scaling modernization, reducing legacy dependency and building a more agile insurance technology landscape. It proved that decades of COBOL complexity can be converted into Duck Creek-ready business rules in days, while preserving premium parity, compliance alignment and business continuity. For the client, this was more than a successful proof point. It was a foundation for scaling modernization, reducing legacy dependency and building a more agile insurance technology landscape for the future.
Assess Your Legacy Migration Risk. Validate A Premium-Parity Path. Talk with Persistent.
