How GenAI is impacting Legacy Policy Administration Systems

Most insurance carriers are running their policy administration on platforms designed when the iPhone didn't exist. The PAS is the system of record for premiums, policies, endorsements, renewals, billing and claims linkage — and for many insurers it is also the single biggest blocker to launching new products, opening new digital channels, and meeting regulator timelines. The cost of doing nothing is no longer theoretical: it shows up in lost quotes, failed integrations, audit findings and a steady drain of engineers who refuse to touch COBOL.

Generative AI changes the economics of fixing this. Modernisation programmes that used to take three to five years and consume entire IT budgets can now be re-scoped into 9-to-18-month tracks, because GenAI absorbs the most expensive parts of the work: reading undocumented code, mapping business rules, generating tests, and rewriting modules in modern languages. This piece sets out where GenAI is actually moving the needle on PAS modernisation, and what insurers should do about it before competitors close the gap.

Why legacy PAS platforms have become a strategic liability

The PAS sits at the centre of every insurance operation. It calculates premiums, issues policies, manages endorsements, drives renewals, handles billing, holds customer data, and feeds compliance reporting. When it works well, nobody notices. When it doesn't, every downstream system — claims, distribution portals, partner APIs, broker platforms, analytics — inherits the problem.

The platforms most insurers depend on were built in the 1990s or early 2000s. They have served well, but three structural problems now dominate boardroom conversations.

Technical debt that compounds every quarter

Older PAS estates run on COBOL, PL/I, Fortran or proprietary 4GLs that few new graduates learn. The architecture is monolithic, the documentation is partial at best, and product logic is woven into stored procedures and batch jobs that nobody is willing to refactor. Every new product launch adds another layer of conditional logic, making the next change harder than the last.

Integration limitations that block digital strategy

Modern distribution depends on APIs, event streams and embedded experiences. Legacy PAS platforms expose flat files, nightly batches and screen-scrape interfaces. Connecting them to a mobile app, a broker portal, an aggregator, or a partner ecosystem requires middleware, manual reconciliation and a long tail of exception handling. The result: digital initiatives stall at the PAS boundary.

Operational risk the regulator will eventually price in

Outdated encryption, weak access controls, missing audit trails and gaps against current data-protection regimes turn legacy PAS into a compliance exposure. Add the knowledge-retention problem — the engineers who actually understand the code are retiring — and the operational risk profile worsens every year, even with no new business growth.

What Generative AI actually brings to PAS modernisation

Generative AI is not a single product. For insurance modernisation it is a set of capabilities — large language models, code-trained models, retrieval-augmented reasoning, agentic workflows — applied to specific PAS problems. Unlike rule-based automation, GenAI can read unfamiliar code, infer intent, summarise behaviour, generate equivalents in modern languages, and propose test cases that exercise both the documented and the undocumented paths.

Adoption is moving fast. Insurance industry surveys point to a roughly 65% jump in GenAI use across policy administration workloads since 2022, with carriers reporting double-digit gains in developer productivity and measurable improvements in code quality on modernisation programmes. The leaders are no longer experimenting; they are operationalising.

The most valuable plays we see at Redian, working with banks, insurers and brokers across our BFSI practice, cluster into seven areas.

Decomposing monoliths into services

GenAI can read a legacy module, identify cohesive business capabilities, and propose a service boundary — quotes, endorsements, renewals, billing, claims handoff. This shortens the most painful part of any modernisation: deciding what to carve out first.

Reconstructing business rules

Hidden inside every legacy PAS are thousands of rules — rating factors, eligibility checks, underwriting referrals, loadings, discounts. GenAI extracts these from code, batch jobs and stored procedures, and renders them in a readable rule catalogue that product teams can finally review.

Generating test cases for code nobody dares touch

The reason most PAS modernisations stall is fear of regression. GenAI generates unit, integration and scenario tests from the legacy code itself, including edge cases the original developers never documented. That test harness becomes the safety net that allows refactoring to proceed.

Code translation, not just code generation

GenAI translates COBOL or PL/I into Java, C#, Python or Kotlin while preserving behaviour. It is not perfect — humans still review every output — but it converts a multi-year line-by-line rewrite into a multi-month assisted migration.

Automated API and documentation generation

For every module that survives or gets rewritten, GenAI generates OpenAPI specs, sequence diagrams and developer documentation. The system finally becomes legible.

Intelligent claims and policy automation

On the operational side, GenAI drafts policy documents, summarises submissions, classifies first-notice-of-loss inputs, and triages claims — feeding the modernised PAS with cleaner data and reducing manual handoffs.

Predictive customer and product insight

Once data is liberated from the legacy core, GenAI surfaces churn risk, cross-sell opportunities and product-mix gaps that the old reporting layer could never produce.

A realistic modernisation pattern that works

We do not recommend ripping out a working PAS. The most successful programmes follow a phased pattern that lets the business keep selling while engineering rebuilds the foundation. Redian's policy administration system practice typically structures these programmes in four overlapping tracks.

Track 1 — Discovery and rule recovery. GenAI ingests the legacy codebase and produces a capability map, a rule catalogue, and a dependency graph. Business analysts validate the output. This alone often delivers more documentation than the carrier has accumulated in twenty years.

Track 2 — Strangler-pattern carve-outs. New microservices are built alongside the legacy core for the highest-value capabilities first — typically quotes and new business issuance, followed by endorsements and renewals. Traffic is routed gradually. The legacy PAS keeps running for everything not yet migrated.

Track 3 — Data and integration modernisation. A modern integration layer replaces the batch-and-flat-file plumbing, exposing clean APIs to distribution channels, broker platforms and partner ecosystems. This is where digital channels and aggregator integrations finally become straightforward.

Track 4 — Decommission and run. Once every capability has been migrated and proven, the legacy PAS is retired. The new platform runs with AI-assisted observability, automated regression suites, and continuous documentation.

This is not a paper exercise. We have run variants of it for insurers and banks across our BFSI engagements, and the pattern works because GenAI compresses the slowest parts — discovery, rule extraction, test creation — by an order of magnitude.

Where GenAI fits across the insurance value chain

PAS modernisation is the foundation, but the same GenAI capabilities unlock value across adjacent systems. A well-modernised PAS pairs naturally with an ML pricing and rating engine that produces risk-adjusted premiums in real time, a modern claims management platform that triages and settles low-complexity claims automatically, and an insurance broker management system that gives intermediaries the digital experience they now demand from every carrier they place business with.

Reinsurance is another area where GenAI is changing the math. Treaty placement, facultative submissions and bordereaux reconciliation are document-heavy, exception-heavy processes that LLMs handle well. A modernised PAS feeding a reinsurance placement system closes one of the longest-standing data quality problems in the industry.

The hard parts insurers underestimate

GenAI is not a magic wand, and PAS modernisation programmes still fail when carriers underestimate four issues.

Data quality is the silent killer. GenAI amplifies whatever data it is given. If policy history, party records and claims data are inconsistent, the new platform inherits the mess at higher speed. Data cleansing has to start in week one, not month twelve.

Model governance is non-negotiable. Insurance regulators are sharpening their stance on AI use in underwriting, pricing and claims. Every GenAI-assisted decision needs lineage, explainability and human-in-the-loop checkpoints. Our AI/ML consulting practice builds these guardrails into the programme from day one.

Change management eats technology for breakfast. Underwriters, claims handlers and product managers need new tooling, new workflows and new skills. A modernised PAS that the front office refuses to adopt is worth nothing.

Vendor lock-in can return through the back door. Choosing the wrong LLM provider, the wrong cloud, or the wrong proprietary low-code layer simply replaces COBOL lock-in with a newer flavour. Architecture choices need to keep optionality open.

What "good" looks like in 18 months

A carrier that runs this programme properly should expect, within 18 months: 60-80% of the legacy rule base catalogued and validated; quotes, new business and endorsements running on modern services with API access; a regression test suite that did not exist before; underwriting and claims teams using GenAI copilots embedded in their daily workflow; and a measurable drop in time-to-market for new products — from quarters to weeks. The savings on mainframe MIPS, licence fees and exception handling typically fund the next phase.

These are not aspirational numbers. They are the outcomes Redian engineers deliver on real BFSI programmes — documented in our case studies, including core banking modernisation in Cameroon and insurance aggregation in Kenya.

Build with Redian

Redian Software has been embedded in BFSI modernisation since 2016, with delivery hubs in Noida, Nairobi, Dubai, London and New York, and a CMMI Level 3 appraised engineering practice. We bring the rare combination of deep insurance domain knowledge and applied GenAI engineering — the two disciplines that PAS modernisation actually demands. If your policy administration system is holding back the rest of your business, start a conversation with our insurance team and we will scope a phased programme that fits your risk appetite and your budget.