Skip to main content

AI that works

AI-Driven Legacy Migration for Real Garant

How Exxeta modernised 1.6 million lines of legacy code using AI in under 18 months.
#Modernisierung #Datenmigration
A car driving through a tunnel.

Unser Impact

  • AI-Driven Migration from RPG to Java

  • Hybrid Cloud Architecture on AWS with OpenShift

  • 30–60% More Efficient Than Purely Manual Redevelopment

The Challenge

Many companies know the situation: the core system keeps running, and has been for decades. But it is built on technology that barely anyone still knows how to work with. Real Garant, an automotive insurance company and subsidiary of the Zurich Group, faced exactly this challenge. Their central product system, GIVIT, has been running for 25 years and is used to manage extended warranties. But it runs on an IBM mainframe environment, programmed in RPG, a language that was once widely used in financial services but for which fewer and fewer developers are available today.

White line drawing on a black background: A person leaning over an open laptop, pressing down on the keyboard with both hands.

The Consequences Are Tangible:

  • Every code change is costly, as developers often have to be brought out of retirement to make it work.

  • System vendors build custom workarounds just to keep the system running at all.

  • New features? Barely possible due to the tightly coupled code.

  • AI integration? Out of the question. Building modern technology on top of an outdated backend is simply not viable.

White line drawing on a black background: A person stands to the left of a wall and throws a bomb marked with a code symbol at it. On the right, another person stands with hands on hips and a skeptical expression.

The Worst Case on Paper

The project has, in theory, everything that makes a migration difficult:

  • 1.6 million lines of RPG code on an AS/400 platform

  • No modularisation

  • No isolated tests for the behaviour of the existing code

  • No documentation. The source code is the only source of truth, and everything must be recovered through reverse engineering.

A standard migration under these conditions has historically been extremely costly and a long-term undertaking. Our goal is to complete it in under 18 months, using artificial intelligence.

The Solution

We took a different approach. Instead of expensive specialists, we brought in AI agents, built in part on modern code analysis models. The result: Java developers can produce clean, maintainable code without ever having to understand RPG themselves. The AI handles the analysis, the comprehension, and the untangling.

Our approach works in three steps:

Stylized browser window with the heading “Decompose”. Below it are three buttons labeled “Extract”, “Analyse”, and “Infer”.

Step 1:

Decompose – Untangling the Spaghetti Code

Before a single line of new code is written, the AI systematically maps the existing system:

  • Automated logic extraction: AI analysis identifies business logic, subprocedures, and database access patterns directly from the RPG code.

  • Knowledge graph: Implicit knowledge is translated into a structured, navigable form, making dependencies and entities visible.

  • Context enrichment: External sources such as Jira, Confluence, and Git complete the digital picture of the system.

Stylized browser window with the heading “Transform”. Below it are three buttons labeled “Design”, “Translate”, and “Validate”.

Step 2:

Transform – Rebuilding, Not Just Translating

No 1:1 transpiling. Instead, a genuine technological transformation:

  • Bottom-up: The AI starts at the deepest levels of the call chains to avoid redundant translations.

  • Semantic translation: Variable names and functions are not simply carried over, but translated into standardised, English-language business Java.

  • Automatic modularisation: Procedural RPG code is transformed into modern, object-oriented Java structures, with cleanly separated entities, repositories, and services.

tylized browser window with the heading “Synthesize”. Below it are three buttons labeled “Package”, “Test”, and “Deploy”.

Step 3:

Synthesize – Integration into the Target Architecture

The results flow into a future-ready design:

  • Hexagonal architecture: Business logic and infrastructure are strictly separated, following the ports-and-adapters principle.

  • Human in the Loop: Experienced developers validate the AI output at quality gates before it enters the CI/CD pipeline.

  • Hybrid cloud operations: The new application runs on AWS with OpenShift (ROSA), while the DB2 database initially remains on the AS/400, with the legacy system and new application running in parallel.

The backend migration is part of a broader transformation: by 2030, Real Garant's entire technology stack is set to be modernised, moving from a mainframe environment to the cloud, from monolithic structures to microservices, and from legacy systems to technologies for which developers are always available.

Stylized browser window showing a checklist. Four items are checked: “fast”, “efficient”, “scalable”, and “future-proof”.

The Impossible, Made Deliverable

The project demonstrates what AI makes possible today. Not because it writes code automatically, but because it overcomes the barrier of not understanding, enabling development teams to work on systems they previously could never have touched.

  • Under 18 months to go live, despite 1.6 million lines of unmodularised legacy code

  • 30–60% more efficient than purely manual redevelopment

  • Scalable: the modernised system supports 52 clients across 25 countries

  • Future ready: AI integration becomes possible for the first time, on a clean, maintainable backend

The principle is not limited to insurance. Wherever historically grown systems are built on RPG, COBOL, ColdFusion, or other legacy languages, the same challenge arises: the system runs, but it can no longer be changed. And at some point, it stops running altogether.

What the experts say

  • »Our starting point was clear: a system that was running, but could no longer be changed. Together with Exxeta, we found a path we would not have taken on our own. The AI driven approach changed not just the speed of the migration, but what is possible for us at all. We are not just modernising code. We are building the foundation for everything that comes next.«
  • »The real gain from AI: I only need strong Java developers. With AI support, they can understand enough RPG to produce something new, without having learned the language from scratch. We can take codebases that we technically don't master and make them manageable, then lift them into something modern.«

Get in Touch

Powerful Knowledge