BBD leveraged AI-assisted code conversion to transform a legacy mainframe address management system into a modern Spring Boot microservice – reducing development time by more than 90% while improving maintainability, testability, and scalability.
The challenge
Sanlam’s CDM team had already migrated an existing mainframe system to Java Spring Boot – a process that spanned over three years, with most of the work done manually. Our challenge was to take a small portion of that already-migrated system and test whether AI could accelerate the process. Specifically, we set out to use agentic AI to translate COBOL into business documentation, and then convert that documentation into a Spring Boot API with minimal developer intervention.
Our AI-led solution
Operating within Sanlam’s environment meant we were limited to using GitHub Copilot. To push the boundaries, we also leveraged the Claude 3.7 model to:
- Translate COBOL and Assembly code into detailed technical and business documentation, cutting a process that normally takes weeks down to just minutes.
- Convert the documentation into a Spring Boot API using VS Code and Copilot’s new agent mode, ensuring alignment with modern industry standards.
- Successfully modernise a mainframe application responsible for CRUD operations and address validation into a Java Spring Boot microservice, resulting in improved maintainability, testability, and architectural design.
- Although AI drastically accelerated delivery, critical architectural decisions were still made upfront by BBD’s experienced developers. Careful planning, segmented prompting, and strong governance ensured the AI-generated code met the highest standards of security, scalability, and reliability.
How we approached it
BBD applied an agile, iterative methodology, embedding responsible AI practices throughout:
- Proof of concept – trialled AI-driven translation on one module to validate the approach.
- Incremental conversion – progressively migrated mainframe programs, with ongoing developer oversight.
- Two methods tested:
- Per component approach: breaking components into layers and prompting individually.
- Per layer approach: treating the service holistically, prompting each layer individually.
- Governance and quality checks – every stage aligned to business rules and best practices.
- Scaling the approach – once validated, AI prompts were structured into discrete, verifiable steps for repeatable success.
The impact
- Massive time savings – migration of core address functionality was reduced from roughly two months to just 3–4 days (over 90% faster).
- Immediate client value – Sanlam teams quickly adopted the COBOL-to-documentation method for broader use, ensuring future migrations would be faster and more efficient.
- Preserved institutional knowledge – by documenting legacy code, the risk of losing critical information as mainframe experts retire was significantly reduced.
- Reduced dependency on mainframe developers – enabling easier, more cost-effective migration going forward.
Why AI with BBD
BBD’s strength lies in combining deep technical expertise with practical AI enablement. We don’t just migrate code – we unlock business value by modernising systems with enhanced monitoring, metrics, and error handling.
Our approach to AI is pragmatic: automation is paired with human oversight to guarantee accuracy, security, scalability, and alignment with business goals. The result is faster, smarter, and more resilient modernisation projects.
