Processing retirement requests requires precision. Personal data must be updated correctly. Regulations must be respected. Errors are not acceptable.
Our client, an insurance company in Germany faced these constraints daily. High volumes of retirement requests were handled manually. Data was read from submitted forms, entered into internal systems, validated, and updated in SAP S/4HANA.
The result: slow processing times, repetitive manual work, and avoidable mistakes.
Our experts proposed to use an LLM for form data extraction within a Camunda 8 workflow, modelled in BPMN, to automate the full retirement process while keeping human validation where it matters.
The Challenge:
Our client operates in the insurance sector. Retirement requests contain sensitive personal data and must comply with regulatory standards.
Several issues were limiting performance:
- Manual and error-prone data entry from submitted forms
- Complex validation logic and exception handling
- High compliance requirements
- Tight coupling between systems
- Mandatory integration with SAP S/4HANA
Each retirement request required careful checks. Every data change had to be traceable. Auditability was not optional.
The organization needed to reduce manual effort without compromising control.
Our Approach:
We designed an end-to-end retirement process using Camunda 8. The entire workflow was modelled in BPMN and executed through Zeebe workers.
The objective was clear: automate form data extraction using AI while embedding human validation only where required.
1. BPMN as the Single Source of Truth
We started with process analysis and BPMN modelling.
The executable BPMN model:
- Represents the full retirement journey
- Connects AI tasks, business rules, and human validation
- Provides transparency across business and IT
- Supports audit requirements
This BPMN-driven structure ensures that every step is visible and traceable.
2. LLM for Form Data Extraction
Retirement forms were previously read manually.
We introduced an LLM for form data extraction to:
- Read structured and unstructured form inputs
- Extract relevant personal data fields
- Prepare validated payloads for downstream systems
The AI component reduced manual data entry and improved data consistency.
Human validation was triggered conditionally. If confidence thresholds were not met or specific regulatory criteria applied, a review task was assigned via Camunda Tasklist.
This human-in-the-loop design keeps control where needed.
3. Event-Driven Orchestration with Camunda
The process runs on Camunda 8 with Zeebe as the workflow engine.
We implemented:
- Event-driven orchestration
- Dedicated Java/Spring Boot workers
- REST API communication
- SAP S/4HANA integration
The architecture separates process logic from technical services. Workers handle:
- AI interaction
- Validation logic
- Persistence
- SAP updates
This approach reduces system coupling and supports future evolution of services.
4. SAP S/4HANA Integration
Once validated, personal data updates are sent to SAP S/4HANA.
The integration is fully embedded in the Camunda process. SAP updates trigger the next retirement steps automatically.
All interactions are recorded within the orchestration layer, supporting audit and compliance requirements.
Final Result:
After implementation, the retirement process moved from manual coordination to orchestrated automation.
Key outcomes:
- Significant reduction in manual effort
- Improved data quality through AI-assisted extraction
- Conditional human validation instead of blanket reviews
- Full transparency through executable BPMN
- Event-driven integration with SAP S/4HANA
- Shorter processing time for retirement requests
The entire flow—from form submission to retirement initiation—is now managed inside a single Camunda-based process.
The organization maintains compliance while processing requests faster and with fewer errors.
Why This Matters for Organizations Handling Regulated Processes
Many regulated industries still rely on fragmented workflows. AI tools are often added on top of existing systems without orchestration.
Here, we show a different path.
An LLM for form data extraction becomes powerful when embedded inside a BPMN-driven process engine like Camunda. AI is not isolated. It is orchestrated. Human validation is not removed. It is targeted.
We design processes where:
- AI supports structured decision points
- BPMN provides visibility
- Camunda orchestrates execution
- Compliance remains measurable
With the experts from our Strategy, Product & Transformation service line and our Software Engineering service line, we advise, build, and run process automation solutions that connect AI, business logic, and enterprise systems in a controlled way.
If your organization handles high volumes of regulated requests, we can help you design a process that reduces manual workload while keeping governance intact.
Contact us to book a call with our experts and discuss how we could help you.
Technology works best when it is aligned with people and processes.