How HBF cut processing costs by 95% with AI claims automation
At a glance
Key services
- AWS Serverless Architecture
- AWS Bedrock
- Full-Stack Cloud-Native Application Development
- SSO Integration, CI/CD Automation
The challenge
HBF, a leading Australian not-for-profit health insurer and market leader in WA, had relied on a third-party provider for processing customer-submitted health insurance claims. This outsourced, largely manual process was costly and slow, with some claims taking up to two weeks to process.
HBF faced several key challenges:
- High cost, low automation: The third-party claims process carried significant per-claim costs with limited automation, placing a growing operational burden on the business.
- Unstructured and variable documents: Claims are submitted as mobile phone photos, scanned documents, and printed receipts across diverse providers and formats, including poor-quality images, varied orientations, and handwritten content, making reliable data extraction time-consuming.
- Third-party dependency: HBF needed a pathway to bring claims processing in-house and remove this strategic dependency.
- Member experience: Manual processing delays directly impacted the member experience. HBF required a solution that could reduce claim turnaround times without increasing the volume of claims requiring manual intervention by internal staff.
The solution
Mantel partnered with HBF to replace this process with an intelligent claims automation solution, combining Optical Character Recognition (OCR), Vision Language Models (VLMs), and agentic AI to extract, validate, and process health insurance claims at scale.
- Conducted an initial proof of concept (POC) to evaluate document accuracy levels, using AWS Textract for OCR and Amazon Bedrock VLMs for data interpretation and field extraction.
- Engineered a cloud-native, event-driven pipeline using Python and .NET microservices for pre-processing (image quality assessment and transformation), claim extraction, and post-extraction data validation.
- Re-routed manual ancillary claims through the IDP pipeline, enabling straight-through processing for non-pharmacy claims directly into the core production system, E5.
- Deployed a shadow process (dark launch) of the IDP solution, forking the full production claims workload into a parallel pipeline to process and reconcile every claim against existing production data. Following successful validation, the solution was promoted to live production for non-pharmacy claims, with the shadow pipeline retained as a continuous testing and regression environment for new claim types.
- Built agentic AI workflows, including a Missing Information Agent for pharmacy claims, to automate handling of complex claims with missing or incomplete data, enabling the solution to scale across additional claim types including pharmacy, multi-claim, medical, ambulance, and handwritten claims.
The outcome
Mantel worked closely with HBF across four phases, from proof of concept through to production deployment, delivering the following outcomes:
- 95% reduction in cost per claim:
The AI-powered claims automation solution reduced the per-claim processing cost, delivering significant operational savings. - Higher extraction accuracy than human reviewers:
The IDP solution achieved 94%+ field extraction accuracy across non-pharmacy printed claims, surpassing legacy service and manual reviewer benchmarks. - Production deployment of the ancillary claims pipeline:
Non-pharmacy claims are now processed end-to-end via straight-through processing (STP), with successful claims submitted directly into HBF’s core production system. - Significant reduction in third-party claim volume: By re-routing manual ancillary claims through the IDP pipeline, the volume of claims sent to the incumbent provider was reduced, directly lowering outsourcing costs and dependency.
- Zero disruption to existing operations: The solution was deployed with no material change to the volume of claims requiring manual intervention by internal HBF claims experts, maintaining operational stability throughout the transition.
- Production-ready foundation for full third-party removal: The deployed architecture provides a scalable foundation designed to remove the need for the provider in the claims value chain, with the roadmap now extending to pharmacy, multi-claim, medical, ambulance, and handwritten claim types.
- Agentic AI roadmap for complex claim automation: Developed agentic AI workflows, including a Missing Information Agent for pharmacy claims, capable of reasoning, decision-making, and retrieval of missing or incomplete data such as the pharmaceutical active ingredient name. This capability has been built and validated, and is positioned for production deployment to further increase straight-through processing rates.
“From days to seconds and from dollars to cents, as a not-for-profit health fund, our new claims automation solution has significantly improved our members' manual claiming experience while delivering substantial cost savings, meaning our members benefit from faster, more seamless service.”
Felicity PittawayGM, Member Operations | HBF
Third-party applications or solutions used
- E5 for claims tracking
- Control-M, used for moving claims between the production and staging AWS environments by writing to S3
- Python and .NET
How AWS was used as part of the solution
Fit-for-purpose AWS services were used to ensure scalability, high-accuracy data extraction, and minimal operational burden. The following AWS technologies were used in this solution:
- Bedrock
- AgentCore
- Textract
- Lambda
- Elastic Container Registry (ECR)
- Simple Queue Service (SQS)
- Simple Notification Service (SNS)
- DynamoDB
- S3
- CloudWatch.
