Business Problem
AIA Malaysia faced rising medical costs, which it suspected was partly due to significant health claims leakage going undetected. The health insurer also wanted to improve customer experience by speeding up turnaround times, which were constrained by a slow and largely manual claims adjudication process.
Accordingly, AIA Malaysia wanted to implement an automated claims adjudication solution that could check across numerous policy rules and arrangements, and detect unusual claim patterns that suggest potential medical Fraud, Waste and Abuse (FWA) – to provide rapid and intelligent support on health claim decisions.
Technical Challenge
To automate claims adjudication, AIA Malaysia needed to capture all relevant
information from health claims they receive in a standardised, digital format, that can be analysed by a digital claims processing solution in a consistent way. However, constructing this standardised data pipeline was challenging due to the following reasons:
- Lack of digitised health claims data – majority of health claims are submitted in a non-digitised format, such as physical paper forms or images.
- Lack of consistency in medical invoice formats – no standardisation in how information is presented in medical claim invoices across various healthcare providers (i.e. in terms of fees, formats and levels of details) who use different billing systems.
- Lack of standard clinical terminology – use of varied service names in medical bills to describe the same medical cases and treatments, makes it difficult to analyse each medical service billed in its exact medical context.
Amplify Health’s Solution
We developed an integrated AI-driven solution that enabled AIA Malaysia to digitally capture and clinically enrich their health claims data, and automatically identify specific claim lines that breach policy agreements or suggest potential FWA:
- NLP-enhanced data capture – deployed an Optical Character Recognition (OCR) solution trained on Asian market data (increased accuracy compared to tools trained on predominantly USA or Europe data) and enhanced with proprietary Natural Language Processing (NLP) techniques to extract all relevant information from claims invoices and organise them into standardised health data.
- AI-enabled clinical enrichment – deployed multi-stage generative AI algorithms to rapidly map health claims text into a comprehensive range of clinical codes across 17 medical domains, with systematic human-in-the-loop quality assurance.
- Rapid custom policy guidelines tracking – deployed rools rule engine with Camunda workflow processing orchestration to automatically track compliance with corporate and individual policy arrangements – including complex clinical rules, negotiated hospital discounts, duplicate claims detection.
- ML-powered granular FWA detection – deployed machine learning models to granularly analyse each itemised claim and identify potential instances of FWA – such as overcharging, over servicing, procedure and diagnosis mismatches – not just by flagging suspicious claims but pinpointing specific line items that human assessors should review.
Business Impact
By transforming AIA Malaysia’s health claims data, Amplify Health helped raised the health insurer’s auto-adjudication and FWA detection levels, resulting in significant cost savings within the first year of deployment.
- Enriched claims data: Automated enrichment of 91% of all health claim line data with standardised clinical codes, enabling AIA Malaysia to assess medical claims in their medical context.
- Automated claims adjudication: Enabled 2x increase in number of claims available for straight-through processing, by digitising a variety of unstructured healthcare paper claims received by AIA Malaysia.
- Enhanced FWA detection: Located over billing, over servicing, diagnosis-claim mismatch in individual claim line items with up to 95% accuracy, through the use of multiple granular FWA detection models.
- Significant cost savings: Prevented medical leakage by identifying ~2.1% savings on total claims processed by the end of year one, which is projected to grow to 5.1% by year four.
