The use of the latest technology to help search and detect financial fraud has been an area of focus for decades. For Frazer Walker’s partners, our first involvement was in the 1990’s developing systems to detect insurance fraud. In this article we look back at what was possible then and contrast it with what is possible today with the use of AI.
The approach we took in the 90’s was to take complete copies of data from mainframe IMS systems and transform it into dedicated DB2 relational database that we could interrogate.
What was leading edge at the time was in how we hand coded systems that used dynamic SQL to hunt for patterns in the data based on pre-defined knowledge based rules, “known alias rings” that red-flagged individuals and the weighting on scores to produce reports on what combination of patterns were found. This provided investigations teams with a prioritised set of potential fraud activities to investigate with the SQL based tools they needed to dig deeper.
Back then the main challenge was the limitations in processing power. We put a lot of effort into optimising the databases and indexing, but the flip side of our dynamic SQL coding was the system itself decided at execution time what paths to take. Targeting reports that took minutes one day, took many hours the next. Throwing additional hardware at the problem could only go so far.
What is possible today using AI and modern infrastructure
Today the use of cloud infrastructure and AI technology has changed the approach and what is possible radically. Storage and processing limitations have largely been overcome with improvements in hardware and database design, whilst AI and Machine Learning has largely replaced the human insight driven development of rules and patterns that predict fraud.
This shift from manual reviews and rule-based systems to AI enabled automated detection methods is being built on:
Real-Time Analysis at Scale
AI systems can now score millions of claims in real time using techniques like automated business rules, machine learning methods, text mining, anomaly detection, and network link analysis.
Pattern Recognition and Predictive Analytics
Machine learning algorithms learn from historical data and improve their ability to detect fraud over time, helping recognize new fraud schemes as they emerge. By learning from historical patterns, these systems continuously enhance their ability to identify intricate fraud schemes.
Multimodal Data Analysis
Modern AI systems combine data from multiple sources—text, images, audio, and video—to identify patterns and anomalies, reducing false positives while increasing detection rates. Technologies like optical character recognition (OCR) and natural language processing (NLP) enable AI to analyze documents such as claim forms and medical records, extracting information, comparing inconsistencies, and highlighting red flags
Improved Detection Rates
The impact has been substantial for both soft fraud (inflating legitimate claims) and hard fraud (premeditated false claims). AI is improving the detection rates while reducing the workload on human investigators.
Whilst the innovations in AI and infrastructure make this all possible, challenges remain for insurance companies to get their internal data AI-ready and to have the governance and AI-ethics frameworks in place to ensure that the outcomes of the automation can be relied on and managed effectively. At Frazer Walker we can help your organisation to harness the opportunities and avoid the pitfalls of using AI for financial fraud detection and investigation.