Making the Complex Simple: Smart Claims Fraud Detection
The insurance industry provides essential products and services that help consumers manage their financial risk every day – a role that is more important than ever in the current economic climate.
Fraudulent insurance claims are costing honest customers £1billion a year; but this is only the tip of the iceberg. The Association of British Insurers [ABI] estimates that there is a further £2 billion of undetected fraud. At present, circa 140,000 bogus claims are detected every year, with fraudulent household insurance claims the most common and fraudulent motor insurance claims the most expensive.
Reducing fraud losses is a high priority for all insurers, with the market recognising they must contain fraud as a ‘managed risk’. There is no room for underestimating the fraudsters targeting the industry, who continually evolve their methodologies in direct response to insurers’ counter fraud strategies. The opportunities afforded to fraudsters by new technologies are driving innovative techniques for perpetrating fraud over the internet. A dynamic counter fraud model is essential for insurers to protect their honest customers and bottom line.
So how can we simplify the complex and work smartly to detect claims fraud?
A cohesive, structured approach to defining, developing applying and maintaining a claims fraud strategy is pivotal to success; as is combining
innovative technology with enriched data, sophisticated analytics and a consistently high standard of claims handler expertise at an operational level.
Claims fraud indicators can be grouped into four main areas:
1. Expert Rules: drawn from the insurers’ claims expertise and based on the company’s internal data to identify typical fraud indicators eg. five or more people involved in an accident; a claim within 30 days of policy inception.
2. Geographical Rules: localising the individuals, the place and the entities involved in the claim and where possible integrating external geographical data to assess environmental risk.
3. Relational Rules & Link Analysis: risk assessment based on the relationships between the people involved in the claim and any links to previous suspicious claims history.
4. Analytics Rules: Data analysis using advanced statistics techniques and predictive models to identify representative behaviour of fraudulent claimants.
In order to maximize the detection capabilities of these indicators, insurers can enrich the data sources used in the model by complementing their internal data with external data sources. There are numerous data sources available to integrate with the solution, both open source and via industry bodies and commercial data providers. These can cover, but are not limited to, insurance customer history, socio-demographic, geographic, vehicle, and credit information.
The importance of identifying and managing the claims fraud risk at the earliest possible stage cannot be over emphasised. To ensure genuine claims are fast tracked and suspicious claims investigated, a three-layered approach to claims processing is highly effective. The first layer applies the automated model drawing on the four key groups of indicators to refer suspicious cases. The second layer of expert claim handlers establishes a triage phase at FNOL [first notification of loss] validating the referred cases
and allocating genuine claims for standard claims handling procedures and suspicious claims to a third layer: the special investigation unit. The special investigation unit can then manage referred cases very early in the claims life cycle and put prompt actions in place to stop fraudulent activities.
Detecting and analysing suspicious claims in this structured, holistic way assists insurers in understanding the fraud typologies affecting their portfolio and informs their counter fraud strategies and internal counter fraud culture. Assuming the cohesive model is based on a flexible and easy to maintain set of rules and procedures, insurers can create an environment supporting a ‘self-learning’ system allowing them to track fraud trends and adapt their model accordingly.
