Big Data Analytics - The Solution to Insurance Fraud?

The issue of insurance fraud has once again hit the headlines, with complaints having been levelled at the IFB for the slow pace at which it is running the Insurance Fraud Register. As the IFR only lists convicted fraudsters, there is concern that it will provide only limited assistance in combatting insurance fraud.

In fact, a poll taken at Hill Dickinson's recent Fraud Conference revealed that only 28% of the industry delegates in attendance considered the IFR an effective deterrent to fraudsters. Whatever the rights or wrongs of this particular argument, the cost of insurance fraud in the UK now stands at £21billion per year - an amount the industry can ill afford to sustain.

A solution can be found to this issue by modernising the way in which insurers detect fraud. The digitisation of business processes, combined with new ways of engaging with the customer - including online, social media and mobile channels, in addition to the more traditional channels of branch and telephone - are creating huge volumes of data which, if not analysed with sufficient rigor, make it more difficult than ever to detect fraud.

To address this, the priority should be to move away from statistical modelling - the traditional approach to fraud detection. Statistical modelling is not suited for our new, data rich world for three reasons:

1. The sampling methods used to analyse data means that some instances of fraud go undetected
2. The method relies on previously existing fraud cases, which means that insurers take a hit every time a new fraud occurs
3. The method works in siloes and is not capable of handling ever-growing sources of data in an integrated way

Insurers wishing to effectively reduce fraud should look instead to the world of big data analytics. Analytics allows insurers to build fraud detection systems that sift through all critical data, not just a sample, ideal for picking up low-incidence events.

The approach also enables insurers to take a global perspective on anti-fraud efforts, allowing them to link associated information within the organisation. So regardless of where the information comes from (mobile, social media, branch, etc.), or where in the insurance ecosystem it occurs (claims or surrender, premium, application, etc.), it is more likely to be detected.

For insurers looking at how to modernise their fraud detection systems there are three innovative methods that will prove important:

1. Social network analysis combines organisational business rules, statistical methods, pattern analysis and network linkage analysis to uncover large amounts of data on specific cases and show otherwise hidden relationships via links and nodes. Public records such as judgements, foreclosures, criminal records and bankruptcies can be integrated into a model that flags claim that are likely to be fraudulent
2. Predictive analytics for Big Data includes the use of text analytics and sentiment analysis to look at big data for fraud detection. Big data analytics helps by sifting through unstructured data - including claim adjustors' written reports - and can easily spot if a claimant's story changes over time
3. Social CRM - When social media is integrated within multiple layers of an organisation it enables greater transparency with customers. Social CRM can listen to the ‘chatter' on social media platforms and feed this information into the insurers' case management system, helping investigators to decide whether a claim is fraudulent or not

The message to insurers is therefore simple: the technology exists today to help create a more rigorous approach to fraud detection. While there will be an initial time investment in deploying such solutions, the impact on reducing the overall costs of fraud will be more than worth it.