By Georgios Kapetanvasileiou, Analytics Consultant, SAS UK & Ireland
Recent figures from KPMG show a 78 per cent increase in the volume of fraud brought to UK courts last year. The value of insurance fraud in 2018 alone eclipsed the entire period from 2014 to 2017 combined.
Insurance fraud is not a victimless crime. In the UK, for example, many of the over 1,500 whiplash claims made a year are fraudulent. They cost insurers £2 billion a year and add £90 to the price of the average customer’s motor insurance premium.
The digitisation efforts of insurers are focused on improving certain processes; for example, the provision of policy decisions, the correcting of pricing policies, integration with aggregators and the reduction of customer friction during first notice of loss and claims. To achieve this, a significant adoption of new technologies is underway; encouraged by multiple factors such as the rise of Insurtechs and the wider availability of artificial intelligence (AI) and machine learning (ML) techniques.
Yet, while customer interactions and technologies are advancing and becoming more efficient, fraud detection techniques are falling behind. It’s crucial that our methods of detection keep pace with both customer expectations and ever-evolving fraud techniques.
A new breed of fraud
In some respects, insurers have gotten better at protecting against fraud. Closer collaboration and information sharing internally and with organisations such as the Insurance Fraud Bureau, have been instrumental in fighting certain kinds. However, the fraud landscape is like a seesaw; pressure on one side leads to the rise of the other.
Fraudsters regularly change their methods of attack and are becoming more advanced and coordinated. The appearance of new digital channels allows fresh forms of fraud to emerge. With a multitude of new digital channels but few regular checks, it has become easier than ever for a fraudster to falsify their identity, exploit systems and target victims online.
Past their best
Technological innovation shouldn’t have to coincide with a weakened security posture. Yet one of the major challenges holding insurers back are outdated detection methods and data infrastructures that are no longer fit for purpose.
Most insurance companies depend on specialised anti-fraud personnel and business rules-based software to protect themselves. While both are important components in any fraud strategy, they can no longer do it all alone. When an experienced staff member leaves the company, there often follows an immediate drop in effectiveness. They take their expertise with them, and the anti-fraud systems they helped build and design can become inscrutable to those left behind.
While it’s possible to replicate their knowledge as business rules, rules are easily broken. Ultimately, they represent a threshold that, once crossed, triggers an alert. So long as a fraudster can find strategies to avoid breaching that threshold, they can remain undetected indefinitely.
Fraud teams aren’t helped when the data environments they work with. Often disorganised, siloed, and unstructured data sources make it very difficult to assemble the data needed to create a complete customer view or identify a fraudster profile. Poor data quality and integration diminish accuracy and slow down the detection process, nevermind the fact that third-party data must also be compiled and integrated.
Prevention over protection
As in business, the most agile party usually prospers. So long as fraudsters can advance and evolve faster than their victims, insurers can’t win. While companies rely on outdated detection techniques and legacy infrastructures, they will remain vulnerable.
Fortunately, future technologies like AI and ML provide a way to even the score. New AI-driven techniques allow insurers to leverage massive quantities of data, and in real-time, not just to detect fraud but to prevent it from ever happening.
The challenge then becomes which option to choose. These should be driven by the unique needs and vulnerabilities of the business. Once an algorithm has been trained with data to recognise fraud, it can recognise patterns or features faster than any human. As an example, predictive modelling is adept at spotting the familiar warning signs before any fraud has actually taken place. These allow teams to step in before any damage is done.
On the other end of the scale, unsupervised learning is designed to hunt down fraud that has no precedent or historical data reference points. Using machine learning, the system decides for itself what is suspicious and warrants extra investigation. This capability alone makes it a gamechanger, and essential for companies that want protection from the unexpected.
Finally, social network analysis rounds out these capabilities. While other solutions are effective at smoking out opportunistic fraud, social network analysis unearths organised fraud. By bringing together multiple disconnected data sources, and establishing the links between them, the technology saves time and reveals connections no one would have expected. One large UK insurer used social network analysis to compare IP addresses, mobile phone numbers and email addresses to identify an organised fraud ring, saving £7 million a year as a result.
Next steps
If there was a tried-and-tested method for the application of insurance detection techniques then it wouldn’t be the problem it is.
Unfortunately, no one size fits all, and the fraud landscape is rarely static. It’s startling to see how AI and ML techniques are being used by fraudsters to maximise their profits. The best strategy is to keep all your bases covered, with a hybrid approach to detection, and infrastructure that’s flexible to change.
By using a hybrid approach and strong data governance policies, Allianz Insurance was able to investigate 26 per cent more cases and save CZK 110 million a year in fraudulent claims detected. What matters isn’t the capabilities you have, but how they work together.
|
