Looking at claims management in fraud detection, Moscow, Russia-based MAINS LAB CEO and Co-Founder Yury Kuvshinov explained that the different types of health fraud include over-treatment; cases that contradict the table or benefit; duplication or technical mistakes; or entirely fabricated claims, where the treatment didn’t occur at all.
Yury rationalised that fraud detection can be split into two categories: machine learning and rules and algorithms. He noted that while machine learning can help to identify complex correlations, it can sometimes be difficult to interpret and requires human expertise; and although rules and algorithms can trigger concrete fraud cases and are explainable and interpretable, they need to be constantly updated. He also stated that both these approaches are based on data and touched upon the problems and solutions associated with these forms of data.
“You really need to understand the structure of fraud in your country,” Yury reasoned, referring to the different levels and types of fraud. He stated that in Romania, Poland and Russia, for example, over-treatment accounts for eight per cent of the overall claims amount.
Yury’s presentation sparked a discussion on the importance of acknowledging company vulnerabilities. When it comes to fighting fraud, Yury noted that it is possible to try to reduce it and keep it as a downward trend, but that, ultimately, the key is to catch trends early and adapt to deter them.
Next on the podium was Steve Paton, Head of Anti-Fraud Services (Europe) at Verisk. His presentation covered the use of technology in the fight against fraud. He talked of the ‘race to zero’, with insurance industries in the US and UK making changes to see who can give the most efficient automated service. The essence of Steve’s empowering dialogue was that players in the insurance industry need to work as one to fight fraud.
He asserted that sharing intelligence internally throughout the different departments was a crucial first step, and that only after organisations had implemented this system could they use other technological capabilities to their best advantages when fighting fraud.
The key is to catch trends early and adapt to deter them
He also said that fraud trends are increasing and that new anti-fraud technologies are therefore now necessary. In the UK alone, there is a lot of activity around detecting insider fraud and Direct Line is receiving awards for simply sharing intel internally, Steve pointed out. He added that The Direct Line Human Resources Department routinely checks its intel system before it recruits people.
“Intelligence management is key. If you can share the intelligence, then that is the strongest asset that you will have,” Steve stated, adding that ‘power is collaboration’. As claims are ongoing, the fraud detection process needs to be constant and advanced anti-fraud strategies need continuous attention. Steve also touched upon AI and how this can be instated into fraud detection to help authenticate claims.
Delving into the implications of using AI to help detect and defer fraud, Estelle Lebar, Global Offering Leader, Travel Insurance & Assistance at France-based Shift Technology, began by explaining the company’s progression from providing fraud detection solutions to claim automation solutions.
She explained that issues facing those working within the fraud management process are the large volume of claims that needed to be processed, limited time when processing each claim and limited information available for each claim. For Shift Technology at least, solutions for these issues have been found, including cross-referencing of an insurer’s data, arranging claims according to their fraud risk and providing insight into the underlying reasons for suspicion.
Fraud detection can be split into two categories: machine learning and rules and algorithms
Estelle went on to detail that Shift Technology has developed a denoising algorithm, called Force, that addresses errors, duplicates, typos and inaccurate entries – which databases often incur – by reconstructing entities and cleaning data. “Of course, we are using AI, but this is a human decision tool,” said Estelle, noting that, in the end, a claim will not be marked as fraudulent without the final decision of the claim manager. The audience members concurred, many of them agreeing that although technology provides ample advancements in the fight against fraud, there will always be a percentage of fraudulent cases that cannot be identified by AI and, as such, the human touch is still necessary.
In a Q&A session, delegates asked who the fraudsters tend to be and whether Shift Technology had ever come across fraudulent assistance providers. Estelle responded by highlighting that, indeed, fraud could occur both internally and externally, and this was another way in which Force had helped detect instances of fraud. The discussion moved on to considering the many ways in which data could be corrupted. The session concluded with the importance of fraud detection procedures firmly established.