Speaking volumes

According to the latest figures from the Association of British Insurers, what used to be referred to simply as theft but is now more accurately referred to as fraud, costs insurers £1.9bn a year.

In an industry where there is often little unanimity, the rising level of fraud is one thing everyone can agree upon. And fraudulent claims are not only burning a hole in the pockets of the insurers - they bump up every individual's premium by an uncomfortable £44.

Experts may claim the recession has had an impact on fraud levels but the fraud supervisor at Provident, Andy Haslam, believes the psychology behind this kind of deception runs deeper than a mere reaction to impoverished circumstances.

Over the past five years, Provident has developed a unique and effective approach to tackling fraud. A crucial part of the insurer's armour is Digilog, a fraud detection solution it first piloted in 2005. In-house, the team refer to its line of attack as pre- and post-Digilog, which hints at the almost salvational respect it has earned. If anyone thinks this is indulging in hyperbole, they should check out the results. Before Provident implemented the software, roughly 10% of claims were refused per year. However, in the first year of Digilog the refusal rate shot to 25% and it has risen year on year to hit 30% in 2009.

When asked how the system works, Mr Haslam explains: "Digilog looks for the truth in a conversation. When you are telling the truth, your version will be lucid with a fair amount of detail. If you were to lie, then when you spoke, your brain would be recalling a different experience. Therefore every question you are asked after that involves your brain reprogramming itself."

Combined approach

Digilog operates on two unique functions: voice stress analysis (VSA) and cognitive interview techniques. The software itself is responsible for analysing how stress levels in someone's voice can indicate if they are potentially lying. However the quality of the cognitive questions relies on the capability of whoever is asking them. Mr Haslam says: "It's not just the Digilog process; it is the way that our staff operate the software that makes it the success that it is."

Rather than outsourcing Digilog, working in-house has allowed the department to slowly evolve and, as it became more experienced, develop and personalise the software to include new questions. These focus on specific areas of risk identified and consequently the package has adapted to suit Provident's individual needs.

The claims department is split into a primary and secondary team. With no filtering, the primary team deals with every claim where a motor vehicle has been stolen. Firstly the claimant's voice is calibrated by asking a couple of questions that require a straightforward answer. The remainder of the call consists of 25 questions designed to talk the policyholder through the theft. During this time, the VSA monitors what aspects of the story are deemed to be 'high risk' and at the same time the claim handler logs specific behaviours that indicate possible deception. These include elaboration, thinking time, problems with detail, abnormal emphasis, lapses in chronology, unusual responses, tense confusion and muddled language indicating a lack of ownership.

If at the end of the conversation the handler believes the caller to be 'high risk', they relay a carefully worded speech requesting the customer call back within the next three days. The handler continues by subtly implying there is an issue with the claim and if the claimant chooses to simply not call back, the claim will not be followed up. It is an easy out for an opportunist fraudster who has thought better of chancing their arm.

If the high risk caller does ring back, the secondary team fields the call, employing a less formulaic approach, probing the problem areas to see what kind of reaction is evoked. At the end of this process, if it is believed the risk posed by the claimant directly affects the policy, the case moves to the investigation phase. Only 5% of suspected theft claims currently end up making it this far.

Mr Haslam is keen to point out that Digilog alone cannot target fraud: "We never use Digilog as evidence and we would never repudiate a claim solely based on the fact that it was high risk on Digilog. We use it as an indicator for claims that should be looked at more closely and then work hard to gather the necessary evidence. I have no desire to use Digilog as the sole identifier of fraud and I think it would actually be quite a dangerous precedent to set."

Therefore, the system is used alongside a host of tools to gather sufficient evidence to challenge the claim. Police reports, sending keys for forensic analysis, accessing the claimant's data history, information about the vehicle's previous owner, witness testimony, Google maps and industry databases such as Hunter II are all methods that help the team to collate enough evidence to challenge a potentially fraudulent claim.

Growth potential

Provident is quick to acknowledge that Digilog has limits and it is currently only used in technical cases where there has been no personal injury. Rather than seeing this as a hindrance, Mr Haslam believes it is a challenge for development. Growth is something he feels his team is more than ready for, saying: "We've got this great tool that's working very well for motor accident claims but what else can we offer? There is a massive collection of insurer data so can we interrogate it to look at things like credit-hire fraud, household, employers' liability?"

What really excites Mr Haslam about the future is finding a way to tap into the £1.9bn currently lost to the netherworld of undetected fraud. His latest project aims to turn the tables on how insurers approach the problem. He argues that it is all well and good identifying fraud when someone makes a claim but questions why can't the software be used to highlight fraudsters at the underwriting stage before anyone has signed on the dotted line?

Provident may not be the biggest insurer out there, and it doesn't try to be. Innovation is not dependent on size and this new perception of risk could inform how the industry combats fraud. After all, prevention, or pre-emption perhaps, is always better than cure. n