Fraud detection: machine versus human?

March 9, 2022

Share on LinkedIn Share on Facebook Share on X

By Ian Carman, director, UK investigation services

Artificial intelligence (AI) and machine learning technologies are massively helpful in detecting suspicious behaviors and activities.

However, insurance fraud identification and claims validation remain one area where technology requires human intervention.

It’s not necessarily machine versus human; in fact, they go hand-in-hand. In sifting through countless claims, forensic imaging analysis and predictive analytics, you’ll find these are all things that claims handlers or fraud investigators would be unable to do with the same speed and level of accuracy. The technology successfully reduces false positives and produces reliable results while speeding up the fraud screening process.

Blended approach

There is no one solution to identifying suspect claims, which is why a blended approach is needed – using technology to process large volumes quickly, AI to identify suspicious patterns and behavioural science to help manage discussions with the customer. While AI doesn’t make decisions, it does point claims handlers and investigators in the right direction. By quickly recognising concerns, AI can help identify specific issues that highlight the claim as worth investigating. A fraud investigator will always be required to process the output. Not to mention, the data increases two-fold following a virtual or physical visit, which underlines the importance of one-to-one investigations.

The detection of policyholder deception is sensitive and requires strong conversation management skills — supported by digital voice risk analysis. It must be a carefully managed process — structured so that the fraudster knows when they’ve been exposed. More often than not, they will choose to walk away from the claim – a reaction that insurers can then consider on a case-by-case basis.

Wider landscape

New trends are also constantly emerging, with fraudsters continually finding alternative ways of committing fraud. Luckily, technology now enables us to expose fresh activity spikes at the earliest stage. Machine learning can even detect opportunist fraud – picking up on where customers are beginning to realise and take advantage of claims thresholds, for example. In liability and personal injury work, in addition to the classic fraud patterns, technology can also detect solicitors and doctors with the same connections, reference social media posts and put everything together to single out potentially fraudulent behaviour.

AI looks at the wider landscape, taking a big data view on fraudulent behaviours ­– from claims exaggeration to sophisticated organised criminal activity – compared to what’s happening in the market. The application of these new technologies makes counter-fraud teams more efficient, ensuring the investigators’ efforts and business expenditure is invested in the most robust cases.

Data insight

Identifying fraud is just part of the picture. Technology can also provide early insight into new fraud trends, giving a micro-level of detail – postcodes, types of claims and people – which empowers investigators with accurate information as it happens. This can help underwriters and sales teams react quickly and switch off areas or books of business that are becoming suspect.

Sharing information

Beyond the innovative technology lies quality data. While there is excellent collaboration and data sharing through motor-based systems, CUE and MIAFTR, the insurance industry could benefit from a wider pooling of data sets. Categorising fraud in a specific and uniform way, rather than using very broad terms, such as ‘motor’ or ‘personal injury’, would also be a step forward. The Association of British Insurers (ABI) has identified a set of agreed fraud types which should lead to greater standardisation of how various frauds are classified. Some insurers have also been sharing opponent-based strategies, certainly in commercial property, which is another positive sign. It’s this sort of industry-wide collaboration – a ‘treaty of the fraud teams’ – that will pay dividends in years to come.

The use of innovative technologies is bound to grow. Today, everything is about speed and performance. Honest customers expect their claims to be settled quickly and with minimum disruption. But fraud detection doesn’t have to conflict with a positive customer journey, and fast identification of questionable cases is crucial to stamping out emerging fraud trends.

This blog includes excerpts from a recent Sedgwick webinar hosted by Ian Carman, where a panel of industry experts debated the pairing of artificial and human intelligence in the fight against fraud.

Many thanks to the following panellists for their participation:

  • Dan Edwards, liability claims manager, Enterprise Rent-A-Car
  • Arnaud Grapinet, chief data scientist, Shift Technology
  • Kevin Kingdon, commercial property fraud manager, Aviva
  • Simon Roylance, claims crime prevention, LV
  • Stephen Dalton, head of intelligence and investigations, IFB