Neural networks could be the next step in tackling fraudulent claims, says Andy Cook

The world of insurance is getting used to the idea of advanced scientific concepts being applied to some of its problems.

Just two or three years ago the idea of cognitive interviewing or using voice stress analysis for countering fraud would have seemed ridiculous. Now it is those insurers who are not using such techniques that are in danger of seeming ridiculous, as recent results from trials show.

Well just as you were becoming comfortable with some of these terms, a new scientific idea is set to hit the industry - neural networks. It seems that at least one insurer has started using this concept to tackle claims and fraud in particular.

So how does it work? It seems that information on claims can be analyzed and the system teaches itself what the characteristics of a good claim are and consequently leaves a residue of iffy claims. It's a bit like using Amazon, where the site learns what kind of books and music you like and then starts to make recommendations to you.

The great benefit of using such a system is it is a low cost initial filter for claims. It is low-cost because apart from investing in the technology, there aren't many people involved. This means that an insurer can then focus on analysing the potential fraudulent claims. While it is early days, it is encouraging that our industry has the courage to try new ideas and this bodes well for shaking off the dull and dusty image.

And a quick apology to R&SA. Insurance Times moved its offices upstairs last week and due to a production error, R&SA's subsidence map was credited to Norwich Union in the headline.