One vital rule for insurers to remember about predictive analytics
A software firm estimates that one in ten insurance claims are fraudulent and that its models can point insurers towards which ones they are. It’s called predictive analytics and is one of the ‘big data’ toolkits that is about to send all sorts of ripples through the insurance sector, including some that have ethical implications.
Predictive analytics takes your data and applies all sorts of statistical techniques to it, in order to identify patterns and relationships. As an output, the insurer would get a list of claims that are judged to be fraudulent in some way. The insurer is then able to focus the appropriate resources onto the further investigation of those claims. All of which is fine, so long as the insurer and its software supplier remember one absolutely vital rule.
Predictive analytics will only tell you the probability of something having a particular characteristic (such as being fraudulent). It will not tell you that something actually has that characteristic. In technical terms, it shows you correlations, not causations. So that list of claims will tell you the probabilities of each being fraudulent; it will not tell you which ones are fraudulent. It’s a distinction that needs to be pinned to every terminal in every claims department, because if it’s forgotten for even one claimant, that insurer risks jeopardising public support for its anti-fraud programme, however much it might protest about the overall impact that insurance fraud has on their business.
“…for even one claimant”: isn’t that a bit much, some of you may protest? Not if a fundamental tenet of English law is to be upheld. It’s called Blackstone’s Formulation, after an 18th century English judge (pictured) whose magnus opus was a set of highly influential books on the Common Law. In those works, he expressed his now famous formulation in this way:
“All presumptive evidence of felony should be admitted cautiously; for the law holds it better that ten guilty persons escape, than that one innocent party suffer.”
You may find it expressed differently (Benjamin Franklin referred to 100 guilty persons), but the message to err on the side of innocence is constant. Why should this be so? Let’s move only a few years on from the publication of Blackstone’s great works, to 1770 and the words of John Adams, then a Boston lawyer but later to become the second President of the United States of America.
“It is more important that innocence should be protected, than it is that guilt be punished; …when innocence itself is brought to the bar and condemned…, the subject will exclaim, ‘it is immaterial to me whether I behave well or ill, for virtue itself is no security.’ And if such a sentiment as this were to take hold in the mind of the subject, that would be the end of all security whatsoever”
In other words, the damage done by condemning one innocent person far outweighs the benefit from convicting ten guilty people, for if ‘virtue itself is no security’, then public support for the cause purportedly being upheld becomes fatally undermined.
For insurers, this would mean their pleas about insurance fraud being ignored and their handling of the conflicts of interest at the heart of claims negotiations seriously challenged (more on that here). As John Adams makes clear, that’s a very slippery slope for anyone to play near the top of.
So when you come across claims like ‘one in ten insurance claims are fraudulent’, remember William Blackstone and John Adams and why their take on those ‘ten’ and ‘one’ numbers has been at the heart of our law for several centuries.
In my next post, I’ll look at some serious ethical issues that could arise from how underwriters are talking about using predictive analytics.