Predictive analytics (also referred to as ‘predictive modelling’): is it the future?

It is if you listen to firms that are taking a keen interest in, or even specialize in, ‘big data’. Firms like BioSignia, Deloitte, Towers Watson and Milliman are intent on getting the ear of carriers, and even reinsurers like RGA and Swiss Re are getting in on the act. According to Deloitte, there are 40 data vendors out there and around 100 databases to draw upon. Also, they have a database that includes every adult in the US, all derived from publicly available information. Sounds like powerful stuff. Scary even.

Of course data mining, analysis and modelling is nothing new in auto and home insurance, in which carriers long ago recognized the value of matching claims statistics with demographics and risk characteristics to derive meaningful risk factors, sometimes with surprising results. And any self-respecting marketer will tell you that data analysis will tell you a lot about consumer behaviour, including propensity to buy (and buy again), value of purchases and, in the case of insurance, propensity to lapse. Life companies everywhere are rather behind this particular curve, and only recently have they been taking an interest in using data (including, increasingly, that from underwriting engines) to understand better the quality of business they are writing and the quality of their producers.

Mortality prediction

But predicting mortality? If you process enough cases via an engine (or otherwise code underwritten cases) and examine the data, you can get a better understanding of why mortality is turning out the way it is and the contribution of various risk factors. Depending on the volume and the detail you may be able to apply some of the new knowledge to individual case pricing. If you don’t have your own stats the data specialists will provide you with some and work with you to apply it to your business. Even if you have your own data it can be supplemented with external sources for a bigger and more detailed picture.

BioSignia, for example, has two predictive analytics models for underwriting, one a ‘full underwriting solution’ and the other a more limited model that does without assays from blood. They say these have been tested against traditional underwriting models and that they work. In fact they say they work better than the traditional models, which over-price some risk and under-price others. And this is one of the claimed benefits of predictive analytics, that they are more accurate – and more accurate in a significant proportion of cases.

In a way, underwriting accuracy is not a big issue provided the overall experience is satisfactory… but then there are the twin issues of cross-subsidy between policyholders and, inevitably, competition. Hit-and-miss risk pricing is not that big an issue if everyone assesses and prices risk in the same way, but if your competitors are doing it better then you could be in danger of losing the good risks and winning the not-so-good ones.

Another claimed benefit is the ability to underwrite on less evidence, maybe to do without lab tests or even a medical. As we have observed before, people are only buying insurance; shouldn’t carriers be trying to make it as easy as possible for them? Of course preferred criteria are based on a variety of factors, including physical characteristics and blood and urine profiles that dictate more than an app. But will ‘preferred’ always be defined in this way?

Whither preferred?

Think about it. Preferred at the moment, even if the assessment process involves a degree of scoring as opposed to an ‘in/out of whatever category approach’, is a bit simplistic, a bit clunky. Suppose the risk pricing could be based on many more factors, and driven by a sophisticated algorithm, for a more holistic risk appraisal? And actually, with good data readily available you can extend the preferred concept away from the current, core, traditionally underwritten model. You can apply the preferred concept to almost any product proposition and among a variety of markets using a variety of channels.

So the term ‘preferred’ becomes redundant, and ‘matrix pricing’, recognizing multiple risk factors, the name of the game.

Someone commented about predictive analytics that: “The tough part isn’t building the model, it’s implementing it, using it effectively.” And they’re right. No point in having something unless it adds real value to your business, unless you get a decent return on investment. You also need to maximize that return, avoiding under-utilization or under-leveraging.

But there’s a more mundane issue. The model might give you answers but generally you have to be able to justify them to customers and maybe producers too. The answers may be right in the overall context of the model’s output, but they may have to make sense to mere humans, they have to be saleable. (That doesn’t always follow, especially in direct-to-consumer environments and especially where portals or aggregator sites are involved, but the issue needs to be borne in mind.)

What is underwriting?

This is starting to beg the question “What is underwriting, what is individual risk pricing?” Because it is becoming no longer ‘quick and dirty’ (for convenience) versus ‘longer and careful’ (for bespoke pricing). ‘Quick and complex’ is emerging as an option. But despite the potential power of the model, and depending on the composition of the predictive data, it won’t get the answer right all of the time (just like human underwriters and the traditional process in fact). Does it matter? It depends. Insurers have a choice of substituting predictive analytics for some of the traditional risk information, or using these tools to enhance the traditional approach.

So where is all this leading the world of life and disability risk management? There’s a need to face up to the fact that we are living in an era in which technology is big. It is empowering consumers via information and by expanding the range of offers available – and doing all this quickly and conveniently. Medicine is applying it to develop new investigations and new therapies, and also to interact with patients in new ways – such as direct transmission of blood pressure and similar values and auto-updating of electronic health records. Technology is empowering businesses too – although some might say not always in a good way.

And don’t forget that insurers in the US and Canada have long been drawing upon resources such as the MIB and MVR records, and more recently on Rx databases. Think of these as predictive analytics ‘lite’. Add to those the scoring models now being offered by the major labs… You can see where underwriting is going. And already some disability insurers are using predictive analytics to improve claims management, particularly the identification of claims likely to be difficult or that otherwise warrant particular attention. Think too about ‘telematics’ systems which enable auto insurers to log how, how much and where a vehicle is driven for premium calculation purposes. (Imagine ‘health telematics’ being sent direct to your life or health insurer…)

So face up to it. Things are going to change and change a lot. New tools and techniques should be gladly embraced – provided they can be applied responsibly for consumer benefit (and not for convenient profiteering). Just don’t go sleepwalking into the new world – you might find it an uncomfortable place.

Postscript

If you think all this predictive analytics stuff is all bunkum, a quicker, cheaper approach might be to enter an applicant’s details into a ‘death clock’ calculator like this one http://www.death-clock.org/. I didn’t take this too seriously… but the result I got did make me think I should start improving my lifestyle…