The process of life and disability underwriting varies from market to market. Those variations as they exist today have been driven by the prevalence of risk factors and the availability of reliable risk information, and not a little by history and culture (of consumers and of the industry itself) too. That differences in practice and process will persist is certain, as is that markets will develop at varying pace.
But one cannot help but wonder whether underwriting is on the cusp of a real revolution. That revolution will be driven by ‘predictive modelling’, by which we mean the use of data from internal and external sources to estimate or calculate risk. In the US predictive modelling – let us refer to it as PM – is a big talking point and various consultancy firms, sometimes in conjunction with reinsurers, have done some fascinating work using external databases – and there are many available there – finding combinations of factors that, together, indicate risk in a way that can be substituted for traditional underwriting using application forms and supplementary evidence like medical reports and laboratory tests.
Maybe not surprisingly the researchers and modellers are reticent about which of these alternative risk factors form the best combinations, for there is money to be made out of the combinations and algorithms that offer genuine potency. But potent they can be, with claims – based on comparative tests – that PM can price individual risks as well as, if not better than, traditional underwriting.
Just to be clear, that means that if an individual applies for life insurance (and for the time being PM is largely concerned with mortality risk), an insurer can price that individual risk as well as, if not more accurately than traditional underwriting (and all that entails from the app to lab tests and physician reports). Well more strictly, given that underwriters cannot predict the future of individuals (if they could they would have supernatural powers), it means that for a portfolio of lives you could happily leave risk selection to PM and achieve mortality experience just as good as, if not better than currently with classic underwriting.
So, will underwriters be redundant? Well, it looks like it… although actually that depends. And it depends on whether PM is a complete or partial substitute for traditional underwriting, or whether PM has no place in the proposition to the customer. Imagine at one end of a spectrum an unsolicited, price-personalised offer of insurance being made to an individual: no underwriting, apart from a few questions about recent (largely medical) history to obviate anti-selection. Somewhere in the middle of the spectrum a more traditional app, but with PM applied to supplement or avoid some of the traditional risk information. And at the other end underwriting as we know it, with highly personalised service, maybe associated with detailed advice as part of a financial planning programme. (However, don’t too get carried away by a conventional idea of ‘personalised’. It could be that databases can yield so much information about each individual that risk assessment and pricing is very personal indeed.)
As we hinted at outset, how or if PM will be applied will vary by market, as will its speed of acceptance and its potential: different markets have different availability of data and the ability to apply it to life and disability insurance. But the more one thinks about PM, the more one is forced to conclude that so much of conventional underwriting is inefficient, costly and, well, archaic. Surely, if there is an effective substitute, what is done now will simply be rendered redundant.
How well will insurers respond to PM? Will they regard it as an opportunity or a challenge, a threat even? Much depends on the nature of data available. But then the insurance industry is not renowned for innovation or rapid adoption of new technology or expertise. Many insurers, even in the US, are not especially tech-oriented: the underwriting process remains very traditional with big reliance on human underwriters, although automated underwriting is steadily gaining traction across the globe.
Smaller companies are most vulnerable, and their demise will lead to concentration of markets. But even larger firms dependent on face-to-face distribution are at risk. Agent and broker forces tend to be ageing and shrinking because new recruits are hard to find. And these firms will miss out on market growth via penetration of the traditionally under-served ‘middle market’. And maybe the high-tech players will be able to offer seriously lower premiums through low-cost processes and economies of scale.
Here’s another thought. PM raises the interesting question of the future significance of wearable device and e-medicine data that are supposed to augment – or even supplant – physician reports. If the predictive models are so good, will all this new e-tech be bypassed?
The power of PM also calls into question the future of underwriting engines as we know them now. It could be that engines are redundant because the model is all that is needed. Or maybe it will be a predictive model in tandem with an engine with a slimmed-down, very specialised rule-set.
Judging by all the PM activity in the US there is clearly a lot of potential there. What about elsewhere? What sources are available in your market? Who is investigating or will investigate them to develop the models? How will your company respond? Who will be your data partner(s)? What new distribution opportunities will you create? How are you using data now to understand which lives you insure have the better mortality and are less likely to lapse?
Will your firm be a leader or a close follower? Or will it sit on the fence, get left behind and struggle in the harsh new landscape?
Pundits reckon that of the occupations likely to be decimated by the march of technology, insurance underwriter is way up near the top of the list. Well, they could be right, although someone needs to be around to assess the lives that don’t fall neatly into the parameters of the models, the lives that have some real-world ‘red flags’ against them. Or, on the grounds of efficiency, will these risks just be rejected right away?