Big Data: What's the big deal?

Big data is one of those terms that seems to have been knocking around forever, without ever being important enough to Google.

This is a shame, as Google knows quite a lot about big data - as does your local supermarket, your motor insurer, those nice people who send you direct mail, Amazon, your bank and even the government.

It's not as if big data doesn't have uses for protection business. It could revolutionise underwriting, distributor management, medical evidence collection, reserving, pricing, marketing, sales and customer retention and change who answers the phone. Though, sadly, even big data cannot make a decent cup of tea without a teapot, so it's not a total panacea.

If it could fulfil even a fraction of this potential you would think the industry would grab it with open arms, but the fact it has yet to do so reflects natural conservatism in the writers of long-term business, and the distraction of regulatory, EU and other changes.

However, there has been progress with big data, and a number of forward-looking insurers and reinsurers have begun implementing big-data projects to improve their business and hone their skills.

There is real momentum building to incorporate big data as a core, multi-discipline tool to help drive businesses forward. Leading players are moving from test and learn to rollout - though the initial visible impact will be low, as deliveries will be incremental to existing processes.

Some players will start doing things that crucial bit better than their competitors, for no obvious reason. And then, at some point, someone will make the first move to revolutionise the market, and big data will move from best supporting actor to full-blown movie star.

The learning zone describes how to go big-data modelling, and two key points need highlighting. First, you do not need lots and lots of your own data to use big data. You can supplement your data with commercially available sources, multiplying the power of the specific data you hold.

Additionally, you can use standalone models built entirely from externally sourced data, such as the Mortascore mortality model from Mortality Metrics (see the case study for more details).

The second point to understand is that big data is just another, albeit very powerful, tool in the protection kitbag. It comes with its own issues and limitations, and it will never replace the need for people with the skill to understand how to put the bits together and the experience to judge the results.

The case study explores a much-cited use for big data in predictive underwriting for direct offers. However, the ability for big data to identify potential claim experience can also be used to complement the traditional underwriting process for intermediated business, enabling the number of underwriting questions to
be reduced and thus speeding up the sales process.

Behind the scenes, policies in existing books can be scored in many ways, from their likelihood of claim through to the chance of them lapsing in the future. Such scoring can provide a forward-looking view of future experience, years ahead of it emerging in future experience monitoring.

Linking clients to behavioural data enables models that can predict softer attributes: the chance of a client non-disclosing or their preferred method of communication.

The protection market is increasingly recognising that competing on product and price is not the way to close the protection gap. It may yet be that the biggest benefit of big data, the arch-mathematical tool, will be to help us to relate to clients as individuals and to really provide them with the protection they both want and need.

Nigel Bradshaw works through Redmayne Consulting on actuarial matters, through Mortality Metrics on the delivery of big-data solutions and through Make Sense Partners when developing and running new propositions

Case study: Predictive Underwriting

Mortascore is a big-data model from Mortality Metrics that delivers mortality scores at an individual, rather than postcode or aggregate, level, expressed as a percentage of population mortality. Scores are based on socio, economic, geographic and lifestyle factors. It is based on commercial data and can therefore be used by any organisation.

Results can be accessed from Mortality Metrics' data partner CallCredit, the consumer information business. CallCredit can meet all data-protection needs and provide immediate results or large batches of scores.

Mortascore has many uses in both protection and annuity insurance business. It can also be used by pension consultants and individual advisers wanting to advise their clients on their longevity risk in a post-compulsory annuity world.

One use of Mortascore is to predictively underwrite, using data not provided by the client to alter the price quoted for an insurance product and/or subsequent underwriting requirements.

So, for example, if you had a bank of potential customers to whom you wanted to make a protection offer then the traditional approach would be to contact them with a standard rate quotation and then put them through an underwriting process.

However, the alternative is to obtain their Mortascore scores in advance and then mail them appropriately priced quotes. Those with a score of 80% or less may get a 10% discount to the standard price; those with a score of 120% to 135% get a 25% loading and so on.

As ever, with new tools it is important to understand the breadth of their capabilities and the natural limitations in their application. It is unlikely there will ever be sufficient appropriate or timely data available to identify people just diagnosed with a major medical problem or who are awaiting hospital test results.

Such people would rightly be more motivated both to seek insurance and to seek a higher sum assured. This selects against other clients who would ultimately pay for the early claims through higher premiums. Therefore it is still necessary to protect against such claims using either one or two specific questions, a claims moratorium period or a pre-existing conditions clause.

The overall result is a protection offer that can be made to customers at an appropriate price for immediate acceptance. Easier to sell, easier to buy and easier to insure.

Learning zone: a beginner's guide to big-data modelling

There are many definitions of big data; a simple one is any set of data too big to be successfully analysed by a human without the aid of a computer. This can be quite small, but with data size does matter. Thousands, and ideally millions, of items of data do provide greater insight and more confidence in the results.

For those thinking they have not got such large amounts of data, or lack the technical or legal ability to use it, do not panic. Commercial data sets from providers such as CallCredit can be used to enhance even modest internal records by matching their clients to an ever-increasing range of data items. Even large data users enhance their data in this way.

It's then into a process: one well trod, but still best done by experts. The first step is to mistrust the data. It is so much less painful than mistrusting the results. Data errors need to be identified and corrected or cleansed. Indicators that have changes over time need to be stripped out - for example, smartphone usage indicated early adopters five years ago, but certainly not now.

The second step is statistical analysis. There are standard techniques and even free software to do this processing, meaning the process is low risk. Users can be assured the black box is not a black hole.

The result of these labours is a mathematical model that can produce scores on demand. It makes predictions about the future from past experience. It is a tool and, as with any tool, users should check they are happy with what it is doing and understand its limitations. Recent models predicting the World Cup winner mainly succeeded in confirming the need for healthy scepticism.

However, the general nature of such modelling enables it to be used for many purposes. The model outcomes could be expected mortality rates, the likelihood of lapse in five years' time, probable sales at different prices, or the chance that a client will pick up the phone if you call them. Nearly anything on which you hold data can be modelled, and the results can be used to help you make more informed business decisions.