Solutions for Life Insights banner

The conversation about automated pricing and underwriting in the life insurance space is influenced by the growth of InsurTech and the desire to make more use of data, both of which converge. Think about the idea of giving your customers wearables, which was big four or five years ago but which seems to have become so big that it bored the industry. Now, there seems to have been some resurgence of interest. John Hancock has required its customers to have some form of wearable if they want a policy, and that has put a few ideas back on the table.

However, there is a paradoxical element. Targeting healthy policyholders means that their mortality is going to be very low, so it will be five or ten years before you get a rich data set to allow detailed analyses. Companies that adopt this approach will not have done very accurate risk calculations, but will have instead estimated where their potential customers are in the social demographic spectrum, perhaps thinking about what sort of person has a wearable and what the known risk characteristics of that type of person are.

Differentiating policyholders accurately

Quantifying ‘FitBit mortality’ can be difficult. Research has been done into the impact of steps and their relationship between morbidity and mortality, but the amount of steps done is less predictive than the intensity of the exercise. A Canadian InsurTech firm called Viametric has access to a relatively small data set of over 15,000 people over 20 years and claims that some of the exercise-related figures are even more relevant to mortality than smoking.

If you are disinclined to use wearables to count steps or measure exercise because of the practical or logistical problems, how do you use data to improve your underwriting? If you’re a large banking group or another big organisation, you may find that you already have information about your potential policyholder in a different part of the group, which would make the underwriting process quicker and more predictive. If you do need to ask underwriting questions, there is growing research into the risk impact of some ‘easy to answer’ questions where the potential policyholder would find it hard to know if their answer would lead to a lower premium, which is ideal for obtaining more honest responses.

One firm is using selfies, but then they might be risking a behavioural change. Some people look young for their age while some look old for their age. That could alter the type of people who apply for such a policy. There were apparently 800 engineers working on the camera of the iPhone XS, with most of those deployed on machine learning, so it is not unreasonable to imagine a way that your phone could make you appear younger.

There is also a huge amount of bias in where data comes from, what sort of people are contributing to the data and even how it is prepared and cleaned. When you’re doing the analysis, you now have 20 or 30 different model types to feed that data into, so there’s a big risk of overfitting. If you’re trying 25 different model types and one of them is particularly predictive for that data set, does it mean that it’s the best model in reality? No. It just means that it’s the best model for that particular data set. And there is an issue of ethics and how you treat what are seen as protected characteristics.

From model choice to pricing

As all this feeds into pricing, it can be seen that there is much more to it than simply choosing and running the right model. One of the extra factors is how a model can take account of the value that customers attach to your brand. If you have a good brand, perhaps you don’t need to be the most competitive in the marketplace but can charge a slightly higher premium. From the risk perspective, this premium should be large and from a demand perspective, it should be low. Price optimisation is about trying to find the sweet spot in which you cover risk but sell enough policies.

But it’s not just about using pricing to promote better consumer behaviour or drive sales.  Firms also want to act on price with more flexibility and agility. If they go to their IT department and say that they want to add an exciting new factor to their pricing basis, whatever that factor is, the answer can often prove prohibitive. One company we spoke to was quoted a million pound cost and a year of delay by their IT department just to add one new factor, which is nonsensical.  Firms need pricing software that allows quick changes to their pricing basis, and can get new rates out to the market in days not months.

This is what the Radar suite of products offers – a way to allow for consumer behaviour, and a way to price with more agility. Radar also allows pricing to be more clearly conduct-compliant, by explicitly coding a range of considerations into the premium basis, for instance, what variation in premiums you will allow between policyholders differing in a particular characteristic, or between different distribution channels. With Radar, you can code all such considerations into your price algorithm and be sure that everything has been understood and discussed at senior management levels. When the FCA come calling, you can show the factors that influenced any particular price quotation, whereas a lot of other firms will have a less explicit audit trail.

Niche targets – and some conclusions

One of my favourite subjects has been targeting people who are not traditionally regarded as being an insurance company’s best friend, such as people who are obese, smokers or people with diabetes. Royal London have come up with policies that help diabetics to improve their behaviour, which seems like a win-win for the insurance company. You are no longer a monolith that people pay money to, but hopefully a firm that you engage with in a beneficial way. As a further example, one InsurTech firm in the US called Mira is aiming to go through all of the different life insurers and work out how you can optimise your cover if you’ve got diabetes.

We can see that lots of work is happening in InsurTech, and lots of that work is aimed at helping clients improve their data and, in turn, improve their profitability. We have the Radar suite of products to help our clients price better – meaning premium bases that are more granular, more flexible, more consumer-centric. This is an exciting and interesting time.