Alastair Black looks at how the increasing use of analytics is affecting life insurance pricing
During the past few years, life insurance pricing has become far more sophisticated. This is most noticeable in the UK protection market, which is performing more granular analysis of the risk cost of different market segments and using ‘demand models’ to consider customer behaviour – both of which lead to better understanding of profitability. Improved technology is allowing many providers to re-price far more quickly, and thus more frequently.
The traditional pricing approach typically relied on high-level monitoring of competitiveness and sales volumes, plus a large amount of commercial judgment and probably some cumbersome spreadsheets, to evaluate any potential price change. As a result, selecting the size and direction of a price change was more an art than a science. The impact of price changes could be hard to predict in advance, and hard to monitor subsequently. The time taken to evaluate, decide and sign off on a price change could be lengthy. Once approved, implementing a price change was often another lengthy IT process of (at least) several weeks.
Data and analytics usage and better technology have combined to help insurers improve every aspect of pricing, and many now take a data-driven approach.
New market information and sales data are rapidly analysed, and bespoke, relevant management information (MI) produced quickly. This means that proposed price changes can be analysed rapidly, and the impact of these price changes evaluated faster and more accurately. Many insurers have also implemented systems improvements to allow them to change prices in less than a day.
This was happening before COVID-19 arrived, but the pandemic has accelerated existing digitalisation trends. There is increasing customer-centricity and a focus on faster, more direct customer journeys, as well as growing influence from the direct-to-consumer market and from aggregators that are enabled by better use of analytics and technology.
Figure 1 shows the drivers for increased analytics use, from the perspective of insurers surveyed recently by Willis Towers Watson (WTW). While in-force management and competitive pressures were the first and third biggest drivers, the desire to improve the customer experience was a close second. There is a desire to use increased sophistication to benefit both the insurer and the customer.
Understanding customer behaviour
Many improvements have been driven by better data management and collection, more automated MI production, and the use of associated technology to enact changes more quickly. In other words, the improvements have come without the use of more analytics. However, when it comes to customer behaviour, analytics has come to the fore. Better understanding of customer lapse behaviour has already driven significant insight. In recent years, the key commercial use has been in understanding customer purchasing behaviour: customer demand and elasticity.
Insurers are now modelling customer demand and elasticity to find the key segments in which the insurer needs to be competitive to win business (or, equivalently, answering the question ‘how far off first place can we be and still win business due to the strength of our brand/service offering/product features?’). They are also modelling these factors so they can predict the impact of price changes more scientifically. These changes have allowed insurers to improve pricing margins in certain segments and increase volumes by better targeting price reductions.
The key data items required for understanding demand are information on quotes, prices quoted, and whether these quotes converted to a sale. These can then be enhanced with other data sources, such as competitiveness data (for example how the insurer’s quote compared with peers’ quotes) and any other data on customers, potential customers and distribution networks. Including data from a randomised price trial allows demand models to also take customer elasticity into account.
“Increased segmentation may mean some segments of the market are priced much higher”
Analytical approaches can then be used to understand what factors affect a customer’s propensity to purchase, and how this varies by competitive position. As there is potential for correlated factors in the data (for example sum assured and distribution channel), generalised linear models (GLMs) are typically the best technique to start with. GLMs allow for these relationships in the data while remaining relatively simple to use and understand. Many life insurers already use GLMs in other areas (such as mortality modelling), so these techniques may already be familiar.
More advanced analytic techniques, such as machine learning, are also much touted. In the property and casualty (P&C) market, sophisticated pricing techniques have been standard practice for many years – particularly in car insurance, where there is a lot of data. As a result, many P&C insurers have moved from GLMs to other techniques in search of improved predictiveness and competitive advantage. By far the most common next step has been gradient boosting machines (GBMs).
GBMs essentially fit a large number of models with weak predictions, each successive model being fitted to the residual of the previous step. Cumulatively, these models provide an overall improved model – typically with better predictive power than ‘just’ a GLM. However, like many advanced analytic techniques, GBMs are generally hard to explain and understand. One approach is to use a GBM to gain additional insight into the significance of factors, but then to re-fit using a GLM for ease of understanding and deployment. This highlights the need to weigh predictive power against ease of use.
One issue raised by the increased use of analytics, particularly more segmented pricing, is that of fairness. Increased segmentation may mean some segments of the market are priced much higher (just as others would be priced lower). Pricing insurance according to individual risk may seem fair to actuaries, but can go against market perceptions that equate fairness with pooling of risk. Ultimately, this may cause problems around uninsurable risks and access to the market. Senior management must ensure they have a clear position on these sorts of issues.
Increased use of customer behaviour can also lead to increased differentials in prices offered to consumers based on their propensity to purchase, rather than their underlying risk. Similarly, the use of randomised price trials noted above leads to price differentials that are random by design. In our experience of implementations of these techniques in the UK life market, the outcome hasn’t been wider differentials – just better targeting of existing commercial differentials. Nevertheless, this remains a potential conduct risk if not carefully managed.
The Financial Conduct Authority’s (FCA) September 2020 General insurance pricing practices: Final Report (bit.ly/3ulXp5P) is relevant here. While it is focused on general insurance – particularly differentials between renewal quotes and new business quotes, which are generally not a concern for retail life insurance products – there are still implications for life insurers.
The FCA’s overall view of fair pricing would also be applicable to life products (as demonstrated by the FCA’s recently released consultation paper on consumer duty, which can be found at bit.ly/3uvvgcq), and the governance requirements in the report are explicitly extended to ‘pure protection’ products.
“Better understanding of customer lapse behaviour has already driven significant insight”
Crucially, the FCA does consider the approach of price optimisation (effectively pricing based on not just underlying risk, but also customer demand) acceptable, so long as it is used appropriately. The FCA requires firms to consider fair value in their product governance processes, and makes it clear that a firm’s board has ultimate responsibility for this. ‘Fairness’ is not easy to define, and a number of sometimes contradictory definitions can be used. Insurers should choose the key fairness criteria that they want to be measured on, set these out transparently with board approval, and then rigorously measure their products and pricing against these criteria.
These pricing trends, and increased data and analytics use, will continue. In the protection market, the market’s sophistication (and frequency of re-prices) will grow. The pricing techniques are likely to become more widely used for products other than protection, with annuities being the next product type to which these techniques are most applicable, although with even more sensitive fairness considerations. In the WTW survey, insurers also identified savings products as a key area for increased analytics use.
These techniques are also likely to extend more widely across the customer journey. We are already seeing increased use of analytics in underwriting (for example to increase straight-through processing), and to improve claims triage and payment times.
The end vision would be for data to be collated, and analytics embedded (ethically), across the customer lifecycle to improve the focus on and understanding of customer needs and behaviour. Alongside increased technology use, this should allow better, more responsive and tailored products for customers, as well as more targeted marketing and cross-selling, faster underwriting and claims, and improved access to insurance.
Alastair Black is a director in Willis Towers Watson’s Life Insurance practice, where he leads the protection offering, with a particular focus on analytics