Published by Michael Crawford on
Other, more pessimistic, onlookers may think this is a vague area of developments which will require huge levels of investment.
Fortunately, in reality, the phrase is less unnerving and introduces an easily obtainable opportunity for insurers. In this blog we introduce the topic which is at the forefront of Barnett Waddingham’s thinking and provide an example of the benefit that data analytics and machine learning can bring to the insurance industry.
One of the cleverest and most useful attributes of the new model was the ability to layer a number of different data sources into the analysis.
Machine learning is the ability to use computers to mine data and extract information that would otherwise be out of reach of human calculations. Up until the late twentieth century, statistical theory far surpassed our ability to apply the calculations to reality. However, as computer power has developed exponentially in the last 20 years, we can now develop systems that would have been unfathomable a decade ago.
Tools can be used to investigate policyholder behaviour, pricing models, underwriting or distribution channels to name but a few. The key to developing these capabilities is to have both the coding and computational theory alongside the understanding of the business. This is what makes Barnett Waddingham’s new Data Analytics team such an exciting addition to the partnership.
We are already seeing clear examples of using machine learning to the benefit of insurance companies and we believe this is just the tip of the iceberg. Below we highlight one of the ways data analytics has provided value to insurers.
Life insurance companies can experience a significant level of lapses or surrenders on their book of contracts. Ideally they would like to understand the cause of the poor experience and to find ways to improve it. This is an area where machine learning can provide insight.
Problems such as this are, typically, grouped under the term 'Time to Event Models' and they have been studied extensively in medical trials and operational research. Time to lapse or surrender, can be modelled in much the same way as time to failure of a machine or recovery/death in a medical study.
Time to event modelling allows us to separate out and quantify the various effects on lapse or surrender that are inherent in a book of policies and one of the key benefits of this approach is the ability to quickly improve the model by testing new hypothesis as we see fit.
The benefits don’t end there however. It is easy to extend a model such as this using public data such as credit scores or socio-economic data. Performing such an exercise can add significant layers of insight not initially available to the insurance company.
Modern visualisation techniques also allow the introspection of the data in many different dimensions and this can also lead to a deep understanding of the problem of lapsing.
Once the factors underlying lapsing have been understood and quantified the insights can be used to put in place management actions aimed at improving the quality of the book of contracts. Furthermore, the insights can also be used to when making future pricing, reserving, underwriting and distribution decisions.