Life insurers are embracing the use of machine learning (ML) and artificial intelligence (AI) models and techniques in all areas of their business. Historically, the non-life sector has shown better integration of the use of data science techniques in their business. Innovative techniques are now more and more used in life insurance. From fraud detection to simplified underwriting, life insurers are showing a clear willingness to accelerate their digital transformation.

This blog post focuses on some of the uses of data science in life insurance as described in the Milliman paper “The use of artificial intelligence and data analytics in life insurance”. The paper describes the technical foundations and key principles of data analytics techniques, then outlines the use cases that have been developed by Milliman teams and by life insurers, reinsurers and insurtechs.

Underwriting

ML and AI have been used in the underwriting process for large insurers. Some insurers use these techniques to fully automate the underwriting process allowing decisions to accept applicants in minutes whilst other use partial AI models in the underwriting process to analyse large data files efficiently. More automated underwriting aims to improve customer satisfaction in interactions with their insurer and to enable faster decisions when accepting a policy.

Pricing and product design

Insurers have shifted their focus beyond just traditional financial orientated goals to the wider ecosystem mindset by increasing customer engagement and customer well-being initiatives. Insurers have incentivised the sharing of data from wearables by offering discounts and incentives for giving up smoking or tracking over 10K steps a day.

ML methods can help improve the speed and accuracy of risk assessments in underwriting and being better at segmenting risk profiles of customers and updating pricing assumptions on more data.

Sales and marketing

More sales are occurring online and robo advisors have emerged to automate the process of purchasing insurance, guiding customers through the process and recommending appropriate products. This type of automated advice generally works well with simpler products and can be used to market products to existing customers throughout their different life stages.

Interactions with customer service and selling agents can also be gathered to extract value that could improve modelling or understand customers better.

Claims management

Claims management is a significant cost for insurers. Life insurance claims often contain unstructured data which needs to be processed using techniques such as natural language processing speeding up claims processing.

In France, insurers have to demonstrate that they have sought to find unclaimed claims where the beneficiary or insured are lost. AI techniques are used to trawl through public records and websites to find lost beneficiaries.

A use case was carried out by Milliman consultants to enhance straight through processing and decision making for invoices for an insurance client. Some emails accompanying invoices contained important instructions regarding the invoice but more often, unnecessary information was also contained in the emails. A model was developed to reduce the amount of emails which accompanied invoices that were sent for further investigation. The implementation of the model led to reduced unnecessary email processing leaving more time for answering more relevant queries.

Fraud detection

Insurers have a wider range of tools to detect fraud such as looking at the claimants internet presence for smoking and drug use. Machine learning and analytics can be used to detect fraud relating to customers submitting false documents or misrepresenting health, family history or occupation.

Identity fraud or account takeover risk is growing and machine learning techniques can be used to detect abnormal activities on accounts. Anomaly detection models have been used to identify fraudsters accessing existing customer accounts.

Investment

In the current economic environment with low interest rates, it is challenging for insurers to produce strong investment returns for shareholders. The use of AI and analytics allow information to be processed from textual data and images such as from articles, broker recommendations or central bank updates. These models can extract new insights and signals allowing for quicker investment decisions.

Risk management and actuarial modelling

The use of neural networks and accessible public databases have opened up new opportunities in mortality modelling and forecasting. Milliman researchers carried out a use case to remove errors caused by a sudden change in fertility patterns that were incorrectly captured as cohort effects for countries that do not have monthly fertility records. As a result, Milliman researchers was able to produce corrected mortality tables for those countries.

Computationally intensive cashflow projection models such as those used in ALM calculations may benefit from machine learning techniques to group policy data. A use case was carried out by Milliman consultants to cluster policy data in a representative example portfolio containing traditional and universal life products. The clustered data resulted in smaller errors in the cashflow patterns than naïve grouping. Applications of these techniques may prove useful in IFRS 17 modelling as the techniques allow for the segmentation of variables which are required for IFRS 17 cohorts.

Machine learning techniques may be used for SCR calculations and variable annuities where many stochastic scenarios are required, and calculations can be time-consuming. Interpretability of the results from using these techniques remain challenging.

Milliman research paper

The full research paper can be found on Milliman’s website. The paper includes detailed descriptions on the various machine learning and AI tools used in the case studies and identifies the models that were best suited for each circumstance. Finally, feedback is obtained from insurers and pension providers on what they deem the most important use of data science techniques for their business.