Risk management processes form an integral part of the insurance industry. Insurers consider every available quantifiable factor to develop profiles of high and low insurance risk for their prospective policyholders. Level of risk determines the insurance premiums of these policies. To this end, insurers collect a vast amount of information about policyholders and insured objects. Increasingly, statistical methods and tools based on data mining techniques are being used to determine insurance policy risk levels. Research by LIMRA, the Life Insurance and Market Research Association, states that “nearly nine in 10 financial services companies have or are exploring the use of big data analytics to compliment the underwriting processes.” We discuss the use of predictive analytics for assessing insurance risk in the following sections.
Insurance Underwriting
The process of insurance underwriting primarily involves the following three steps:
- Collecting client information such as personal credit history, vehicle reports including vehicle identification numbers, and medical history amongst others
- Analyzing the aforesaid information in detail to determine the client’s risk score
- Based on the predefined underwriting guidelines, accepting or declining the application, followed by the calculation of the premium amount of the customer
Transition with the Advent of Predictive Modelling
A few years ago, underwriters had to rely on specific, predefined guidelines, basic statistical models like profiling and scoring models and their intuitions for evaluating risk of insurance policies. However, with the advent of predictive analytical models, underwriters can now make data-based predictions about a customer’s risk profile with higher accuracy.
One of the key benefits of predictive analytics is cognitive insight. Through cognitive insight, underwriters can drive efficiency and accuracy by leveraging information on more complex portions of the process that facilitate decision making. Predictive analytics that are run at the beginning of the process offer granular visibility into specific risks. For example, it can alert underwriters about additional elements to be considered for evaluation which could have been missed under a traditional process.
However, this data is available to companies in many formats – structured, unstructured, and semi-structured data. Structured data such as internal claims data can be standardized and mined for useful insights. The potential of unstructured data has been difficult to harness. Unstructured data can be in many forms – emails, geo-spatial location, images, videos, electronic health records, data from smart phones and wearables and social media data. Underwriters can access this data to evaluate an insurance application, the risks involved and make data-driven predictions. Additionally, unstructured data such as geo-spatial data in conjunction with other data, can be useful in identifying trends in insurance fraud.
Information can be collected from social media, credit agencies, government agencies, counsellors, and third-party vendors. Together, these data points are combined into a single dataset that can then be used for analytical purposes. This is followed by data preparation which provides a solid foundation for model development. Data preparation can be summarized into four steps which are described below:
- Variable generation: The process of creating variables from the raw data
- Exploratory data analysis: The process of analyzing the distributional properties of each variable. Descriptive statistics such as min, max, mean, median, mode, and frequency can provide useful insights to insurers
- Variable transformation: The aforesaid exploratory data analysis might reveal some imperfections in the data which must be addressed before constructing the predictive model
- Partitioning model set for model build: The process of dividing the data set into three approximately equal parts commonly called as the ‘test’, ‘train’, and ‘validation’ sets. The last two sets are used for model building while the ‘test’ is used to assess the model results
While data mining is about summarizing large datasets, predictive modelling involves model creation using a set of mathematical tools. These models throw light on how a policy might perform in the future, giving you a detailed insight of the risk involved by unravelling the hidden patterns of data. These models form an integral part of the enterprise rule engines and integrated seamlessly into the application landscape.
It is expected that more and more insurance companies will start using unconventional underwriting evidence as means of data collection in the near future.
A thorough training of insurers in analytics is imperative for efficient risk management processes. Without proper knowledge and understanding about data and analytics, insurers will be unable to fully realize the potential of data-driven predictions. Also, this should not be just looked at as an either/or proposition. Rather, insurance companies should empower skilled underwriters to supplement these predictive models with their own insights and knowledge gained from experience for better decisions.
Business Impact and Benefits
Predictive models analytics and risk analytics enable underwriters to get an automated result to guide them in the decision-making stage. Given the dynamic nature of data collected, it is important to employ predictive modelling to get consistent and automated results. In doing so, insurance companies can also reduce costs, improve the experience of their clients, and generate substantial business growth.
Predictive modelling solutions can process structured, semi-structured and unstructured data formats. This also reduces the time of processing applications and helps organizations save time. According to a Deloitte study of 15 life insurers, the median service time to issue a new policy ranged between 30 and 35 days for policies with face amounts between $0.1 million to $5 million. This means that a life insurer typically spends approximately one month and several hundred dollars underwriting each applicant. And these numbers continue to significantly grow.
Since predictive modelling and risk analytics can reveal hidden patterns in data, it gives a clearer insight of the impending risks. This in turn helps organizations to take preventive measures as well as adds to their learning experience. In the long run, organizations can strategize a new methodology to avoid risks, given their rich experience with big data analytics. Predictive analytics can be used by underwriting to upsell and cross-sell as well – when the risk assessment is a little more automated they can focus on customization and selling of products as well.
Predictive analytics has revolutionized how companies dig data and extract actionable insights, but its full potential is yet to be fully realized in the insurance industry. Insurance organizations are now turning to predictive modelling to extract maximum value out of their data.
Leveraging advanced analytics for insurance: At LatentView Analytics, we draw on our deep insurance experience to help organizations unlock new areas of value. We work with insurers to find opportunities that deliver profitable growth while protecting and optimizing their enterprise. To know more about LatentView’s end-to-end custom analytics solutions, please write into: marketing@latentview.com
