A recent survey of North American insurance carriers revealed approximately 82% rely on predictive modeling and analytics to help with their product lines. Modeling provides strong analytical forecasting that insures a company is correctly pricing its lines of business and evaluating its risk. Predictive models also play a large part in expansions for companies looking to grow their books or potentially move into other lines.
Many companies are touting predictive analytics as a best practice for rating, essentially eliminating a great deal of the uncertainty that was attributed to a very manual underwriting and actuarial process. That being said, the data is only as good as the processes and platforms that are being used to analyze it. Predictive analytics and business intelligence are often referred to interchangeably; however, business intelligence is a subset of predictive analytics, and not necesssary the complete science of analytics.
Carriers have always relied on forecasting. In the case of property and casualty insurance in Cat exposed areas, earthquake, hurricane and tornado losses are often modeled using historical damage data and loss data to determine the potential risk of writing business in geographical areas that are potentially prone to large losses. This enables companies to determine their appetites in areas where they can write higher premiums but are prone to more risk.
Many companies outsource their analytics to actuarial or disaster modeling companies that have a greater level of expertise. And, many insurance companies rely on scorecards to determine the score for each potential risk. There are many factors that play into this and the requirement of not only knowing your own data, but training a predictive system to know your data and evolve with your data over time can be a daunting task.
Data mining is a component of predictive analytics that is used to document trends, patterns and relationships between various data sources. This information is then used to create a predictive model. Predictive analytics, along with the predictive model and data mining techniques, rely on increasingly sophisticated statistical methods, including multi and univariate analysis techniques in order to determine trends and relationships that may not be readily apparent, but adversely affect a company’s top and bottom lines.
Data availability and the ability to aggregate data from various data sources into a meaningful format is predictively one of the largest areas where companies fail in their efforts to create accurate and worthwhile predictive models. This results in inaccurate and skewed data, which can make or break a company if they get it wrong.
Although property and casualty companies generate a whopping $400 billion in premiums annually, the growth rate of premiums in these lines has risen only 5% in the last decade. This trend has many P&C companies scrambling to find ways to bring in more premiums and increase the value of their lines versus competitors. P&C companies are beginning to embrace predictive analytics as a means for not only extending their product lines, but also as a way to carve out more profitability in their current lines by evaluating the cost of claims, reserving practices, reinsurance and operational costs.
The area where companies have failed across the board has been the use of predictive analytics to measure consumer behavior. Social channels and the Internet have been largely ignored, as have demographics and customer buying habits based on statistical data such as age, location, income, credit scores and many other factors that could be used to provide a competitive edge to the insurer, while creating perceived value to the insured by offering competitive pricing and a broader range of product and pricing options.
The law of “adverse selection” is also present, which states the person who is the most at risk is most likely to be the one who buys insurance. An insurer needs to capture enough adequate information about the risk in order to determine whether it is worth writing.
Although there is a lot of information available, it can be very expensive for the insurer to collect, and there is concern than the money could potentially be spent on a prospect that does not convert to a piece of business. When companies cannot distinguish between high and low risk, they cannot adequately price their product to insure high risk candidates are eliminated from buying their products.
Next week, PC360 will publish part 2 of this predictive analytics series, "Find the real value in predictive analytics."